Initial commit: Research Intelligence System

Self-hosted research paper triage with AI scoring, preference learning,
and a setup wizard for first-time configuration.

.dockerignore

@@ -0,0 +1,17 @@
+.git
+.gitignore
+.env
+.claude
+__pycache__
+*.pyc
+*.pyo
+data/*.db
+data/*.db-wal
+data/*.db-shm
+data/weeks/
+.pytest_cache
+.mypy_cache
+.ruff_cache
+README.md
+CLAUDE.md
+config.yaml

.env.example

@@ -0,0 +1,5 @@
+# Required: Anthropic API key for paper scoring
+ANTHROPIC_API_KEY=your-key-here
+
+# Optional: GitHub token for higher API rate limits
+GITHUB_TOKEN=

.gitignore

@@ -0,0 +1,32 @@
+# Environment
+.env
+
+# Database
+data/*.db
+data/*.db-wal
+data/*.db-shm
+data/backups/
+data/weeks/
+
+# Python
+__pycache__/
+*.pyc
+*.pyo
+.pytest_cache/
+.mypy_cache/
+.ruff_cache/
+
+# IDE / Editor
+.claude/
+.vscode/
+.idea/
+*.swp
+*.swo
+*~
+
+# OS
+.DS_Store
+Thumbs.db
+
+# Generated config (created by setup wizard)
+config.yaml

CLAUDE.md

@@ -0,0 +1,73 @@
+# Research Intelligence System
+
+## Architecture
+
+- **Web dashboard**: FastAPI + Jinja2 + HTMX on port 8888
+- **Database**: SQLite at `data/researcher.db` (configurable in `config.yaml`)
+- **Config**: YAML-driven via `config.yaml` (generated by setup wizard on first run)
+- **Pipelines**: `src/pipelines/aiml.py` (HF + arXiv), `src/pipelines/security.py` (arXiv cs.CR)
+- **Scoring**: `src/scoring.py` — Claude API batch scoring with configurable axes
+- **Preferences**: `src/preferences.py` — learns from user signals (upvote/downvote/save/dismiss)
+- **Scheduler**: APScheduler runs on configurable cron schedule
+
+## Key Files
+
+| File | Purpose |
+|------|---------|
+| `src/config.py` | YAML config loader, scoring prompt builder, defaults |
+| `src/db.py` | SQLite schema + query helpers |
+| `src/scoring.py` | Unified Claude API scorer |
+| `src/preferences.py` | Preference computation from user signals |
+| `src/pipelines/aiml.py` | AI/ML paper fetching (HF + arXiv) |
+| `src/pipelines/security.py` | Security paper fetching (arXiv cs.CR) |
+| `src/pipelines/github.py` | GitHub trending projects via OSSInsight |
+| `src/pipelines/events.py` | Conferences, releases, RSS news |
+| `src/web/app.py` | FastAPI routes, middleware, report generation |
+| `src/scheduler.py` | APScheduler weekly trigger |
+
+## Config System
+
+`src/config.py` loads `config.yaml` and exposes module-level constants:
+
+- `FIRST_RUN` — True when `config.yaml` doesn't exist (triggers setup wizard)
+- `SCORING_CONFIGS` — Dict of domain scoring configs (axes, weights, prompts)
+- `DB_PATH` — Path to SQLite database
+- `ANTHROPIC_API_KEY` — From `.env` or environment
+
+Scoring prompts are built dynamically from `scoring_axes` and `preferences` in config.
+
+## Working with the Database
+
+```bash
+sqlite3 data/researcher.db
+
+# Top papers
+SELECT title, composite, summary FROM papers
+WHERE domain='aiml' AND composite IS NOT NULL
+ORDER BY composite DESC LIMIT 10;
+
+# Signal counts
+SELECT action, COUNT(*) FROM signals GROUP BY action;
+
+# Preference profile
+SELECT * FROM preferences ORDER BY abs(pref_value) DESC LIMIT 20;
+```
+
+## Docker
+
+```bash
+docker compose up --build
+# Dashboard at http://localhost:9090
+# Setup wizard runs on first visit
+
+# Trigger pipelines
+curl -X POST http://localhost:9090/run/aiml
+curl -X POST http://localhost:9090/run/security
+```
+
+## Allowed Tools
+
+When working with this project in Claude Code:
+- **Bash**: python, sqlite3, curl, docker commands
+- **WebSearch/WebFetch**: arXiv, GitHub, HuggingFace for paper details
+- **Read/Edit**: all project files and data/

Dockerfile

@@ -0,0 +1,28 @@
+FROM python:3.12-slim
+
+WORKDIR /app
+
+# Install dependencies (cached layer)
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+# Create non-root user
+RUN useradd -m -s /bin/bash appuser
+
+# Copy source
+COPY src/ src/
+COPY data/seed_papers.json data/seed_papers.json
+COPY entrypoint.sh .
+RUN chmod +x entrypoint.sh
+
+# Create data directory with correct ownership
+RUN mkdir -p data/weeks && chown -R appuser:appuser /app
+
+USER appuser
+
+EXPOSE 8888
+
+HEALTHCHECK --interval=30s --timeout=10s --retries=3 --start-period=15s \
+    CMD python -c "import urllib.request; urllib.request.urlopen('http://localhost:8888/api/status')"
+
+ENTRYPOINT ["./entrypoint.sh"]

README.md

@@ -0,0 +1,128 @@
+# Research Intelligence
+
+A self-hosted research triage system that monitors academic papers (AI/ML and Security) and trending GitHub projects, scores them with Claude, and learns your preferences over time.
+
+## Features
+
+- **Paper monitoring** — Fetches new papers from arXiv and HuggingFace daily/weekly
+- **AI scoring** — Claude scores each paper on configurable axes (novelty, code availability, practical impact)
+- **Preference learning** — Rate papers with thumbs up/down; the system learns what you care about and re-ranks accordingly
+- **GitHub tracking** — Monitors trending repositories across curated collections
+- **Event tracking** — Conference deadlines, releases, and RSS news feeds
+- **Weekly reports** — Auto-generated markdown summaries of top papers
+- **Dark-theme dashboard** — Fast, responsive web UI built with HTMX
+
+## Quick Start
+
+### Docker (recommended)
+
+```bash
+git clone https://github.com/yourname/researcher.git
+cd researcher
+cp .env.example .env
+# Edit .env and add your Anthropic API key
+
+docker compose up --build
+```
+
+Visit **http://localhost:9090** — the setup wizard will guide you through configuration.
+
+### Local
+
+```bash
+git clone https://github.com/yourname/researcher.git
+cd researcher
+pip install -r requirements.txt
+cp .env.example .env
+# Edit .env and add your Anthropic API key
+
+python -m uvicorn src.web.app:app --host 0.0.0.0 --port 8888
+```
+
+Visit **http://localhost:8888** and follow the setup wizard.
+
+## Setup Wizard
+
+On first launch (before `config.yaml` exists), you'll be guided through:
+
+1. **API Key** — Enter your Anthropic API key (validated with a test call)
+2. **Domains** — Enable/disable AI/ML and Security monitoring, adjust scoring weights
+3. **GitHub** — Toggle GitHub project tracking
+4. **Schedule** — Set pipeline frequency (daily, weekly, or manual-only)
+
+After setup, you can optionally **pick seed papers** to bootstrap your preference profile.
+
+## Configuration
+
+All settings live in `config.yaml` (generated by the setup wizard). You can also edit it directly:
+
+```yaml
+domains:
+  aiml:
+    enabled: true
+    scoring_axes:
+      - name: "Code & Weights"
+        weight: 0.30
+      - name: "Novelty"
+        weight: 0.35
+      - name: "Practical Applicability"
+        weight: 0.35
+  security:
+    enabled: true
+    scoring_axes:
+      - name: "Has Code/PoC"
+        weight: 0.25
+      - name: "Novel Attack Surface"
+        weight: 0.40
+      - name: "Real-World Impact"
+        weight: 0.35
+
+schedule:
+  cron: "0 22 * * 0"  # Weekly on Sunday at 22:00 UTC
+```
+
+## Architecture
+
+| Component | Technology |
+|-----------|-----------|
+| Web server | FastAPI + Jinja2 + HTMX |
+| Database | SQLite |
+| Scoring | Claude API (Anthropic) |
+| Scheduling | APScheduler |
+| Container | Docker |
+
+### Key Files
+
+| File | Purpose |
+|------|---------|
+| `src/config.py` | YAML config loader with defaults |
+| `src/db.py` | SQLite schema and queries |
+| `src/scoring.py` | Claude API batch scorer |
+| `src/preferences.py` | Preference learning from user signals |
+| `src/pipelines/aiml.py` | AI/ML paper fetcher (HF + arXiv) |
+| `src/pipelines/security.py` | Security paper fetcher (arXiv cs.CR) |
+| `src/pipelines/github.py` | GitHub trending projects |
+| `src/pipelines/events.py` | Conferences, releases, RSS |
+| `src/web/app.py` | Web routes and middleware |
+| `src/scheduler.py` | Cron-based pipeline scheduler |
+
+## Running Pipelines Manually
+
+From the dashboard, click the pipeline buttons. Or via API:
+
+```bash
+curl -X POST http://localhost:9090/run/aiml
+curl -X POST http://localhost:9090/run/security
+curl -X POST http://localhost:9090/run/github
+curl -X POST http://localhost:9090/run/events
+```
+
+## Requirements
+
+- Python 3.12+
+- Anthropic API key (for paper scoring)
+- Optional: GitHub token (for higher API rate limits)
+
+## License
+
+MIT

data/seed_papers.json

@@ -0,0 +1,182 @@
+[
+  {
+    "arxiv_id": "2401.04088",
+    "title": "DeepSeek-Coder-V2: Breaking the Barrier of Closed-Source Models in Code Intelligence",
+    "domain": "aiml",
+    "summary": "Open-source code LLM matching GPT-4 Turbo on coding benchmarks with MoE architecture."
+  },
+  {
+    "arxiv_id": "2403.05530",
+    "title": "GaLore: Memory-Efficient LLM Training by Gradient Low-Rank Projection",
+    "domain": "aiml",
+    "summary": "Reduces memory usage for LLM training via gradient projection, enabling 7B training on consumer GPUs."
+  },
+  {
+    "arxiv_id": "2402.13616",
+    "title": "World Model on Million-Length Video and Language with RingAttention",
+    "domain": "aiml",
+    "summary": "Trains world models on million-token video sequences using ring attention for long context."
+  },
+  {
+    "arxiv_id": "2403.03206",
+    "title": "The Claude 3 Model Family",
+    "domain": "aiml",
+    "summary": "Multimodal LLM family with strong vision capabilities and extended context windows."
+  },
+  {
+    "arxiv_id": "2402.17764",
+    "title": "Sora: A Review on Background, Technology, Limitations, and Opportunities",
+    "domain": "aiml",
+    "summary": "Analysis of video generation model capabilities, architecture, and limitations."
+  },
+  {
+    "arxiv_id": "2401.02954",
+    "title": "MoE-Mamba: Efficient Selective State Space Models with Mixture of Experts",
+    "domain": "aiml",
+    "summary": "Combines Mamba state-space model with mixture-of-experts for efficient scaling."
+  },
+  {
+    "arxiv_id": "2403.09611",
+    "title": "Quiet-STaR: Language Models Can Teach Themselves to Think Before Speaking",
+    "domain": "aiml",
+    "summary": "Self-taught reasoning where LLMs learn to generate internal rationale tokens."
+  },
+  {
+    "arxiv_id": "2402.01032",
+    "title": "OLMo: Accelerating the Science of Language Models",
+    "domain": "aiml",
+    "summary": "Fully open-source LLM with released weights, code, data, and training logs."
+  },
+  {
+    "arxiv_id": "2403.14608",
+    "title": "ReALM: Reference Resolution As Language Modeling",
+    "domain": "aiml",
+    "summary": "Resolves onscreen and conversational references using LLMs for device agents."
+  },
+  {
+    "arxiv_id": "2402.14261",
+    "title": "Griffin: Mixing Gated Linear Recurrences with Local Attention for Efficient Language Models",
+    "domain": "aiml",
+    "summary": "Hybrid architecture combining gated linear RNNs with local attention, matching transformer quality."
+  },
+  {
+    "arxiv_id": "2401.14196",
+    "title": "GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers",
+    "domain": "aiml",
+    "summary": "One-shot quantization method reducing LLM size to 3-4 bits with minimal accuracy loss."
+  },
+  {
+    "arxiv_id": "2403.07691",
+    "title": "Stealing Part of a Production Language Model",
+    "domain": "security",
+    "summary": "Extracts internal architecture details from production LLM APIs through crafted queries."
+  },
+  {
+    "arxiv_id": "2402.06132",
+    "title": "SoK: Where's the Bug? A Study of Bug Localization Tools",
+    "domain": "security",
+    "summary": "Systematizes bug localization approaches and evaluates 23 tools on real-world CVEs."
+  },
+  {
+    "arxiv_id": "2401.16727",
+    "title": "A Survey of Side-Channel Attacks on Intel SGX",
+    "domain": "security",
+    "summary": "Comprehensive analysis of side-channel attacks targeting Intel SGX enclaves."
+  },
+  {
+    "arxiv_id": "2403.02783",
+    "title": "SyzVegas: Beating Kernel Fuzzing Odds with Reinforcement Learning",
+    "domain": "security",
+    "summary": "RL-guided kernel fuzzer that outperforms Syzkaller in bug discovery rate."
+  },
+  {
+    "arxiv_id": "2402.15483",
+    "title": "BSIMM: An Empirical Study of 130 Software Security Programs",
+    "domain": "security",
+    "summary": "Large-scale study of enterprise security maturity across 130 organizations."
+  },
+  {
+    "arxiv_id": "2403.14469",
+    "title": "Reverse Engineering eBPF Programs: Challenges and Approaches",
+    "domain": "security",
+    "summary": "Novel techniques for reverse engineering eBPF bytecode in Linux kernel security."
+  },
+  {
+    "arxiv_id": "2401.09577",
+    "title": "WiFi-Based Keystroke Inference Attack Using Adversarial CSI Perturbation",
+    "domain": "security",
+    "summary": "Exploits WiFi channel state information to infer keystrokes from nearby devices."
+  },
+  {
+    "arxiv_id": "2402.08787",
+    "title": "Binary Code Similarity Detection via Graph Neural Networks",
+    "domain": "security",
+    "summary": "GNN-based approach to detect similar binary functions across compilers and architectures."
+  },
+  {
+    "arxiv_id": "2403.01218",
+    "title": "Practical Exploitation of DNS Rebinding in IoT Devices",
+    "domain": "security",
+    "summary": "Demonstrates DNS rebinding attacks against 15 popular IoT devices in home networks."
+  },
+  {
+    "arxiv_id": "2401.15491",
+    "title": "GPU.zip: Side Channel Attacks on GPU-Based Graphical Data Compression",
+    "domain": "security",
+    "summary": "First cross-origin pixel-stealing attack through GPU hardware data compression."
+  },
+  {
+    "arxiv_id": "2402.03367",
+    "title": "CryptoFuzz: Fully Automated Testing of Cryptographic API Misuse",
+    "domain": "security",
+    "summary": "Automated fuzzer detecting cryptographic API misuse patterns in Java applications."
+  },
+  {
+    "arxiv_id": "2403.08946",
+    "title": "Video Generation Models as World Simulators",
+    "domain": "aiml",
+    "summary": "Explores how video generation models learn physical world dynamics as implicit simulators."
+  },
+  {
+    "arxiv_id": "2402.05929",
+    "title": "V-JEPA: Video Joint Embedding Predictive Architecture",
+    "domain": "aiml",
+    "summary": "Self-supervised video representation learning that predicts in latent space rather than pixel space."
+  },
+  {
+    "arxiv_id": "2401.10020",
+    "title": "AlphaGeometry: Solving Olympiad Geometry without Human Demonstrations",
+    "domain": "aiml",
+    "summary": "AI system solving IMO-level geometry problems through neurosymbolic reasoning."
+  },
+  {
+    "arxiv_id": "2403.04132",
+    "title": "Design2Code: How Far Are We From Automating Front-End Engineering?",
+    "domain": "aiml",
+    "summary": "Benchmarks multimodal LLMs on converting visual designs to functional HTML/CSS code."
+  },
+  {
+    "arxiv_id": "2402.14905",
+    "title": "YOLOv9: Learning What You Want to Learn Using Programmable Gradient Information",
+    "domain": "aiml",
+    "summary": "New YOLO architecture using programmable gradient information for better object detection."
+  },
+  {
+    "arxiv_id": "2401.06066",
+    "title": "MagicVideo-V2: Multi-Stage High-Aesthetic Video Generation",
+    "domain": "aiml",
+    "summary": "Multi-stage video generation pipeline producing high-quality aesthetic videos from text."
+  },
+  {
+    "arxiv_id": "2402.01680",
+    "title": "Grandmaster-Level Chess Without Search",
+    "domain": "aiml",
+    "summary": "Transformer achieving grandmaster chess play through pure pattern recognition without tree search."
+  },
+  {
+    "arxiv_id": "2403.04706",
+    "title": "SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering",
+    "domain": "aiml",
+    "summary": "LLM agent that autonomously fixes GitHub issues by interacting with code repositories."
+  }
+]

docker-compose.yml

@@ -0,0 +1,27 @@
+services:
+  researcher:
+    build: .
+    ports:
+      - "9090:8888"
+    volumes:
+      - ./data:/app/data
+    environment:
+      - ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY}
+      - GITHUB_TOKEN=${GITHUB_TOKEN}
+      - PYTHONUNBUFFERED=1
+    restart: unless-stopped
+    healthcheck:
+      test: ["CMD", "python", "-c", "import urllib.request; urllib.request.urlopen('http://localhost:8888/api/status')"]
+      interval: 30s
+      timeout: 10s
+      retries: 3
+      start_period: 15s
+    logging:
+      driver: json-file
+      options:
+        max-size: "10m"
+        max-file: "3"
+    deploy:
+      resources:
+        limits:
+          memory: 2g

entrypoint.sh

@@ -0,0 +1,7 @@
+#!/bin/bash
+set -e
+
+echo "=== Research Intelligence ==="
+echo "Starting web server + scheduler on port 8888 ..."
+
+exec python -m uvicorn src.web.app:app --host 0.0.0.0 --port 8888

requirements.txt

@@ -0,0 +1,17 @@
+# Core web
+fastapi>=0.115,<1
+uvicorn>=0.34,<1
+jinja2>=3.1,<4
+python-multipart>=0.0.18
+
+# Data
+arxiv>=2.1,<3
+requests>=2.31,<3
+anthropic>=0.40,<1
+feedparser>=6.0,<7
+
+# Config
+pyyaml>=6.0,<7
+
+# Scheduling
+apscheduler>=3.10,<4

scripts/backup-db.sh

@@ -0,0 +1,35 @@
+#!/bin/bash
+# Daily SQLite backup — safe online backup using .backup command
+# Add to crontab: 0 3 * * * /path/to/researcher/scripts/backup-db.sh
+
+set -e
+
+SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
+PROJECT_DIR="$(dirname "$SCRIPT_DIR")"
+
+DB_PATH="${PROJECT_DIR}/data/researcher.db"
+BACKUP_DIR="${PROJECT_DIR}/data/backups"
+KEEP_DAYS=14
+
+mkdir -p "$BACKUP_DIR"
+
+TIMESTAMP=$(date +%Y%m%d-%H%M%S)
+BACKUP_FILE="$BACKUP_DIR/researcher-$TIMESTAMP.db"
+
+# Use SQLite online backup (safe with WAL mode)
+python3 -c "
+import sqlite3, shutil
+src = sqlite3.connect('$DB_PATH')
+dst = sqlite3.connect('$BACKUP_FILE')
+src.backup(dst)
+dst.close()
+src.close()
+"
+
+# Compress
+gzip "$BACKUP_FILE"
+echo "Backup: ${BACKUP_FILE}.gz ($(du -h "${BACKUP_FILE}.gz" | cut -f1))"
+
+# Prune old backups
+find "$BACKUP_DIR" -name "researcher-*.db.gz" -mtime +"$KEEP_DAYS" -delete
+echo "Pruned backups older than $KEEP_DAYS days"

src/config.py

@@ -0,0 +1,505 @@
+"""Configuration loader — reads from config.yaml, falls back to defaults."""
+
+import logging
+import os
+import re
+import sys
+from pathlib import Path
+
+# ---------------------------------------------------------------------------
+# Logging (always available, before config loads)
+# ---------------------------------------------------------------------------
+
+LOG_FORMAT = "%(asctime)s [%(name)s] %(levelname)s: %(message)s"
+LOG_LEVEL = os.environ.get("LOG_LEVEL", "INFO").upper()
+
+logging.basicConfig(
+    format=LOG_FORMAT,
+    level=getattr(logging, LOG_LEVEL, logging.INFO),
+    stream=sys.stdout,
+)
+
+# Quiet noisy libraries
+logging.getLogger("httpx").setLevel(logging.WARNING)
+logging.getLogger("httpcore").setLevel(logging.WARNING)
+logging.getLogger("apscheduler").setLevel(logging.WARNING)
+
+log = logging.getLogger(__name__)
+
+# ---------------------------------------------------------------------------
+# Config file path
+# ---------------------------------------------------------------------------
+
+CONFIG_PATH = Path(os.environ.get("CONFIG_PATH", "config.yaml"))
+FIRST_RUN = not CONFIG_PATH.exists()
+
+# ---------------------------------------------------------------------------
+# Environment
+# ---------------------------------------------------------------------------
+
+ANTHROPIC_API_KEY = os.environ.get("ANTHROPIC_API_KEY", "")
+GITHUB_TOKEN = os.environ.get("GITHUB_TOKEN", "")
+
+
+def validate_env():
+    """Check required environment variables at startup. Warn on missing."""
+    if not ANTHROPIC_API_KEY:
+        log.warning("ANTHROPIC_API_KEY not set — scoring will be disabled")
+    if not GITHUB_TOKEN:
+        log.info("GITHUB_TOKEN not set — GitHub API calls will be rate-limited")
+
+
+# ---------------------------------------------------------------------------
+# Load config.yaml (or defaults)
+# ---------------------------------------------------------------------------
+
+def _load_yaml() -> dict:
+    """Load config.yaml if present, otherwise return empty dict."""
+    if CONFIG_PATH.exists():
+        try:
+            import yaml
+            with open(CONFIG_PATH) as f:
+                data = yaml.safe_load(f) or {}
+            log.info("Loaded config from %s", CONFIG_PATH)
+            return data
+        except Exception as e:
+            log.error("Failed to load %s: %s — using defaults", CONFIG_PATH, e)
+    return {}
+
+
+_cfg = _load_yaml()
+
+# ---------------------------------------------------------------------------
+# Claude API
+# ---------------------------------------------------------------------------
+
+CLAUDE_MODEL = _cfg.get("claude_model", "claude-sonnet-4-5-20250929")
+BATCH_SIZE = _cfg.get("batch_size", 20)
+
+# ---------------------------------------------------------------------------
+# Database
+# ---------------------------------------------------------------------------
+
+DB_PATH = Path(_cfg.get("database", {}).get("path", os.environ.get("DB_PATH", "data/researcher.db")))
+
+# ---------------------------------------------------------------------------
+# Web
+# ---------------------------------------------------------------------------
+
+WEB_HOST = _cfg.get("web", {}).get("host", "0.0.0.0")
+WEB_PORT = _cfg.get("web", {}).get("port", 8888)
+
+# ---------------------------------------------------------------------------
+# Schedule
+# ---------------------------------------------------------------------------
+
+SCHEDULE_CRON = _cfg.get("schedule", {}).get("cron", "0 22 * * 0")
+
+# ---------------------------------------------------------------------------
+# Domains from config
+# ---------------------------------------------------------------------------
+
+_domains_cfg = _cfg.get("domains", {})
+
+# ---------------------------------------------------------------------------
+# Shared constants
+# ---------------------------------------------------------------------------
+
+HF_API = "https://huggingface.co/api"
+GITHUB_URL_RE = re.compile(r"https?://github\.com/[A-Za-z0-9_.-]+/[A-Za-z0-9_.-]+")
+MAX_ABSTRACT_CHARS_AIML = 2000
+MAX_ABSTRACT_CHARS_SECURITY = 1500
+HF_MAX_AGE_DAYS = 90
+
+# ---------------------------------------------------------------------------
+# AI/ML pipeline constants
+# ---------------------------------------------------------------------------
+
+_aiml_cfg = _domains_cfg.get("aiml", {})
+
+ARXIV_LARGE_CATS = _aiml_cfg.get("arxiv_categories", ["cs.CV", "cs.CL", "cs.LG"])
+ARXIV_SMALL_CATS = ["eess.AS", "cs.SD"]
+
+_aiml_include = _aiml_cfg.get("include_patterns", [])
+_aiml_exclude = _aiml_cfg.get("exclude_patterns", [])
+
+_DEFAULT_INCLUDE = (
+    r"video.generat|world.model|image.generat|diffusion|text.to.image|text.to.video|"
+    r"code.generat|foundation.model|open.weight|large.language|language.model|"
+    r"text.to.speech|tts|speech.synth|voice.clon|audio.generat|"
+    r"transformer|attention.mechanism|state.space|mamba|mixture.of.expert|\bmoe\b|"
+    r"scaling.law|architecture|quantiz|distillat|pruning|"
+    r"multimodal|vision.language|\bvlm\b|agent|reasoning|"
+    r"reinforcement.learn|rlhf|dpo|preference.optim|"
+    r"retrieval.augment|\brag\b|in.context.learn|"
+    r"image.edit|video.edit|3d.generat|nerf|gaussian.splat|"
+    r"robot|embodied|simulat|"
+    r"benchmark|evaluat|leaderboard|"
+    r"open.source|reproducib|"
+    r"instruction.tun|fine.tun|align|"
+    r"long.context|context.window|"
+    r"token|vocab|embedding|"
+    r"training.efficien|parallel|distributed.train|"
+    r"synthetic.data|data.curat"
+)
+
+_DEFAULT_EXCLUDE = (
+    r"medical.imag|clinical|radiology|pathology|histolog|"
+    r"climate.model|weather.predict|meteorolog|"
+    r"survey.of|comprehensive.survey|"
+    r"sentiment.analysis|named.entity|"
+    r"drug.discover|protein.fold|molecular.dock|"
+    r"software.engineering.practice|code.smell|technical.debt|"
+    r"autonomous.driv|traffic.signal|"
+    r"remote.sens|satellite.imag|crop.yield|"
+    r"stock.predict|financial.forecast|"
+    r"electronic.health|patient.record|"
+    r"seismic|geophys|oceanograph|"
+    r"educational.data|student.perform|"
+    r"blockchain|smart.contract|\bdefi\b|decentralized.finance|cryptocurrency|"
+    r"jailbreak|guardrail|red.teaming|llm.safety|"
+    r"safe.alignment|safety.tuning|harmful.content|toxicity"
+)
+
+INCLUDE_RE = re.compile(
+    "|".join(_aiml_include) if _aiml_include else _DEFAULT_INCLUDE,
+    re.IGNORECASE,
+)
+
+EXCLUDE_RE = re.compile(
+    "|".join(_aiml_exclude) if _aiml_exclude else _DEFAULT_EXCLUDE,
+    re.IGNORECASE,
+)
+
+# ---------------------------------------------------------------------------
+# Security pipeline constants
+# ---------------------------------------------------------------------------
+
+_sec_cfg = _domains_cfg.get("security", {})
+
+SECURITY_KEYWORDS = re.compile(
+    r"\b(?:attack|vulnerability|exploit|fuzzing|fuzz|malware|"
+    r"intrusion|ransomware|phishing|adversarial|"
+    r"defense|defence|secure|security|privacy|"
+    r"cryptograph|authentication|authorization|"
+    r"injection|xss|csrf|cve\-\d|penetration.test|"
+    r"threat|anomaly.detect|ids\b|ips\b|firewall|"
+    r"reverse.engineer|obfuscat|sandbox|"
+    r"side.channel|buffer.overflow|zero.day|"
+    r"botnet|rootkit|trojan|worm)\b",
+    re.IGNORECASE,
+)
+
+ADJACENT_CATEGORIES = ["cs.AI", "cs.SE", "cs.NI", "cs.DC", "cs.OS", "cs.LG"]
+
+SECURITY_EXCLUDE_RE = re.compile(
+    r"blockchain|smart.contract|\bdefi\b|decentralized.finance|"
+    r"memecoin|meme.coin|cryptocurrency.trading|\bnft\b|"
+    r"comprehensive.survey|systematization.of.knowledge|"
+    r"differential.privacy.(?:mechanism|framework)|"
+    r"stock.predict|financial.forecast|crop.yield|"
+    r"sentiment.analysis|educational.data",
+    re.IGNORECASE,
+)
+
+SECURITY_LLM_RE = re.compile(
+    r"jailbreak|guardrail|red.teaming|"
+    r"llm.safety|safe.alignment|safety.tuning|"
+    r"harmful.(?:content|output)|toxicity|content.moderation|"
+    r"prompt.injection|"
+    r"reward.model.(?:for|safety|alignment)",
+    re.IGNORECASE,
+)
+
+# ---------------------------------------------------------------------------
+# Dynamic scoring prompt builder
+# ---------------------------------------------------------------------------
+
+def _build_scoring_prompt(domain: str, axes: list[dict], preferences: dict) -> str:
+    """Build a Claude scoring prompt from config axes + preferences."""
+    boost = preferences.get("boost_topics", [])
+    penalize = preferences.get("penalize_topics", [])
+
+    if domain == "aiml":
+        return _build_aiml_prompt(axes, boost, penalize)
+    elif domain == "security":
+        return _build_security_prompt(axes, boost, penalize)
+    return ""
+
+
+def _build_aiml_prompt(axes: list[dict], boost: list[str], penalize: list[str]) -> str:
+    """Generate AI/ML scoring prompt from axes config."""
+    axis_fields = []
+    axis_section = []
+    for i, ax in enumerate(axes, 1):
+        name = ax.get("name", f"axis_{i}")
+        desc = ax.get("description", "")
+        field = name.lower().replace(" ", "_").replace("&", "and").replace("/", "_")
+        axis_fields.append(field)
+        axis_section.append(f"{i}. **{field}** — {name}: {desc}")
+
+    boost_line = ", ".join(boost) if boost else (
+        "New architectures, open-weight models, breakthrough methods, "
+        "papers with code AND weights, efficiency improvements"
+    )
+    penalize_line = ", ".join(penalize) if penalize else (
+        "Surveys, incremental SOTA, closed-model papers, "
+        "medical/climate/remote sensing applications"
+    )
+
+    return f"""\
+You are an AI/ML research analyst. Score each paper on three axes (1-10):
+
+{chr(10).join(axis_section)}
+
+Scoring preferences:
+- Score UP: {boost_line}
+- Score DOWN: {penalize_line}
+
+Use HF ecosystem signals: hf_upvotes > 50 means community interest; hf_models present = weights available;
+hf_spaces = demo exists; github_repo = code available; source "both" = higher visibility.
+
+Also provide:
+- **summary**: 2-3 sentence practitioner-focused summary.
+- **reasoning**: 1-2 sentences explaining scoring.
+- **code_url**: Extract GitHub/GitLab URL from abstract/comments if present, else null.
+
+Respond with a JSON array of objects, one per paper, each with fields:
+arxiv_id, {", ".join(axis_fields)}, summary, reasoning, code_url
+"""
+
+
+def _build_security_prompt(axes: list[dict], boost: list[str], penalize: list[str]) -> str:
+    """Generate security scoring prompt from axes config."""
+    axis_fields = []
+    axes_section = []
+    for i, ax in enumerate(axes, 1):
+        name = ax.get("name", f"axis_{i}")
+        desc = ax.get("description", "")
+        field = name.lower().replace(" ", "_").replace("&", "and").replace("/", "_")
+        axis_fields.append(field)
+        axes_section.append(f"{i}. **{field}** (1-10) — {name}: {desc}")
+
+    return f"""\
+You are a security research analyst. Score each paper on three axes (1-10).
+
+=== HARD RULES (apply BEFORE scoring) ===
+
+1. If the paper is primarily about LLM safety, alignment, jailbreaking, guardrails,
+   red-teaming LLMs, or making AI models safer: cap ALL three axes at 3 max.
+   Check the "llm_adjacent" field — if true, this rule almost certainly applies.
+
+2. If the paper is a survey, SoK, or literature review: cap {axis_fields[1] if len(axis_fields) > 1 else 'axis_2'} at 2 max.
+
+3. If the paper is about blockchain, DeFi, cryptocurrency, smart contracts: cap ALL three axes at 2 max.
+
+4. If the paper is about theoretical differential privacy or federated learning
+   without concrete security attacks: cap ALL three axes at 3 max.
+
+=== SCORING AXES ===
+
+{chr(10).join(axes_section)}
+
+=== OUTPUT ===
+
+For each paper also provide:
+- **summary**: 2-3 sentence practitioner-focused summary.
+- **reasoning**: 1-2 sentences explaining your scoring.
+- **code_url**: Extract GitHub/GitLab URL from abstract/comments if present, else null.
+
+Respond with a JSON array of objects, one per paper, each with fields:
+entry_id, {", ".join(axis_fields)}, summary, reasoning, code_url
+"""
+
+
+# ---------------------------------------------------------------------------
+# Scoring configs per domain
+# ---------------------------------------------------------------------------
+
+def _build_scoring_configs() -> dict:
+    """Build SCORING_CONFIGS from config.yaml or defaults."""
+    configs = {}
+
+    # AI/ML config
+    aiml_axes_cfg = _aiml_cfg.get("scoring_axes", [
+        {"name": "Code & Weights", "weight": 0.30, "description": "Open weights on HF, code on GitHub"},
+        {"name": "Novelty", "weight": 0.35, "description": "Paradigm shifts over incremental"},
+        {"name": "Practical Applicability", "weight": 0.35, "description": "Usable by practitioners soon"},
+    ])
+    aiml_prefs = _aiml_cfg.get("preferences", {})
+    aiml_weight_keys = ["code_weights", "novelty", "practical"]
+    aiml_weights = {}
+    for i, ax in enumerate(aiml_axes_cfg):
+        key = aiml_weight_keys[i] if i < len(aiml_weight_keys) else f"axis_{i+1}"
+        aiml_weights[key] = ax.get("weight", 1.0 / len(aiml_axes_cfg))
+
+    configs["aiml"] = {
+        "weights": aiml_weights,
+        "axes": ["code_weights", "novelty", "practical_applicability"],
+        "axis_labels": [ax.get("name", f"Axis {i+1}") for i, ax in enumerate(aiml_axes_cfg)],
+        "prompt": _build_scoring_prompt("aiml", aiml_axes_cfg, aiml_prefs),
+    }
+
+    # Security config
+    sec_axes_cfg = _sec_cfg.get("scoring_axes", [
+        {"name": "Has Code/PoC", "weight": 0.25, "description": "Working tools, repos, artifacts"},
+        {"name": "Novel Attack Surface", "weight": 0.40, "description": "First-of-kind research"},
+        {"name": "Real-World Impact", "weight": 0.35, "description": "Affects production systems"},
+    ])
+    sec_prefs = _sec_cfg.get("preferences", {})
+    sec_weight_keys = ["code", "novelty", "impact"]
+    sec_weights = {}
+    for i, ax in enumerate(sec_axes_cfg):
+        key = sec_weight_keys[i] if i < len(sec_weight_keys) else f"axis_{i+1}"
+        sec_weights[key] = ax.get("weight", 1.0 / len(sec_axes_cfg))
+
+    configs["security"] = {
+        "weights": sec_weights,
+        "axes": ["has_code", "novel_attack_surface", "real_world_impact"],
+        "axis_labels": [ax.get("name", f"Axis {i+1}") for i, ax in enumerate(sec_axes_cfg)],
+        "prompt": _build_scoring_prompt("security", sec_axes_cfg, sec_prefs),
+    }
+
+    return configs
+
+
+SCORING_CONFIGS = _build_scoring_configs()
+
+# ---------------------------------------------------------------------------
+# Events config
+# ---------------------------------------------------------------------------
+
+RSS_FEEDS = _cfg.get("rss_feeds", [
+    {"name": "OpenAI Blog", "url": "https://openai.com/blog/rss.xml", "category": "news"},
+    {"name": "Anthropic Blog", "url": "https://www.anthropic.com/rss.xml", "category": "news"},
+    {"name": "Google DeepMind", "url": "https://deepmind.google/blog/rss.xml", "category": "news"},
+    {"name": "Meta AI", "url": "https://ai.meta.com/blog/rss/", "category": "news"},
+    {"name": "HuggingFace Blog", "url": "https://huggingface.co/blog/feed.xml", "category": "news"},
+    {"name": "Krebs on Security", "url": "https://krebsonsecurity.com/feed/", "category": "news"},
+    {"name": "The Record", "url": "https://therecord.media/feed", "category": "news"},
+    {"name": "Microsoft Security", "url": "https://www.microsoft.com/en-us/security/blog/feed/", "category": "news"},
+])
+
+CONFERENCES = _cfg.get("conferences", [
+    {"name": "NeurIPS 2026", "url": "https://neurips.cc/", "domain": "aiml",
+     "deadline": "2026-05-16", "date": "2026-12-07",
+     "description": "Conference on Neural Information Processing Systems."},
+    {"name": "ICML 2026", "url": "https://icml.cc/", "domain": "aiml",
+     "deadline": "2026-01-23", "date": "2026-07-19",
+     "description": "International Conference on Machine Learning."},
+    {"name": "ICLR 2026", "url": "https://iclr.cc/", "domain": "aiml",
+     "deadline": "2025-10-01", "date": "2026-04-24",
+     "description": "International Conference on Learning Representations."},
+    {"name": "CVPR 2026", "url": "https://cvpr.thecvf.com/", "domain": "aiml",
+     "deadline": "2025-11-14", "date": "2026-06-15",
+     "description": "IEEE/CVF Conference on Computer Vision and Pattern Recognition."},
+    {"name": "ACL 2026", "url": "https://www.aclweb.org/", "domain": "aiml",
+     "deadline": "2026-02-20", "date": "2026-08-02",
+     "description": "Annual Meeting of the Association for Computational Linguistics."},
+    {"name": "IEEE S&P 2026", "url": "https://www.ieee-security.org/TC/SP/", "domain": "security",
+     "deadline": "2026-06-05", "date": "2026-05-18",
+     "description": "IEEE Symposium on Security and Privacy."},
+    {"name": "USENIX Security 2026", "url": "https://www.usenix.org/conference/usenixsecurity/", "domain": "security",
+     "deadline": "2026-02-04", "date": "2026-08-12",
+     "description": "USENIX Security Symposium."},
+    {"name": "CCS 2026", "url": "https://www.sigsac.org/ccs/", "domain": "security",
+     "deadline": "2026-05-01", "date": "2026-11-09",
+     "description": "ACM Conference on Computer and Communications Security."},
+    {"name": "Black Hat USA 2026", "url": "https://www.blackhat.com/", "domain": "security",
+     "deadline": "2026-04-01", "date": "2026-08-04",
+     "description": "Black Hat USA."},
+    {"name": "DEF CON 34", "url": "https://defcon.org/", "domain": "security",
+     "deadline": "2026-05-01", "date": "2026-08-06",
+     "description": "DEF CON hacker conference."},
+])
+
+# ---------------------------------------------------------------------------
+# GitHub projects (OSSInsight) config
+# ---------------------------------------------------------------------------
+
+OSSINSIGHT_API = "https://api.ossinsight.io/v1"
+
+_github_cfg = _cfg.get("github", {})
+
+OSSINSIGHT_COLLECTIONS = {}
+for _coll in _github_cfg.get("collections", []):
+    if isinstance(_coll, dict):
+        OSSINSIGHT_COLLECTIONS[_coll["id"]] = (_coll["name"], _coll.get("domain", "aiml"))
+    elif isinstance(_coll, int):
+        OSSINSIGHT_COLLECTIONS[_coll] = (str(_coll), "aiml")
+
+if not OSSINSIGHT_COLLECTIONS:
+    OSSINSIGHT_COLLECTIONS = {
+        10010: ("Artificial Intelligence", "aiml"),
+        10076: ("LLM Tools", "aiml"),
+        10098: ("AI Agent Frameworks", "aiml"),
+        10087: ("LLM DevTools", "aiml"),
+        10079: ("Stable Diffusion Ecosystem", "aiml"),
+        10075: ("ChatGPT Alternatives", "aiml"),
+        10094: ("Vector Database", "aiml"),
+        10095: ("GraphRAG", "aiml"),
+        10099: ("MCP Client", "aiml"),
+        10058: ("MLOps Tools", "aiml"),
+        10051: ("Security Tool", "security"),
+        10082: ("Web Scanner", "security"),
+    }
+
+OSSINSIGHT_TRENDING_LANGUAGES = ["Python", "Rust", "Go", "TypeScript", "C++"]
+
+GITHUB_AIML_KEYWORDS = re.compile(
+    r"machine.learn|deep.learn|neural.net|transformer|llm|large.language|"
+    r"diffusion|generat.ai|gpt|bert|llama|vision.model|multimodal|"
+    r"reinforcement.learn|computer.vision|nlp|natural.language|"
+    r"text.to|speech.to|image.generat|video.generat|"
+    r"fine.tun|training|inference|quantiz|embedding|vector|"
+    r"rag|retrieval.augment|agent|langchain|"
+    r"hugging.?face|pytorch|tensorflow|jax|"
+    r"stable.diffusion|comfyui|ollama|vllm|"
+    r"tokeniz|dataset|benchmark|model.serv|mlops",
+    re.IGNORECASE,
+)
+
+GITHUB_SECURITY_KEYWORDS = re.compile(
+    r"security|pentest|penetration.test|vulnerability|exploit|"
+    r"fuzzing|fuzz|malware|scanner|scanning|"
+    r"intrusion|ransomware|phishing|"
+    r"reverse.engineer|decompil|disassembl|"
+    r"ctf|capture.the.flag|"
+    r"firewall|ids\b|ips\b|siem|"
+    r"password|credential|auth|"
+    r"xss|csrf|injection|"
+    r"osint|reconnaissance|recon|"
+    r"forensic|incident.response|"
+    r"encryption|cryptograph|"
+    r"burp|nuclei|nmap|metasploit|wireshark",
+    re.IGNORECASE,
+)
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+def get_enabled_domains() -> list[str]:
+    """Return list of enabled domain keys."""
+    if not _domains_cfg:
+        return ["aiml", "security"]
+    return [k for k, v in _domains_cfg.items() if v.get("enabled", True)]
+
+
+def get_domain_label(domain: str) -> str:
+    """Return human-readable label for a domain."""
+    if _domains_cfg and domain in _domains_cfg:
+        return _domains_cfg[domain].get("label", domain.upper())
+    return {"aiml": "AI/ML", "security": "Security"}.get(domain, domain.upper())
+
+
+def save_config(data: dict):
+    """Write config data to config.yaml."""
+    import yaml
+    with open(CONFIG_PATH, "w") as f:
+        yaml.dump(data, f, default_flow_style=False, sort_keys=False)
+    log.info("Config saved to %s", CONFIG_PATH)
+    global _cfg, FIRST_RUN, SCORING_CONFIGS
+    _cfg = data
+    FIRST_RUN = False
+    SCORING_CONFIGS.update(_build_scoring_configs())

src/db.py

@@ -0,0 +1,870 @@
+"""Database layer — SQLite schema, connection, and query helpers."""
+
+import json
+import logging
+import sqlite3
+from contextlib import contextmanager
+from datetime import datetime, timezone
+from pathlib import Path
+
+log = logging.getLogger(__name__)
+
+
+def get_db_path() -> Path:
+    from src.config import DB_PATH
+    return DB_PATH
+
+
+@contextmanager
+def get_conn():
+    """Yield a SQLite connection with WAL mode and foreign keys."""
+    path = get_db_path()
+    path.parent.mkdir(parents=True, exist_ok=True)
+    conn = sqlite3.connect(str(path))
+    conn.row_factory = sqlite3.Row
+    conn.execute("PRAGMA journal_mode=WAL")
+    conn.execute("PRAGMA foreign_keys=ON")
+    try:
+        yield conn
+        conn.commit()
+    except Exception:
+        conn.rollback()
+        log.exception("Database transaction failed")
+        raise
+    finally:
+        conn.close()
+
+
+def init_db():
+    """Create tables if they don't exist."""
+    with get_conn() as conn:
+        conn.executescript(SCHEMA)
+        for sql in _MIGRATIONS:
+            try:
+                conn.execute(sql)
+            except sqlite3.OperationalError as e:
+                if "duplicate column" in str(e).lower() or "already exists" in str(e).lower():
+                    pass  # Expected — column/index already exists
+                else:
+                    log.warning("Migration failed: %s — %s", sql.strip()[:60], e)
+
+
+SCHEMA = """\
+CREATE TABLE IF NOT EXISTS runs (
+    id INTEGER PRIMARY KEY,
+    domain TEXT NOT NULL,
+    started_at TEXT NOT NULL,
+    finished_at TEXT,
+    date_start TEXT NOT NULL,
+    date_end TEXT NOT NULL,
+    paper_count INTEGER DEFAULT 0,
+    status TEXT DEFAULT 'running'
+);
+
+CREATE TABLE IF NOT EXISTS papers (
+    id INTEGER PRIMARY KEY,
+    run_id INTEGER REFERENCES runs(id),
+    domain TEXT NOT NULL,
+    arxiv_id TEXT NOT NULL,
+    entry_id TEXT,
+    title TEXT NOT NULL,
+    authors TEXT,
+    abstract TEXT,
+    published TEXT,
+    categories TEXT,
+    pdf_url TEXT,
+    arxiv_url TEXT,
+    comment TEXT,
+    source TEXT,
+    github_repo TEXT,
+    github_stars INTEGER,
+    hf_upvotes INTEGER DEFAULT 0,
+    hf_models TEXT,
+    hf_datasets TEXT,
+    hf_spaces TEXT,
+    score_axis_1 REAL,
+    score_axis_2 REAL,
+    score_axis_3 REAL,
+    composite REAL,
+    summary TEXT,
+    reasoning TEXT,
+    code_url TEXT,
+    UNIQUE(domain, arxiv_id, run_id)
+);
+
+CREATE TABLE IF NOT EXISTS events (
+    id INTEGER PRIMARY KEY,
+    run_id INTEGER,
+    category TEXT NOT NULL,
+    title TEXT NOT NULL,
+    description TEXT,
+    url TEXT,
+    event_date TEXT,
+    source TEXT,
+    relevance_score REAL,
+    fetched_at TEXT NOT NULL
+);
+
+CREATE TABLE IF NOT EXISTS paper_connections (
+    id INTEGER PRIMARY KEY,
+    paper_id INTEGER NOT NULL REFERENCES papers(id),
+    connected_arxiv_id TEXT,
+    connected_s2_id TEXT,
+    connected_title TEXT,
+    connected_year INTEGER,
+    connection_type TEXT NOT NULL,
+    in_db_paper_id INTEGER,
+    fetched_at TEXT NOT NULL
+);
+
+CREATE INDEX IF NOT EXISTS idx_papers_domain_composite
+    ON papers(domain, composite DESC);
+CREATE INDEX IF NOT EXISTS idx_papers_run ON papers(run_id);
+CREATE INDEX IF NOT EXISTS idx_events_category ON events(category, event_date);
+CREATE INDEX IF NOT EXISTS idx_connections_paper ON paper_connections(paper_id);
+CREATE INDEX IF NOT EXISTS idx_connections_arxiv ON paper_connections(connected_arxiv_id);
+CREATE INDEX IF NOT EXISTS idx_papers_arxiv_id ON papers(arxiv_id);
+CREATE INDEX IF NOT EXISTS idx_papers_published ON papers(published);
+CREATE INDEX IF NOT EXISTS idx_events_run_id ON events(run_id);
+
+CREATE TABLE IF NOT EXISTS github_projects (
+    id INTEGER PRIMARY KEY,
+    run_id INTEGER REFERENCES runs(id),
+    repo_id INTEGER NOT NULL,
+    repo_name TEXT NOT NULL,
+    description TEXT,
+    language TEXT,
+    stars INTEGER DEFAULT 0,
+    forks INTEGER DEFAULT 0,
+    pull_requests INTEGER DEFAULT 0,
+    total_score REAL DEFAULT 0,
+    collection_names TEXT,
+    topics TEXT DEFAULT '[]',
+    url TEXT NOT NULL,
+    domain TEXT,
+    fetched_at TEXT NOT NULL,
+    UNIQUE(repo_name, run_id)
+);
+
+CREATE INDEX IF NOT EXISTS idx_gh_run ON github_projects(run_id);
+CREATE INDEX IF NOT EXISTS idx_gh_domain ON github_projects(domain, total_score DESC);
+CREATE INDEX IF NOT EXISTS idx_gh_repo ON github_projects(repo_name);
+
+CREATE TABLE IF NOT EXISTS user_signals (
+    id INTEGER PRIMARY KEY,
+    paper_id INTEGER NOT NULL REFERENCES papers(id),
+    action TEXT NOT NULL CHECK(action IN ('save','view','upvote','downvote','dismiss')),
+    created_at TEXT NOT NULL,
+    metadata TEXT DEFAULT '{}'
+);
+
+CREATE UNIQUE INDEX IF NOT EXISTS idx_signals_paper_action
+    ON user_signals(paper_id, action) WHERE action != 'view';
+CREATE INDEX IF NOT EXISTS idx_signals_created ON user_signals(created_at);
+CREATE INDEX IF NOT EXISTS idx_signals_paper ON user_signals(paper_id);
+
+CREATE TABLE IF NOT EXISTS user_preferences (
+    id INTEGER PRIMARY KEY,
+    pref_key TEXT NOT NULL UNIQUE,
+    pref_value REAL NOT NULL DEFAULT 0.0,
+    signal_count INTEGER NOT NULL DEFAULT 0,
+    updated_at TEXT NOT NULL
+);
+
+CREATE INDEX IF NOT EXISTS idx_prefs_key ON user_preferences(pref_key);
+"""
+
+# Columns added after initial schema — idempotent via try/except
+_MIGRATIONS = [
+    "ALTER TABLE papers ADD COLUMN s2_tldr TEXT",
+    "ALTER TABLE papers ADD COLUMN s2_paper_id TEXT",
+    "ALTER TABLE papers ADD COLUMN topics TEXT DEFAULT '[]'",
+    "CREATE UNIQUE INDEX IF NOT EXISTS idx_events_unique ON events(title, category)",
+]
+
+
+# ---------------------------------------------------------------------------
+# Run helpers
+# ---------------------------------------------------------------------------
+
+def create_run(domain: str, date_start: str, date_end: str) -> int:
+    """Insert a new pipeline run, return its ID."""
+    now = datetime.now(timezone.utc).isoformat()
+    with get_conn() as conn:
+        cur = conn.execute(
+            "INSERT INTO runs (domain, started_at, date_start, date_end, status) "
+            "VALUES (?, ?, ?, ?, 'running')",
+            (domain, now, date_start, date_end),
+        )
+        return cur.lastrowid
+
+
+def finish_run(run_id: int, paper_count: int, status: str = "completed"):
+    now = datetime.now(timezone.utc).isoformat()
+    with get_conn() as conn:
+        conn.execute(
+            "UPDATE runs SET finished_at=?, paper_count=?, status=? WHERE id=?",
+            (now, paper_count, status, run_id),
+        )
+
+
+def get_latest_run(domain: str) -> dict | None:
+    with get_conn() as conn:
+        row = conn.execute(
+            "SELECT * FROM runs WHERE domain=? ORDER BY id DESC LIMIT 1",
+            (domain,),
+        ).fetchone()
+        return dict(row) if row else None
+
+
+def get_run(run_id: int) -> dict | None:
+    with get_conn() as conn:
+        row = conn.execute("SELECT * FROM runs WHERE id=?", (run_id,)).fetchone()
+        return dict(row) if row else None
+
+
+# ---------------------------------------------------------------------------
+# Paper helpers
+# ---------------------------------------------------------------------------
+
+def _serialize_json(val):
+    """JSON-encode lists/dicts for storage."""
+    if isinstance(val, (list, dict)):
+        return json.dumps(val)
+    return val
+
+
+def insert_papers(papers: list[dict], run_id: int, domain: str):
+    """Bulk-insert papers into the DB."""
+    with get_conn() as conn:
+        for p in papers:
+            conn.execute(
+                """INSERT OR IGNORE INTO papers
+                   (run_id, domain, arxiv_id, entry_id, title, authors, abstract,
+                    published, categories, pdf_url, arxiv_url, comment, source,
+                    github_repo, github_stars, hf_upvotes, hf_models, hf_datasets, hf_spaces)
+                   VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)""",
+                (
+                    run_id, domain,
+                    p.get("arxiv_id", ""),
+                    p.get("entry_id", ""),
+                    p.get("title", ""),
+                    _serialize_json(p.get("authors", [])),
+                    p.get("abstract", ""),
+                    p.get("published", ""),
+                    _serialize_json(p.get("categories", [])),
+                    p.get("pdf_url", ""),
+                    p.get("arxiv_url", ""),
+                    p.get("comment", ""),
+                    p.get("source", ""),
+                    p.get("github_repo", ""),
+                    p.get("github_stars"),
+                    p.get("hf_upvotes", 0),
+                    _serialize_json(p.get("hf_models", [])),
+                    _serialize_json(p.get("hf_datasets", [])),
+                    _serialize_json(p.get("hf_spaces", [])),
+                ),
+            )
+
+
+def update_paper_scores(paper_id: int, scores: dict):
+    """Update a paper's scores after Claude scoring."""
+    with get_conn() as conn:
+        conn.execute(
+            """UPDATE papers SET
+               score_axis_1=?, score_axis_2=?, score_axis_3=?,
+               composite=?, summary=?, reasoning=?, code_url=?
+               WHERE id=?""",
+            (
+                scores.get("score_axis_1"),
+                scores.get("score_axis_2"),
+                scores.get("score_axis_3"),
+                scores.get("composite"),
+                scores.get("summary", ""),
+                scores.get("reasoning", ""),
+                scores.get("code_url"),
+                paper_id,
+            ),
+        )
+
+
+def get_unscored_papers(run_id: int) -> list[dict]:
+    """Get papers from a run that haven't been scored yet."""
+    with get_conn() as conn:
+        rows = conn.execute(
+            "SELECT * FROM papers WHERE run_id=? AND composite IS NULL",
+            (run_id,),
+        ).fetchall()
+        return [_deserialize_paper(row) for row in rows]
+
+
+def get_top_papers(domain: str, run_id: int | None = None, limit: int = 20) -> list[dict]:
+    """Get top-scored papers for a domain, optionally from a specific run."""
+    with get_conn() as conn:
+        if run_id:
+            rows = conn.execute(
+                "SELECT * FROM papers WHERE domain=? AND run_id=? AND composite IS NOT NULL "
+                "ORDER BY composite DESC LIMIT ?",
+                (domain, run_id, limit),
+            ).fetchall()
+        else:
+            # Latest run
+            latest = get_latest_run(domain)
+            if not latest:
+                return []
+            rows = conn.execute(
+                "SELECT * FROM papers WHERE domain=? AND run_id=? AND composite IS NOT NULL "
+                "ORDER BY composite DESC LIMIT ?",
+                (domain, latest["id"], limit),
+            ).fetchall()
+        return [_deserialize_paper(row) for row in rows]
+
+
+def get_paper(paper_id: int) -> dict | None:
+    with get_conn() as conn:
+        row = conn.execute("SELECT * FROM papers WHERE id=?", (paper_id,)).fetchone()
+        return _deserialize_paper(row) if row else None
+
+
+SORT_OPTIONS = {
+    "score": "composite DESC",
+    "date": "published DESC",
+    "axis1": "score_axis_1 DESC",
+    "axis2": "score_axis_2 DESC",
+    "axis3": "score_axis_3 DESC",
+    "title": "title ASC",
+}
+
+
+def get_papers_page(domain: str, run_id: int | None = None,
+                    offset: int = 0, limit: int = 50,
+                    min_score: float | None = None,
+                    has_code: bool | None = None,
+                    search: str | None = None,
+                    topic: str | None = None,
+                    sort: str | None = None) -> tuple[list[dict], int]:
+    """Paginated, filterable paper list. Returns (papers, total_count)."""
+    with get_conn() as conn:
+        if not run_id:
+            latest = get_latest_run(domain)
+            if not latest:
+                return [], 0
+            run_id = latest["id"]
+
+        conditions = ["domain=?", "run_id=?", "composite IS NOT NULL"]
+        params: list = [domain, run_id]
+
+        if min_score is not None:
+            conditions.append("composite >= ?")
+            params.append(min_score)
+
+        if has_code:
+            conditions.append("(code_url IS NOT NULL AND code_url != '')")
+
+        if search:
+            conditions.append("(title LIKE ? OR abstract LIKE ?)")
+            params.extend([f"%{search}%", f"%{search}%"])
+
+        if topic:
+            conditions.append("topics LIKE ?")
+            params.append(f'%"{topic}"%')
+
+        where = " AND ".join(conditions)
+        order = SORT_OPTIONS.get(sort, "composite DESC")
+
+        total = conn.execute(
+            f"SELECT COUNT(*) FROM papers WHERE {where}", params
+        ).fetchone()[0]
+
+        rows = conn.execute(
+            f"SELECT * FROM papers WHERE {where} ORDER BY {order} LIMIT ? OFFSET ?",
+            params + [limit, offset],
+        ).fetchall()
+
+        return [_deserialize_paper(row) for row in rows], total
+
+
+def count_papers(domain: str, run_id: int | None = None, scored_only: bool = False) -> int:
+    with get_conn() as conn:
+        if not run_id:
+            latest = get_latest_run(domain)
+            if not latest:
+                return 0
+            run_id = latest["id"]
+        sql = "SELECT COUNT(*) FROM papers WHERE domain=? AND run_id=?"
+        if scored_only:
+            sql += " AND composite IS NOT NULL"
+        row = conn.execute(sql, (domain, run_id)).fetchone()
+        return row[0] if row else 0
+
+
+def _deserialize_paper(row) -> dict:
+    """Convert a sqlite3.Row to a dict, parsing JSON fields."""
+    d = dict(row)
+    for key in ("authors", "categories", "hf_models", "hf_datasets", "hf_spaces", "topics"):
+        val = d.get(key)
+        if isinstance(val, str):
+            try:
+                d[key] = json.loads(val)
+            except (json.JSONDecodeError, TypeError):
+                d[key] = []
+    return d
+
+
+# ---------------------------------------------------------------------------
+# Event helpers
+# ---------------------------------------------------------------------------
+
+def insert_events(events: list[dict], run_id: int | None = None):
+    now = datetime.now(timezone.utc).isoformat()
+    with get_conn() as conn:
+        for e in events:
+            conn.execute(
+                """INSERT OR IGNORE INTO events
+                   (run_id, category, title, description, url, event_date,
+                    source, relevance_score, fetched_at)
+                   VALUES (?,?,?,?,?,?,?,?,?)""",
+                (
+                    run_id,
+                    e.get("category", ""),
+                    e.get("title", ""),
+                    e.get("description", ""),
+                    e.get("url", ""),
+                    e.get("event_date", ""),
+                    e.get("source", ""),
+                    e.get("relevance_score"),
+                    now,
+                ),
+            )
+
+
+def get_events(category: str | None = None, limit: int = 50) -> list[dict]:
+    with get_conn() as conn:
+        if category:
+            rows = conn.execute(
+                "SELECT * FROM events WHERE category=? ORDER BY event_date DESC LIMIT ?",
+                (category, limit),
+            ).fetchall()
+        else:
+            rows = conn.execute(
+                "SELECT * FROM events ORDER BY fetched_at DESC LIMIT ?",
+                (limit,),
+            ).fetchall()
+        return [dict(row) for row in rows]
+
+
+def count_events() -> int:
+    with get_conn() as conn:
+        return conn.execute("SELECT COUNT(*) FROM events").fetchone()[0]
+
+
+# ---------------------------------------------------------------------------
+# Dashboard helpers
+# ---------------------------------------------------------------------------
+
+def get_all_runs(limit: int = 20) -> list[dict]:
+    with get_conn() as conn:
+        rows = conn.execute(
+            "SELECT * FROM runs ORDER BY id DESC LIMIT ?", (limit,)
+        ).fetchall()
+        return [dict(row) for row in rows]
+
+
+# ---------------------------------------------------------------------------
+# Paper connections (Semantic Scholar)
+# ---------------------------------------------------------------------------
+
+def insert_connections(connections: list[dict]):
+    """Bulk-insert paper connections."""
+    now = datetime.now(timezone.utc).isoformat()
+    with get_conn() as conn:
+        for c in connections:
+            conn.execute(
+                """INSERT INTO paper_connections
+                   (paper_id, connected_arxiv_id, connected_s2_id,
+                    connected_title, connected_year, connection_type,
+                    in_db_paper_id, fetched_at)
+                   VALUES (?,?,?,?,?,?,?,?)""",
+                (
+                    c["paper_id"],
+                    c.get("connected_arxiv_id", ""),
+                    c.get("connected_s2_id", ""),
+                    c.get("connected_title", ""),
+                    c.get("connected_year"),
+                    c["connection_type"],
+                    c.get("in_db_paper_id"),
+                    now,
+                ),
+            )
+
+
+def get_paper_connections(paper_id: int) -> dict:
+    """Get connected papers grouped by type."""
+    with get_conn() as conn:
+        rows = conn.execute(
+            "SELECT * FROM paper_connections WHERE paper_id=? "
+            "ORDER BY connection_type, connected_year DESC",
+            (paper_id,),
+        ).fetchall()
+
+    result = {"references": [], "recommendations": []}
+    for row in rows:
+        d = dict(row)
+        ctype = d["connection_type"]
+        if ctype in result:
+            result[ctype].append(d)
+    return result
+
+
+def clear_connections(paper_id: int):
+    """Remove existing connections for a paper (before re-enrichment)."""
+    with get_conn() as conn:
+        conn.execute("DELETE FROM paper_connections WHERE paper_id=?", (paper_id,))
+
+
+def update_paper_s2(paper_id: int, s2_paper_id: str, s2_tldr: str):
+    """Update S2 metadata on a paper."""
+    with get_conn() as conn:
+        conn.execute(
+            "UPDATE papers SET s2_paper_id=?, s2_tldr=? WHERE id=?",
+            (s2_paper_id, s2_tldr, paper_id),
+        )
+
+
+def update_paper_topics(paper_id: int, topics: list[str]):
+    """Update topic tags on a paper."""
+    with get_conn() as conn:
+        conn.execute(
+            "UPDATE papers SET topics=? WHERE id=?",
+            (json.dumps(topics), paper_id),
+        )
+
+
+def get_arxiv_id_map(run_id: int) -> dict[str, int]:
+    """Return {arxiv_id: paper_db_id} for all papers in a run."""
+    with get_conn() as conn:
+        rows = conn.execute(
+            "SELECT id, arxiv_id FROM papers WHERE run_id=?", (run_id,)
+        ).fetchall()
+        return {row["arxiv_id"]: row["id"] for row in rows}
+
+
+def get_available_topics(domain: str, run_id: int) -> list[str]:
+    """Get distinct topic tags used in a run."""
+    with get_conn() as conn:
+        rows = conn.execute(
+            "SELECT DISTINCT topics FROM papers "
+            "WHERE domain=? AND run_id=? AND topics IS NOT NULL AND topics != '[]'",
+            (domain, run_id),
+        ).fetchall()
+
+    all_topics: set[str] = set()
+    for row in rows:
+        try:
+            all_topics.update(json.loads(row["topics"]))
+        except (json.JSONDecodeError, TypeError):
+            pass
+    return sorted(all_topics)
+
+
+# ---------------------------------------------------------------------------
+# GitHub project helpers
+# ---------------------------------------------------------------------------
+
+def insert_github_projects(projects: list[dict], run_id: int):
+    """Bulk-insert GitHub projects into the DB."""
+    now = datetime.now(timezone.utc).isoformat()
+    with get_conn() as conn:
+        for p in projects:
+            conn.execute(
+                """INSERT OR IGNORE INTO github_projects
+                   (run_id, repo_id, repo_name, description, language,
+                    stars, forks, pull_requests, total_score,
+                    collection_names, topics, url, domain, fetched_at)
+                   VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?)""",
+                (
+                    run_id,
+                    p.get("repo_id", 0),
+                    p.get("repo_name", ""),
+                    p.get("description", ""),
+                    p.get("language", ""),
+                    p.get("stars", 0),
+                    p.get("forks", 0),
+                    p.get("pull_requests", 0),
+                    p.get("total_score", 0),
+                    p.get("collection_names", ""),
+                    _serialize_json(p.get("topics", [])),
+                    p.get("url", ""),
+                    p.get("domain", ""),
+                    now,
+                ),
+            )
+
+
+GH_SORT_OPTIONS = {
+    "score": "total_score DESC",
+    "stars": "stars DESC",
+    "forks": "forks DESC",
+    "name": "repo_name ASC",
+}
+
+
+def get_github_projects_page(
+    run_id: int | None = None,
+    offset: int = 0,
+    limit: int = 50,
+    search: str | None = None,
+    language: str | None = None,
+    domain: str | None = None,
+    sort: str | None = None,
+) -> tuple[list[dict], int]:
+    """Paginated, filterable GitHub project list."""
+    with get_conn() as conn:
+        if not run_id:
+            latest = get_latest_run("github")
+            if not latest:
+                return [], 0
+            run_id = latest["id"]
+
+        conditions = ["run_id=?"]
+        params: list = [run_id]
+
+        if search:
+            conditions.append("(repo_name LIKE ? OR description LIKE ?)")
+            params.extend([f"%{search}%", f"%{search}%"])
+
+        if language:
+            conditions.append("language=?")
+            params.append(language)
+
+        if domain:
+            conditions.append("domain=?")
+            params.append(domain)
+
+        where = " AND ".join(conditions)
+        order = GH_SORT_OPTIONS.get(sort, "total_score DESC")
+
+        total = conn.execute(
+            f"SELECT COUNT(*) FROM github_projects WHERE {where}", params
+        ).fetchone()[0]
+
+        rows = conn.execute(
+            f"SELECT * FROM github_projects WHERE {where} ORDER BY {order} LIMIT ? OFFSET ?",
+            params + [limit, offset],
+        ).fetchall()
+
+        return [_deserialize_gh_project(row) for row in rows], total
+
+
+def get_top_github_projects(run_id: int | None = None, limit: int = 10) -> list[dict]:
+    """Get top GitHub projects by score."""
+    with get_conn() as conn:
+        if not run_id:
+            latest = get_latest_run("github")
+            if not latest:
+                return []
+            run_id = latest["id"]
+        rows = conn.execute(
+            "SELECT * FROM github_projects WHERE run_id=? ORDER BY total_score DESC LIMIT ?",
+            (run_id, limit),
+        ).fetchall()
+        return [_deserialize_gh_project(row) for row in rows]
+
+
+def count_github_projects(run_id: int | None = None) -> int:
+    with get_conn() as conn:
+        if not run_id:
+            latest = get_latest_run("github")
+            if not latest:
+                return 0
+            run_id = latest["id"]
+        return conn.execute(
+            "SELECT COUNT(*) FROM github_projects WHERE run_id=?", (run_id,)
+        ).fetchone()[0]
+
+
+def get_github_languages(run_id: int) -> list[str]:
+    """Get distinct languages in a GitHub run."""
+    with get_conn() as conn:
+        rows = conn.execute(
+            "SELECT DISTINCT language FROM github_projects "
+            "WHERE run_id=? AND language IS NOT NULL AND language != '' "
+            "ORDER BY language",
+            (run_id,),
+        ).fetchall()
+        return [row["language"] for row in rows]
+
+
+def _deserialize_gh_project(row) -> dict:
+    d = dict(row)
+    for key in ("topics",):
+        val = d.get(key)
+        if isinstance(val, str):
+            try:
+                d[key] = json.loads(val)
+            except (json.JSONDecodeError, TypeError):
+                d[key] = []
+    return d
+
+
+# ---------------------------------------------------------------------------
+# User signal helpers (preference learning)
+# ---------------------------------------------------------------------------
+
+def insert_signal(paper_id: int, action: str, metadata: dict | None = None) -> bool:
+    """Record a user signal. Returns True if inserted, False if duplicate.
+
+    Views are deduped by 5-minute window. Other actions use UNIQUE constraint.
+    """
+    now = datetime.now(timezone.utc).isoformat()
+    meta_json = json.dumps(metadata or {})
+    with get_conn() as conn:
+        if action == "view":
+            # Dedup views within 5-minute window
+            recent = conn.execute(
+                "SELECT 1 FROM user_signals "
+                "WHERE paper_id=? AND action='view' "
+                "AND created_at > datetime(?, '-5 minutes')",
+                (paper_id, now),
+            ).fetchone()
+            if recent:
+                return False
+            conn.execute(
+                "INSERT INTO user_signals (paper_id, action, created_at, metadata) "
+                "VALUES (?, ?, ?, ?)",
+                (paper_id, action, now, meta_json),
+            )
+            return True
+        else:
+            try:
+                conn.execute(
+                    "INSERT INTO user_signals (paper_id, action, created_at, metadata) "
+                    "VALUES (?, ?, ?, ?)",
+                    (paper_id, action, now, meta_json),
+                )
+                return True
+            except sqlite3.IntegrityError:
+                return False
+
+
+def delete_signal(paper_id: int, action: str) -> bool:
+    """Remove a signal (for toggling off). Returns True if deleted."""
+    with get_conn() as conn:
+        cur = conn.execute(
+            "DELETE FROM user_signals WHERE paper_id=? AND action=?",
+            (paper_id, action),
+        )
+        return cur.rowcount > 0
+
+
+def get_paper_signal(paper_id: int) -> str | None:
+    """Return the user's latest non-view signal for a paper, or None."""
+    with get_conn() as conn:
+        row = conn.execute(
+            "SELECT action FROM user_signals "
+            "WHERE paper_id=? AND action != 'view' "
+            "ORDER BY created_at DESC LIMIT 1",
+            (paper_id,),
+        ).fetchone()
+        return row["action"] if row else None
+
+
+def get_paper_signals_batch(paper_ids: list[int]) -> dict[int, str]:
+    """Batch fetch latest non-view signal per paper. Returns {paper_id: action}."""
+    if not paper_ids:
+        return {}
+    with get_conn() as conn:
+        placeholders = ",".join("?" for _ in paper_ids)
+        rows = conn.execute(
+            f"SELECT paper_id, action FROM user_signals "
+            f"WHERE paper_id IN ({placeholders}) AND action != 'view' "
+            f"ORDER BY created_at DESC",
+            paper_ids,
+        ).fetchall()
+    result: dict[int, str] = {}
+    for row in rows:
+        pid = row["paper_id"]
+        if pid not in result:
+            result[pid] = row["action"]
+    return result
+
+
+def get_all_signals_with_papers() -> list[dict]:
+    """Join signals with paper data for preference computation."""
+    with get_conn() as conn:
+        rows = conn.execute(
+            """SELECT s.id as signal_id, s.paper_id, s.action, s.created_at,
+                      p.title, p.categories, p.topics, p.authors, p.domain,
+                      p.score_axis_1, p.score_axis_2, p.score_axis_3, p.composite
+               FROM user_signals s
+               JOIN papers p ON s.paper_id = p.id
+               ORDER BY s.created_at DESC"""
+        ).fetchall()
+    results = []
+    for row in rows:
+        d = dict(row)
+        for key in ("categories", "topics", "authors"):
+            val = d.get(key)
+            if isinstance(val, str):
+                try:
+                    d[key] = json.loads(val)
+                except (json.JSONDecodeError, TypeError):
+                    d[key] = []
+        results.append(d)
+    return results
+
+
+def get_signal_counts() -> dict[str, int]:
+    """Summary stats: count per action type."""
+    with get_conn() as conn:
+        rows = conn.execute(
+            "SELECT action, COUNT(*) as cnt FROM user_signals GROUP BY action"
+        ).fetchall()
+    return {row["action"]: row["cnt"] for row in rows}
+
+
+def save_preferences(prefs: dict[str, tuple[float, int]]):
+    """Bulk write preferences. prefs = {key: (value, signal_count)}."""
+    now = datetime.now(timezone.utc).isoformat()
+    with get_conn() as conn:
+        conn.execute("DELETE FROM user_preferences")
+        for key, (value, count) in prefs.items():
+            conn.execute(
+                "INSERT INTO user_preferences (pref_key, pref_value, signal_count, updated_at) "
+                "VALUES (?, ?, ?, ?)",
+                (key, value, count, now),
+            )
+
+
+def load_preferences() -> dict[str, float]:
+    """Load preference profile. Returns {pref_key: pref_value}."""
+    with get_conn() as conn:
+        rows = conn.execute(
+            "SELECT pref_key, pref_value FROM user_preferences"
+        ).fetchall()
+    return {row["pref_key"]: row["pref_value"] for row in rows}
+
+
+def get_preferences_detail() -> list[dict]:
+    """Load full preference details for the preferences page."""
+    with get_conn() as conn:
+        rows = conn.execute(
+            "SELECT * FROM user_preferences ORDER BY ABS(pref_value) DESC"
+        ).fetchall()
+    return [dict(row) for row in rows]
+
+
+def get_preferences_updated_at() -> str | None:
+    """Return when preferences were last computed."""
+    with get_conn() as conn:
+        row = conn.execute(
+            "SELECT updated_at FROM user_preferences ORDER BY updated_at DESC LIMIT 1"
+        ).fetchone()
+        return row["updated_at"] if row else None
+
+
+def clear_preferences():
+    """Reset all preferences and signals."""
+    with get_conn() as conn:
+        conn.execute("DELETE FROM user_preferences")
+        conn.execute("DELETE FROM user_signals")

src/pipelines/aiml.py

@@ -0,0 +1,327 @@
+"""AI/ML paper pipeline.
+
+Fetches papers from HuggingFace Daily Papers + arXiv, enriches with
+HF ecosystem metadata, and writes to the database.
+"""
+
+import logging
+import re
+import time
+from datetime import datetime, timedelta, timezone
+
+import arxiv
+import requests
+
+from src.config import (
+    ARXIV_LARGE_CATS,
+    ARXIV_SMALL_CATS,
+    EXCLUDE_RE,
+    GITHUB_URL_RE,
+    HF_API,
+    HF_MAX_AGE_DAYS,
+    INCLUDE_RE,
+    MAX_ABSTRACT_CHARS_AIML,
+)
+from src.db import create_run, finish_run, insert_papers
+
+log = logging.getLogger(__name__)
+
+
+# ---------------------------------------------------------------------------
+# HuggingFace API
+# ---------------------------------------------------------------------------
+
+
+def fetch_hf_daily(date_str: str) -> list[dict]:
+    """Fetch HF Daily Papers for a given date."""
+    url = f"{HF_API}/daily_papers?date={date_str}"
+    try:
+        resp = requests.get(url, timeout=30)
+        resp.raise_for_status()
+        return resp.json()
+    except (requests.RequestException, ValueError):
+        return []
+
+
+def fetch_hf_trending(limit: int = 50) -> list[dict]:
+    """Fetch HF trending papers."""
+    url = f"{HF_API}/daily_papers?sort=trending&limit={limit}"
+    try:
+        resp = requests.get(url, timeout=30)
+        resp.raise_for_status()
+        return resp.json()
+    except (requests.RequestException, ValueError):
+        return []
+
+
+def arxiv_id_to_date(arxiv_id: str) -> datetime | None:
+    """Extract approximate publication date from arXiv ID (YYMM.NNNNN)."""
+    match = re.match(r"(\d{2})(\d{2})\.\d+", arxiv_id)
+    if not match:
+        return None
+    year = 2000 + int(match.group(1))
+    month = int(match.group(2))
+    if not (1 <= month <= 12):
+        return None
+    return datetime(year, month, 1, tzinfo=timezone.utc)
+
+
+def normalize_hf_paper(hf_entry: dict) -> dict | None:
+    """Convert an HF daily_papers entry to our normalized format.
+
+    Returns None if the paper is too old.
+    """
+    paper = hf_entry.get("paper", hf_entry)
+    arxiv_id = paper.get("id", "")
+
+    authors_raw = paper.get("authors", [])
+    authors = []
+    for a in authors_raw:
+        if isinstance(a, dict):
+            name = a.get("name", a.get("user", {}).get("fullname", ""))
+            if name:
+                authors.append(name)
+        elif isinstance(a, str):
+            authors.append(a)
+
+    github_repo = hf_entry.get("githubRepo") or paper.get("githubRepo") or ""
+
+    pub_date = arxiv_id_to_date(arxiv_id)
+    if pub_date and (datetime.now(timezone.utc) - pub_date).days > HF_MAX_AGE_DAYS:
+        return None
+
+    return {
+        "arxiv_id": arxiv_id,
+        "title": paper.get("title", "").replace("\n", " ").strip(),
+        "authors": authors[:10],
+        "abstract": paper.get("summary", paper.get("abstract", "")).replace("\n", " ").strip(),
+        "published": paper.get("publishedAt", paper.get("published", "")),
+        "categories": paper.get("categories", []),
+        "pdf_url": f"https://arxiv.org/pdf/{arxiv_id}" if arxiv_id else "",
+        "arxiv_url": f"https://arxiv.org/abs/{arxiv_id}" if arxiv_id else "",
+        "comment": "",
+        "source": "hf",
+        "hf_upvotes": hf_entry.get("paper", {}).get("upvotes", hf_entry.get("upvotes", 0)),
+        "github_repo": github_repo,
+        "github_stars": None,
+        "hf_models": [],
+        "hf_datasets": [],
+        "hf_spaces": [],
+    }
+
+
+# ---------------------------------------------------------------------------
+# arXiv fetching
+# ---------------------------------------------------------------------------
+
+
+def fetch_arxiv_category(
+    cat: str,
+    start: datetime,
+    end: datetime,
+    max_results: int,
+    filter_keywords: bool,
+) -> list[dict]:
+    """Fetch papers from a single arXiv category."""
+    client = arxiv.Client(page_size=200, delay_seconds=3.0, num_retries=3)
+
+    query = arxiv.Search(
+        query=f"cat:{cat}",
+        max_results=max_results,
+        sort_by=arxiv.SortCriterion.SubmittedDate,
+        sort_order=arxiv.SortOrder.Descending,
+    )
+
+    papers = []
+    for result in client.results(query):
+        pub = result.published.replace(tzinfo=timezone.utc)
+        if pub < start:
+            break
+        if pub > end:
+            continue
+
+        if filter_keywords:
+            text = f"{result.title} {result.summary}"
+            if not INCLUDE_RE.search(text):
+                continue
+            if EXCLUDE_RE.search(text):
+                continue
+
+        papers.append(_arxiv_result_to_dict(result))
+
+    return papers
+
+
+def _arxiv_result_to_dict(result: arxiv.Result) -> dict:
+    """Convert an arxiv.Result to our normalized format."""
+    arxiv_id = result.entry_id.split("/abs/")[-1]
+    base_id = re.sub(r"v\d+$", "", arxiv_id)
+
+    github_urls = GITHUB_URL_RE.findall(f"{result.summary} {result.comment or ''}")
+    github_repo = github_urls[0].rstrip(".") if github_urls else ""
+
+    return {
+        "arxiv_id": base_id,
+        "title": result.title.replace("\n", " ").strip(),
+        "authors": [a.name for a in result.authors[:10]],
+        "abstract": result.summary.replace("\n", " ").strip(),
+        "published": result.published.isoformat(),
+        "categories": list(result.categories),
+        "pdf_url": result.pdf_url,
+        "arxiv_url": result.entry_id,
+        "comment": (result.comment or "").replace("\n", " ").strip(),
+        "source": "arxiv",
+        "hf_upvotes": 0,
+        "github_repo": github_repo,
+        "github_stars": None,
+        "hf_models": [],
+        "hf_datasets": [],
+        "hf_spaces": [],
+    }
+
+
+# ---------------------------------------------------------------------------
+# Enrichment
+# ---------------------------------------------------------------------------
+
+
+def enrich_paper(paper: dict) -> dict:
+    """Query HF API for linked models, datasets, and spaces."""
+    arxiv_id = paper["arxiv_id"]
+    if not arxiv_id:
+        return paper
+
+    base_id = re.sub(r"v\d+$", "", arxiv_id)
+
+    for resource, key, limit in [
+        ("models", "hf_models", 5),
+        ("datasets", "hf_datasets", 3),
+        ("spaces", "hf_spaces", 3),
+    ]:
+        url = f"{HF_API}/{resource}?filter=arxiv:{base_id}&limit={limit}&sort=likes"
+        try:
+            resp = requests.get(url, timeout=15)
+            if resp.ok:
+                items = resp.json()
+                paper[key] = [
+                    {"id": item.get("id", item.get("_id", "")), "likes": item.get("likes", 0)}
+                    for item in items
+                ]
+        except (requests.RequestException, ValueError):
+            pass
+
+    time.sleep(0.2)
+    return paper
+
+
+# ---------------------------------------------------------------------------
+# Merge
+# ---------------------------------------------------------------------------
+
+
+def merge_papers(hf_papers: list[dict], arxiv_papers: list[dict]) -> list[dict]:
+    """Deduplicate by arXiv ID. When both sources have a paper, merge."""
+    by_id: dict[str, dict] = {}
+
+    for p in arxiv_papers:
+        aid = re.sub(r"v\d+$", "", p["arxiv_id"])
+        if aid:
+            by_id[aid] = p
+
+    for p in hf_papers:
+        aid = re.sub(r"v\d+$", "", p["arxiv_id"])
+        if not aid:
+            continue
+        if aid in by_id:
+            existing = by_id[aid]
+            existing["source"] = "both"
+            existing["hf_upvotes"] = max(existing.get("hf_upvotes", 0), p.get("hf_upvotes", 0))
+            if p.get("github_repo") and not existing.get("github_repo"):
+                existing["github_repo"] = p["github_repo"]
+            if not existing.get("categories") and p.get("categories"):
+                existing["categories"] = p["categories"]
+        else:
+            by_id[aid] = p
+
+    return list(by_id.values())
+
+
+# ---------------------------------------------------------------------------
+# Pipeline entry point
+# ---------------------------------------------------------------------------
+
+
+def run_aiml_pipeline(
+    start: datetime | None = None,
+    end: datetime | None = None,
+    max_papers: int = 300,
+    skip_enrich: bool = False,
+) -> int:
+    """Run the full AI/ML pipeline. Returns the run ID."""
+    if end is None:
+        end = datetime.now(timezone.utc)
+    if start is None:
+        start = end - timedelta(days=7)
+
+    # Ensure timezone-aware
+    if start.tzinfo is None:
+        start = start.replace(tzinfo=timezone.utc)
+    if end.tzinfo is None:
+        end = end.replace(tzinfo=timezone.utc, hour=23, minute=59, second=59)
+
+    run_id = create_run("aiml", start.date().isoformat(), end.date().isoformat())
+    log.info("Run %d: %s to %s", run_id, start.date(), end.date())
+
+    try:
+        # Step 1: Fetch HF papers
+        log.info("Fetching HuggingFace Daily Papers ...")
+        hf_papers_raw = []
+        current = start
+        while current <= end:
+            date_str = current.strftime("%Y-%m-%d")
+            daily = fetch_hf_daily(date_str)
+            hf_papers_raw.extend(daily)
+            current += timedelta(days=1)
+
+        trending = fetch_hf_trending(limit=50)
+        hf_papers_raw.extend(trending)
+
+        hf_papers = [p for p in (normalize_hf_paper(e) for e in hf_papers_raw) if p is not None]
+        log.info("HF papers: %d", len(hf_papers))
+
+        # Step 2: Fetch arXiv papers
+        log.info("Fetching arXiv papers ...")
+        arxiv_papers = []
+        for cat in ARXIV_LARGE_CATS:
+            papers = fetch_arxiv_category(cat, start, end, max_papers, filter_keywords=True)
+            arxiv_papers.extend(papers)
+            log.info("  %s: %d papers (keyword-filtered)", cat, len(papers))
+
+        for cat in ARXIV_SMALL_CATS:
+            papers = fetch_arxiv_category(cat, start, end, max_papers, filter_keywords=False)
+            arxiv_papers.extend(papers)
+            log.info("  %s: %d papers", cat, len(papers))
+
+        # Step 3: Merge
+        all_papers = merge_papers(hf_papers, arxiv_papers)
+        log.info("Merged: %d unique papers", len(all_papers))
+
+        # Step 4: Enrich
+        if not skip_enrich:
+            log.info("Enriching with HF ecosystem links ...")
+            for i, paper in enumerate(all_papers):
+                all_papers[i] = enrich_paper(paper)
+                if (i + 1) % 25 == 0:
+                    log.info("  Enriched %d/%d ...", i + 1, len(all_papers))
+            log.info("Enrichment complete")
+
+        # Step 5: Insert into DB
+        insert_papers(all_papers, run_id, "aiml")
+        finish_run(run_id, len(all_papers))
+        log.info("Done — %d papers inserted", len(all_papers))
+        return run_id
+
+    except Exception as e:
+        finish_run(run_id, 0, status="failed")
+        log.exception("Pipeline failed")
+        raise

src/pipelines/events.py

@@ -0,0 +1,196 @@
+"""Events pipeline — conferences, releases, and news.
+
+Three sub-collectors:
+1. Conferences: curated list + aideadlin.es scrape
+2. Releases: HF trending models/spaces
+3. News: RSS feeds from key AI/security blogs
+"""
+
+import logging
+import time
+from datetime import datetime, timezone
+
+import feedparser
+import requests
+
+from src.config import CONFERENCES, HF_API, RSS_FEEDS
+from src.db import insert_events
+
+log = logging.getLogger(__name__)
+
+
+def run_events_pipeline() -> int:
+    """Run all event sub-collectors. Returns total events collected."""
+    log.info("Starting events pipeline ...")
+    all_events = []
+
+    # 1. Conference deadlines
+    conf_events = fetch_conference_deadlines()
+    all_events.extend(conf_events)
+    log.info("Conferences: %d", len(conf_events))
+
+    # 2. HF trending releases
+    release_events = fetch_hf_releases()
+    all_events.extend(release_events)
+    log.info("Releases: %d", len(release_events))
+
+    # 3. RSS news
+    news_events = fetch_rss_news()
+    all_events.extend(news_events)
+    log.info("News: %d", len(news_events))
+
+    if all_events:
+        insert_events(all_events)
+
+    log.info("Done — %d total events", len(all_events))
+    return len(all_events)
+
+
+# ---------------------------------------------------------------------------
+# Conferences
+# ---------------------------------------------------------------------------
+
+
+def fetch_conference_deadlines() -> list[dict]:
+    """Return curated conference list as events + try aideadlin.es."""
+    events = []
+
+    # Static curated list
+    for conf in CONFERENCES:
+        deadline = conf.get("deadline", "")
+        conf_date = conf.get("date", "")
+        desc = conf.get("description", "")
+        if deadline and conf_date:
+            desc = f"{desc} Deadline: {deadline}. Conference: {conf_date}."
+        elif deadline:
+            desc = f"{desc} Deadline: {deadline}."
+        elif conf_date:
+            desc = f"{desc} Conference: {conf_date}."
+        events.append({
+            "category": "conference",
+            "title": conf["name"],
+            "description": desc,
+            "url": conf["url"],
+            "event_date": deadline or conf_date or "",
+            "source": "curated",
+        })
+
+    # Try aideadlin.es for dynamic deadlines
+    try:
+        resp = requests.get("https://aideadlin.es/ai-deadlines.json", timeout=15)
+        if resp.ok:
+            deadlines = resp.json()
+            for d in deadlines:
+                if d.get("deadline", "TBA") == "TBA":
+                    continue
+                events.append({
+                    "category": "conference",
+                    "title": d.get("title", d.get("name", "")),
+                    "description": d.get("full_name", ""),
+                    "url": d.get("link", ""),
+                    "event_date": d.get("deadline", ""),
+                    "source": "aideadlin.es",
+                })
+    except (requests.RequestException, ValueError) as e:
+        log.warning("aideadlin.es fetch failed: %s", e)
+
+    return events
+
+
+# ---------------------------------------------------------------------------
+# HF/GitHub releases
+# ---------------------------------------------------------------------------
+
+
+def fetch_hf_releases() -> list[dict]:
+    """Fetch trending models and spaces from HuggingFace."""
+    events = []
+
+    # Trending models
+    try:
+        resp = requests.get(
+            f"{HF_API}/models",
+            params={"sort": "trending", "limit": 15},
+            timeout=15,
+        )
+        if resp.ok:
+            for model in resp.json():
+                events.append({
+                    "category": "release",
+                    "title": model.get("id", ""),
+                    "description": f"Trending model — {model.get('likes', 0)} likes, "
+                                   f"{model.get('downloads', 0)} downloads",
+                    "url": f"https://huggingface.co/{model.get('id', '')}",
+                    "event_date": model.get("lastModified", ""),
+                    "source": "huggingface",
+                    "relevance_score": None,
+                })
+    except (requests.RequestException, ValueError):
+        pass
+
+    time.sleep(0.5)
+
+    # Trending spaces
+    try:
+        resp = requests.get(
+            f"{HF_API}/spaces",
+            params={"sort": "trending", "limit": 10},
+            timeout=15,
+        )
+        if resp.ok:
+            for space in resp.json():
+                events.append({
+                    "category": "release",
+                    "title": f"Space: {space.get('id', '')}",
+                    "description": f"Trending space — {space.get('likes', 0)} likes",
+                    "url": f"https://huggingface.co/spaces/{space.get('id', '')}",
+                    "event_date": space.get("lastModified", ""),
+                    "source": "huggingface",
+                    "relevance_score": None,
+                })
+    except (requests.RequestException, ValueError):
+        pass
+
+    return events
+
+
+# ---------------------------------------------------------------------------
+# RSS news
+# ---------------------------------------------------------------------------
+
+
+def fetch_rss_news() -> list[dict]:
+    """Fetch recent entries from configured RSS feeds."""
+    events = []
+
+    for feed_config in RSS_FEEDS:
+        try:
+            feed = feedparser.parse(feed_config["url"])
+            for entry in feed.entries[:5]:
+                published = ""
+                if hasattr(entry, "published"):
+                    published = entry.published
+                elif hasattr(entry, "updated"):
+                    published = entry.updated
+
+                events.append({
+                    "category": "news",
+                    "title": entry.get("title", ""),
+                    "description": _clean_html(entry.get("summary", ""))[:300],
+                    "url": entry.get("link", ""),
+                    "event_date": published,
+                    "source": feed_config["name"],
+                    "relevance_score": None,
+                })
+        except Exception as e:
+            log.warning("RSS fetch failed for %s: %s", feed_config['name'], e)
+        time.sleep(0.3)
+
+    return events
+
+
+def _clean_html(text: str) -> str:
+    """Strip HTML tags from text."""
+    import re
+    clean = re.sub(r"<[^>]+>", "", text)
+    return clean.replace("\n", " ").strip()

src/pipelines/github.py

@@ -0,0 +1,194 @@
+"""GitHub projects pipeline — discover trending repos via OSSInsight.io API.
+
+Two strategies:
+1. Trending repos — weekly trending filtered by AI/ML and security keywords
+2. Collection rankings — curated collections ranked by star growth
+"""
+
+import logging
+import time
+from datetime import datetime, timedelta, timezone
+
+import requests
+
+from src.config import (
+    GITHUB_AIML_KEYWORDS,
+    GITHUB_SECURITY_KEYWORDS,
+    OSSINSIGHT_API,
+    OSSINSIGHT_COLLECTIONS,
+    OSSINSIGHT_TRENDING_LANGUAGES,
+)
+from src.db import create_run, finish_run, insert_github_projects
+
+log = logging.getLogger(__name__)
+
+_SESSION = requests.Session()
+_SESSION.headers["Accept"] = "application/json"
+
+
+def _safe_int(val, default=0) -> int:
+    """Parse an int from a value that may be empty string or None."""
+    if not val and val != 0:
+        return default
+    try:
+        return int(val)
+    except (ValueError, TypeError):
+        return default
+
+
+def _safe_float(val, default=0.0) -> float:
+    if not val and val != 0:
+        return default
+    try:
+        return float(val)
+    except (ValueError, TypeError):
+        return default
+
+
+def _api_get(path: str, params: dict | None = None) -> list[dict]:
+    """Make an OSSInsight API request and return the rows."""
+    url = f"{OSSINSIGHT_API}{path}"
+    try:
+        resp = _SESSION.get(url, params=params, timeout=30)
+        resp.raise_for_status()
+        data = resp.json().get("data", {})
+        return data.get("rows", [])
+    except (requests.RequestException, ValueError, KeyError) as e:
+        log.warning("OSSInsight API error for %s: %s", path, e)
+        return []
+
+
+def _classify_domain(repo_name: str, description: str, collection_names: str = "") -> str | None:
+    """Classify a repo into aiml, security, or None based on keywords."""
+    text = f"{repo_name} {description} {collection_names}"
+    if GITHUB_SECURITY_KEYWORDS.search(text):
+        return "security"
+    if GITHUB_AIML_KEYWORDS.search(text):
+        return "aiml"
+    return None
+
+
+def fetch_trending_repos() -> list[dict]:
+    """Fetch trending repos across configured languages for the past week."""
+    seen: set[str] = set()
+    projects: list[dict] = []
+
+    # Also fetch "All" to catch cross-language breakouts
+    languages = ["All"] + OSSINSIGHT_TRENDING_LANGUAGES
+
+    for lang in languages:
+        lang_param = lang if lang != "C++" else "C%2B%2B"
+        rows = _api_get("/trends/repos", {"language": lang_param, "period": "past_week"})
+        log.info("Trending %s: %d repos", lang, len(rows))
+
+        for row in rows:
+            repo_name = row.get("repo_name", "")
+            if not repo_name or repo_name in seen:
+                continue
+            seen.add(repo_name)
+
+            description = row.get("description", "") or ""
+            collection_names = row.get("collection_names", "") or ""
+            domain = _classify_domain(repo_name, description, collection_names)
+
+            if domain is None:
+                continue
+
+            projects.append({
+                "repo_id": _safe_int(row.get("repo_id")),
+                "repo_name": repo_name,
+                "description": description,
+                "language": row.get("primary_language", "") or "",
+                "stars": _safe_int(row.get("stars")),
+                "forks": _safe_int(row.get("forks")),
+                "pull_requests": _safe_int(row.get("pull_requests")),
+                "total_score": _safe_float(row.get("total_score")),
+                "collection_names": collection_names,
+                "topics": [],
+                "url": f"https://github.com/{repo_name}",
+                "domain": domain,
+            })
+
+        time.sleep(0.5)
+
+    return projects
+
+
+def fetch_collection_rankings() -> list[dict]:
+    """Fetch top repos from curated AI/ML and security collections."""
+    seen: set[str] = set()
+    projects: list[dict] = []
+
+    for cid, (cname, domain) in OSSINSIGHT_COLLECTIONS.items():
+        rows = _api_get(f"/collections/{cid}/ranking_by_stars", {"period": "past_28_days"})
+        log.info("Collection '%s' (%d): %d repos", cname, cid, len(rows))
+
+        for row in rows:
+            repo_name = row.get("repo_name", "")
+            if not repo_name or repo_name in seen:
+                continue
+            seen.add(repo_name)
+
+            growth = _safe_int(row.get("current_period_growth"))
+            if growth <= 0:
+                continue
+
+            projects.append({
+                "repo_id": _safe_int(row.get("repo_id")),
+                "repo_name": repo_name,
+                "description": "",
+                "language": "",
+                "stars": growth,
+                "forks": 0,
+                "pull_requests": 0,
+                "total_score": _safe_float(growth),
+                "collection_names": cname,
+                "topics": [],
+                "url": f"https://github.com/{repo_name}",
+                "domain": domain,
+            })
+
+        time.sleep(0.5)
+
+    return projects
+
+
+def run_github_pipeline() -> int:
+    """Run the full GitHub projects pipeline. Returns run_id."""
+    now = datetime.now(timezone.utc)
+    start = (now - timedelta(days=7)).date().isoformat()
+    end = now.date().isoformat()
+
+    run_id = create_run("github", start, end)
+    log.info("GitHub pipeline started — run %d (%s to %s)", run_id, start, end)
+
+    try:
+        # Strategy 1: Trending repos
+        trending = fetch_trending_repos()
+        log.info("Trending repos (filtered): %d", len(trending))
+
+        # Strategy 2: Collection rankings
+        collections = fetch_collection_rankings()
+        log.info("Collection repos: %d", len(collections))
+
+        # Merge — trending takes priority (has richer data)
+        seen = {p["repo_name"] for p in trending}
+        merged = list(trending)
+        for p in collections:
+            if p["repo_name"] not in seen:
+                seen.add(p["repo_name"])
+                merged.append(p)
+
+        log.info("Total unique projects: %d", len(merged))
+
+        if merged:
+            insert_github_projects(merged, run_id)
+
+        finish_run(run_id, len(merged))
+        log.info("GitHub pipeline complete — %d projects stored", len(merged))
+        return run_id
+
+    except Exception:
+        finish_run(run_id, 0, status="failed")
+        log.exception("GitHub pipeline failed")
+        raise

src/pipelines/security.py

@@ -0,0 +1,252 @@
+"""Security paper pipeline.
+
+Fetches security papers from arXiv (cs.CR + adjacent categories),
+finds code URLs, and writes to the database.
+"""
+
+import logging
+import re
+import time
+from datetime import datetime, timedelta, timezone
+
+import arxiv
+import requests
+
+from src.config import (
+    ADJACENT_CATEGORIES,
+    GITHUB_TOKEN,
+    GITHUB_URL_RE,
+    SECURITY_EXCLUDE_RE,
+    SECURITY_KEYWORDS,
+    SECURITY_LLM_RE,
+)
+from src.db import create_run, finish_run, insert_papers
+
+log = logging.getLogger(__name__)
+
+
+# ---------------------------------------------------------------------------
+# arXiv fetching
+# ---------------------------------------------------------------------------
+
+
+def fetch_arxiv_papers(start: datetime, end: datetime, max_papers: int) -> list[dict]:
+    """Fetch papers from arXiv: all cs.CR + security-filtered adjacent categories."""
+    client = arxiv.Client(page_size=500, delay_seconds=3.0, num_retries=3)
+    papers: dict[str, dict] = {}
+
+    # Primary: all cs.CR papers
+    log.info("Fetching cs.CR papers ...")
+    cr_query = arxiv.Search(
+        query="cat:cs.CR",
+        max_results=max_papers,
+        sort_by=arxiv.SortCriterion.SubmittedDate,
+        sort_order=arxiv.SortOrder.Descending,
+    )
+
+    for result in client.results(cr_query):
+        pub = result.published.replace(tzinfo=timezone.utc)
+        if pub < start:
+            break
+        if pub > end:
+            continue
+        paper = _result_to_dict(result)
+        papers[paper["entry_id"]] = paper
+
+    log.info("cs.CR: %d papers", len(papers))
+
+    # Adjacent categories with security keyword filter
+    for cat in ADJACENT_CATEGORIES:
+        adj_query = arxiv.Search(
+            query=f"cat:{cat}",
+            max_results=max_papers // len(ADJACENT_CATEGORIES),
+            sort_by=arxiv.SortCriterion.SubmittedDate,
+            sort_order=arxiv.SortOrder.Descending,
+        )
+        count = 0
+        for result in client.results(adj_query):
+            pub = result.published.replace(tzinfo=timezone.utc)
+            if pub < start:
+                break
+            if pub > end:
+                continue
+            text = f"{result.title} {result.summary}"
+            if SECURITY_KEYWORDS.search(text):
+                paper = _result_to_dict(result)
+                if paper["entry_id"] not in papers:
+                    papers[paper["entry_id"]] = paper
+                    count += 1
+        log.info("  %s: %d security-relevant papers", cat, count)
+
+    # Pre-filter: remove excluded topics (blockchain, surveys, etc.)
+    before = len(papers)
+    papers = {
+        eid: p for eid, p in papers.items()
+        if not SECURITY_EXCLUDE_RE.search(f"{p['title']} {p['abstract']}")
+    }
+    excluded = before - len(papers)
+    if excluded:
+        log.info("Excluded %d papers (blockchain/survey/off-topic)", excluded)
+
+    # Tag LLM-adjacent papers so the scoring prompt can apply hard caps
+    for p in papers.values():
+        text = f"{p['title']} {p['abstract']}"
+        p["llm_adjacent"] = bool(SECURITY_LLM_RE.search(text))
+
+    llm_count = sum(1 for p in papers.values() if p["llm_adjacent"])
+    if llm_count:
+        log.info("Tagged %d papers as LLM-adjacent", llm_count)
+
+    all_papers = list(papers.values())
+    log.info("Total unique papers: %d", len(all_papers))
+    return all_papers
+
+
+def _result_to_dict(result: arxiv.Result) -> dict:
+    """Convert an arxiv.Result to a plain dict."""
+    arxiv_id = result.entry_id.split("/abs/")[-1]
+    base_id = re.sub(r"v\d+$", "", arxiv_id)
+
+    return {
+        "arxiv_id": base_id,
+        "entry_id": result.entry_id,
+        "title": result.title.replace("\n", " ").strip(),
+        "authors": [a.name for a in result.authors[:10]],
+        "abstract": result.summary.replace("\n", " ").strip(),
+        "published": result.published.isoformat(),
+        "categories": list(result.categories),
+        "pdf_url": result.pdf_url,
+        "arxiv_url": result.entry_id,
+        "comment": (result.comment or "").replace("\n", " ").strip(),
+        "source": "arxiv",
+        "github_repo": "",
+        "github_stars": None,
+        "hf_upvotes": 0,
+        "hf_models": [],
+        "hf_datasets": [],
+        "hf_spaces": [],
+    }
+
+
+# ---------------------------------------------------------------------------
+# Code URL finding
+# ---------------------------------------------------------------------------
+
+
+def extract_github_urls(paper: dict) -> list[str]:
+    """Extract GitHub URLs from abstract and comments."""
+    text = f"{paper['abstract']} {paper.get('comment', '')}"
+    return list(set(GITHUB_URL_RE.findall(text)))
+
+
+def search_github_for_paper(title: str, token: str | None) -> str | None:
+    """Search GitHub for a repo matching the paper title."""
+    headers = {"Accept": "application/vnd.github.v3+json"}
+    if token:
+        headers["Authorization"] = f"token {token}"
+
+    if token:
+        try:
+            resp = requests.get("https://api.github.com/rate_limit", headers=headers, timeout=10)
+            if resp.ok:
+                remaining = resp.json().get("resources", {}).get("search", {}).get("remaining", 0)
+                if remaining < 5:
+                    return None
+        except requests.RequestException:
+            pass
+
+    clean = re.sub(r"[^\w\s]", " ", title)
+    words = clean.split()[:8]
+    query = " ".join(words)
+
+    try:
+        resp = requests.get(
+            "https://api.github.com/search/repositories",
+            params={"q": query, "sort": "updated", "per_page": 3},
+            headers=headers,
+            timeout=10,
+        )
+        if not resp.ok:
+            return None
+        items = resp.json().get("items", [])
+        if items:
+            return items[0]["html_url"]
+    except requests.RequestException:
+        pass
+    return None
+
+
+def find_code_urls(papers: list[dict]) -> dict[str, str | None]:
+    """Find code/repo URLs for each paper."""
+    token = GITHUB_TOKEN or None
+    code_urls: dict[str, str | None] = {}
+
+    for paper in papers:
+        urls = extract_github_urls(paper)
+        if urls:
+            code_urls[paper["entry_id"]] = urls[0]
+            continue
+
+        url = search_github_for_paper(paper["title"], token)
+        code_urls[paper["entry_id"]] = url
+        if not token:
+            time.sleep(2)
+
+    return code_urls
+
+
+# ---------------------------------------------------------------------------
+# Pipeline entry point
+# ---------------------------------------------------------------------------
+
+
+def run_security_pipeline(
+    start: datetime | None = None,
+    end: datetime | None = None,
+    max_papers: int = 300,
+) -> int:
+    """Run the full security pipeline. Returns the run ID."""
+    if end is None:
+        end = datetime.now(timezone.utc)
+    if start is None:
+        start = end - timedelta(days=7)
+
+    if start.tzinfo is None:
+        start = start.replace(tzinfo=timezone.utc)
+    if end.tzinfo is None:
+        end = end.replace(tzinfo=timezone.utc, hour=23, minute=59, second=59)
+
+    run_id = create_run("security", start.date().isoformat(), end.date().isoformat())
+    log.info("Run %d: %s to %s", run_id, start.date(), end.date())
+
+    try:
+        # Step 1: Fetch papers
+        papers = fetch_arxiv_papers(start, end, max_papers)
+
+        if not papers:
+            log.info("No papers found")
+            finish_run(run_id, 0)
+            return run_id
+
+        # Step 2: Find code URLs
+        log.info("Searching for code repositories ...")
+        code_urls = find_code_urls(papers)
+        with_code = sum(1 for v in code_urls.values() if v)
+        log.info("Found code for %d/%d papers", with_code, len(papers))
+
+        # Attach code URLs to papers as github_repo
+        for paper in papers:
+            url = code_urls.get(paper["entry_id"])
+            if url:
+                paper["github_repo"] = url
+
+        # Step 3: Insert into DB
+        insert_papers(papers, run_id, "security")
+        finish_run(run_id, len(papers))
+        log.info("Done — %d papers inserted", len(papers))
+        return run_id
+
+    except Exception as e:
+        finish_run(run_id, 0, status="failed")
+        log.exception("Pipeline failed")
+        raise

src/pipelines/semantic_scholar.py

@@ -0,0 +1,294 @@
+"""Semantic Scholar enrichment — connected papers, TL;DR, and topic extraction.
+
+Uses the free S2 Academic Graph API. No API key required but rate-limited
+to a shared pool. With a key (x-api-key header), 1 req/sec guaranteed.
+
+Enrichment strategy:
+1. Batch lookup all papers → TL;DR + S2 paper ID  (1 API call per 500 papers)
+2. Top N papers by score → references + recommendations  (2 calls each)
+3. Topic extraction from title/abstract  (local, no API)
+"""
+
+import json
+import logging
+import re
+import time
+
+import requests
+
+log = logging.getLogger(__name__)
+
+from src.db import (
+    clear_connections,
+    get_arxiv_id_map,
+    get_conn,
+    get_top_papers,
+    insert_connections,
+    update_paper_s2,
+    update_paper_topics,
+)
+
+S2_GRAPH = "https://api.semanticscholar.org/graph/v1"
+S2_RECO = "https://api.semanticscholar.org/recommendations/v1"
+S2_HEADERS: dict[str, str] = {}  # Add {"x-api-key": "..."} if you have one
+
+# How many top papers get full connection enrichment
+TOP_N_CONNECTIONS = 30
+# Rate limit pause between requests (seconds)
+RATE_LIMIT = 1.1
+
+
+# ---------------------------------------------------------------------------
+# Main entry point
+# ---------------------------------------------------------------------------
+
+
+def enrich_run(run_id: int, domain: str):
+    """Enrich all scored papers in a run with S2 data + topics."""
+    with get_conn() as conn:
+        rows = conn.execute(
+            "SELECT id, arxiv_id, title, abstract, composite FROM papers "
+            "WHERE run_id=? AND composite IS NOT NULL "
+            "ORDER BY composite DESC",
+            (run_id,),
+        ).fetchall()
+        papers = [dict(r) for r in rows]
+
+    if not papers:
+        log.info("No scored papers in run %d, skipping", run_id)
+        return
+
+    arxiv_map = get_arxiv_id_map(run_id)
+    log.info("Enriching %d papers from run %d (%s)...", len(papers), run_id, domain)
+
+    # Step 1: Batch TL;DR + S2 ID
+    _batch_tldr(papers)
+
+    # Step 2: Connected papers for top N
+    top_papers = papers[:TOP_N_CONNECTIONS]
+    for i, p in enumerate(top_papers):
+        try:
+            _fetch_connections(p, arxiv_map)
+        except Exception as e:
+            log.warning("Error fetching connections for %s: %s", p['arxiv_id'], e)
+        if (i + 1) % 10 == 0:
+            log.info("Connections: %d/%d", i + 1, len(top_papers))
+
+    # Step 3: Topic extraction (local, instant)
+    for p in papers:
+        topics = extract_topics(p["title"], p.get("abstract", ""), domain)
+        if topics:
+            update_paper_topics(p["id"], topics)
+
+    log.info("Done enriching run %d", run_id)
+
+
+# ---------------------------------------------------------------------------
+# Step 1: Batch TL;DR
+# ---------------------------------------------------------------------------
+
+
+def _batch_tldr(papers: list[dict]):
+    """Batch fetch TL;DR and S2 paper IDs."""
+    chunk_size = 500
+    for start in range(0, len(papers), chunk_size):
+        chunk = papers[start : start + chunk_size]
+        ids = [f"arXiv:{p['arxiv_id']}" for p in chunk]
+
+        try:
+            resp = requests.post(
+                f"{S2_GRAPH}/paper/batch",
+                params={"fields": "externalIds,tldr"},
+                json={"ids": ids},
+                headers=S2_HEADERS,
+                timeout=30,
+            )
+            resp.raise_for_status()
+            results = resp.json()
+        except Exception as e:
+            log.warning("Batch TL;DR failed: %s", e)
+            time.sleep(RATE_LIMIT)
+            continue
+
+        for paper, s2_data in zip(chunk, results):
+            if s2_data is None:
+                continue
+            s2_id = s2_data.get("paperId", "")
+            tldr_obj = s2_data.get("tldr")
+            tldr_text = tldr_obj.get("text", "") if tldr_obj else ""
+            update_paper_s2(paper["id"], s2_id, tldr_text)
+            paper["s2_paper_id"] = s2_id
+
+        found = sum(1 for r in results if r is not None)
+        log.info("Batch TL;DR: %d/%d papers found in S2", found, len(chunk))
+        time.sleep(RATE_LIMIT)
+
+
+# ---------------------------------------------------------------------------
+# Step 2: Connected papers (references + recommendations)
+# ---------------------------------------------------------------------------
+
+
+def _fetch_connections(paper: dict, arxiv_map: dict[str, int]):
+    """Fetch references and recommendations for a single paper."""
+    arxiv_id = paper["arxiv_id"]
+    paper_id = paper["id"]
+
+    # Clear old connections before re-fetching
+    clear_connections(paper_id)
+
+    connections: list[dict] = []
+
+    # References
+    time.sleep(RATE_LIMIT)
+    try:
+        resp = requests.get(
+            f"{S2_GRAPH}/paper/arXiv:{arxiv_id}/references",
+            params={"fields": "title,year,externalIds", "limit": 30},
+            headers=S2_HEADERS,
+            timeout=15,
+        )
+        if resp.ok:
+            for item in resp.json().get("data", []):
+                cited = item.get("citedPaper")
+                if not cited or not cited.get("title"):
+                    continue
+                ext = cited.get("externalIds") or {}
+                c_arxiv = ext.get("ArXiv", "")
+                connections.append({
+                    "paper_id": paper_id,
+                    "connected_arxiv_id": c_arxiv,
+                    "connected_s2_id": cited.get("paperId", ""),
+                    "connected_title": cited.get("title", ""),
+                    "connected_year": cited.get("year"),
+                    "connection_type": "reference",
+                    "in_db_paper_id": arxiv_map.get(c_arxiv),
+                })
+    except requests.RequestException as e:
+        log.warning("References failed for %s: %s", arxiv_id, e)
+
+    # Recommendations
+    time.sleep(RATE_LIMIT)
+    try:
+        resp = requests.get(
+            f"{S2_RECO}/papers/forpaper/arXiv:{arxiv_id}",
+            params={"fields": "title,year,externalIds", "limit": 15},
+            headers=S2_HEADERS,
+            timeout=15,
+        )
+        if resp.ok:
+            for rec in resp.json().get("recommendedPapers", []):
+                if not rec or not rec.get("title"):
+                    continue
+                ext = rec.get("externalIds") or {}
+                c_arxiv = ext.get("ArXiv", "")
+                connections.append({
+                    "paper_id": paper_id,
+                    "connected_arxiv_id": c_arxiv,
+                    "connected_s2_id": rec.get("paperId", ""),
+                    "connected_title": rec.get("title", ""),
+                    "connected_year": rec.get("year"),
+                    "connection_type": "recommendation",
+                    "in_db_paper_id": arxiv_map.get(c_arxiv),
+                })
+    except requests.RequestException as e:
+        log.warning("Recommendations failed for %s: %s", arxiv_id, e)
+
+    if connections:
+        insert_connections(connections)
+
+
+# ---------------------------------------------------------------------------
+# Step 3: Topic extraction (local, no API)
+# ---------------------------------------------------------------------------
+
+AIML_TOPICS = {
+    "Video Generation": re.compile(
+        r"video.generat|text.to.video|video.diffusion|video.synth|video.edit", re.I),
+    "Image Generation": re.compile(
+        r"image.generat|text.to.image|(?:stable|latent).diffusion|image.synth|image.edit", re.I),
+    "Language Models": re.compile(
+        r"language.model|(?:large|foundation).model|\bllm\b|\bgpt\b|instruction.tun|fine.tun", re.I),
+    "Code": re.compile(
+        r"code.generat|code.complet|program.synth|vibe.cod|software.engineer", re.I),
+    "Multimodal": re.compile(
+        r"multimodal|vision.language|\bvlm\b|visual.question|image.text", re.I),
+    "Efficiency": re.compile(
+        r"quantiz|distillat|pruning|efficient|scaling.law|compress|accelerat", re.I),
+    "Agents": re.compile(
+        r"\bagent\b|tool.use|function.call|planning|agentic", re.I),
+    "Speech / Audio": re.compile(
+        r"text.to.speech|\btts\b|speech|audio.generat|voice|music.generat", re.I),
+    "3D / Vision": re.compile(
+        r"\b3d\b|nerf|gaussian.splat|point.cloud|depth.estim|object.detect|segmentat", re.I),
+    "Retrieval / RAG": re.compile(
+        r"retriev|\brag\b|knowledge.(?:base|graph)|in.context.learn|embedding", re.I),
+    "Robotics": re.compile(
+        r"robot|embodied|manipulat|locomotion|navigation", re.I),
+    "Reasoning": re.compile(
+        r"reasoning|chain.of.thought|mathemat|logic|theorem", re.I),
+    "Training": re.compile(
+        r"reinforcement.learn|\brlhf\b|\bdpo\b|preference|reward.model|alignment", re.I),
+    "Architecture": re.compile(
+        r"attention.mechanism|state.space|\bmamba\b|mixture.of.expert|\bmoe\b|transformer", re.I),
+    "Benchmark": re.compile(
+        r"benchmark|evaluat|leaderboard|dataset|scaling.law", re.I),
+    "World Models": re.compile(
+        r"world.model|environment.model|predictive.model|dynamics.model", re.I),
+    "Optimization": re.compile(
+        r"optimi[zs]|gradient|convergence|learning.rate|loss.function|multi.objective|adversarial.train", re.I),
+    "RL": re.compile(
+        r"reinforcement.learn|\brl\b|reward|policy.gradient|q.learning|bandit", re.I),
+}
+
+SECURITY_TOPICS = {
+    "Web Security": re.compile(
+        r"web.(?:secur|app|vuln)|xss|injection|csrf|waf|\bbrowser.secur", re.I),
+    "Network": re.compile(
+        r"network.secur|intrusion|\bids\b|firewall|traffic|\bdns\b|\bbgp\b|\bddos\b|fingerprint|scanning|packet", re.I),
+    "Malware": re.compile(
+        r"malware|ransomware|trojan|botnet|rootkit|worm|backdoor", re.I),
+    "Vulnerabilities": re.compile(
+        r"vulnerab|\bcve\b|exploit|fuzzing|fuzz|buffer.overflow|zero.day|attack.surface|security.bench", re.I),
+    "Cryptography": re.compile(
+        r"cryptograph|encryption|decrypt|protocol|\btls\b|\bssl\b|cipher", re.I),
+    "Hardware": re.compile(
+        r"side.channel|timing.attack|spectre|meltdown|hardware|firmware|microarch|fault.inject|emfi|embedded.secur", re.I),
+    "Reverse Engineering": re.compile(
+        r"reverse.engineer|binary|decompil|obfuscat|disassembl", re.I),
+    "Mobile": re.compile(
+        r"\bandroid\b|\bios.secur|mobile.secur", re.I),
+    "Cloud": re.compile(
+        r"cloud.secur|container.secur|docker|kubernetes|serverless|devsecops", re.I),
+    "Authentication": re.compile(
+        r"authentica|identity|credential|phishing|password|oauth|passkey|webauthn", re.I),
+    "Privacy": re.compile(
+        r"privacy|anonymi|differential.privacy|data.leak|tracking|membership.inference", re.I),
+    "LLM Security": re.compile(
+        r"(?:llm|language.model).*(secur|attack|jailbreak|safety|risk|unsafe|inject|adversar)|prompt.inject|red.team|rubric.attack|preference.drift", re.I),
+    "Forensics": re.compile(
+        r"forensic|incident.response|audit|log.analy|carver|tamper|evidence", re.I),
+    "Blockchain": re.compile(
+        r"blockchain|smart.contract|solana|ethereum|memecoin|mev|defi|token|cryptocurrency", re.I),
+    "Supply Chain": re.compile(
+        r"supply.chain|dependency|package.secur|software.comp|sbom", re.I),
+}
+
+
+def extract_topics(title: str, abstract: str, domain: str) -> list[str]:
+    """Extract up to 3 topic tags from title and abstract."""
+    patterns = AIML_TOPICS if domain == "aiml" else SECURITY_TOPICS
+    abstract_head = (abstract or "")[:500]
+
+    scored: dict[str, int] = {}
+    for topic, pattern in patterns.items():
+        score = 0
+        if pattern.search(title):
+            score += 3  # Title match is strong signal
+        if pattern.search(abstract_head):
+            score += 1
+        if score > 0:
+            scored[topic] = score
+
+    ranked = sorted(scored.items(), key=lambda x: -x[1])
+    return [t for t, _ in ranked[:3]]

src/preferences.py

@@ -0,0 +1,343 @@
+"""Preference engine — learns from user signals to personalize paper rankings.
+
+Adds a preference_boost (max +1.5 / min -1.0) on top of stored composite scores.
+Never re-scores papers. Papers with composite >= 8 are never penalized.
+"""
+
+import logging
+import math
+import re
+from collections import defaultdict
+from datetime import datetime, timezone
+
+from src.db import (
+    get_all_signals_with_papers,
+    load_preferences,
+    save_preferences,
+    get_paper_signal,
+    get_paper_signals_batch,
+)
+
+log = logging.getLogger(__name__)
+
+# ---------------------------------------------------------------------------
+# Signal weights
+# ---------------------------------------------------------------------------
+
+SIGNAL_WEIGHTS = {
+    "save": 3.0,
+    "upvote": 2.0,
+    "view": 0.5,
+    "downvote": -2.0,
+    "dismiss": -1.5,
+}
+
+HALF_LIFE_DAYS = 60.0
+
+# Dimension weights for combining into final boost
+DIMENSION_WEIGHTS = {
+    "topic": 0.35,
+    "axis": 0.25,
+    "keyword": 0.15,
+    "category": 0.15,
+    "author": 0.10,
+}
+
+# Scaling factors for tanh normalization (tuned per dimension)
+SCALING_FACTORS = {
+    "topic": 5.0,
+    "axis": 4.0,
+    "keyword": 8.0,
+    "category": 5.0,
+    "author": 6.0,
+}
+
+# Stopwords for keyword extraction from titles
+_STOPWORDS = frozenset(
+    "a an the and or but in on of for to with from by at is are was were "
+    "be been being have has had do does did will would shall should may might "
+    "can could this that these those it its we our their".split()
+)
+
+_WORD_RE = re.compile(r"[a-z]{3,}", re.IGNORECASE)
+
+
+def _extract_keywords(title: str) -> list[str]:
+    """Extract meaningful keywords from a paper title."""
+    words = _WORD_RE.findall(title.lower())
+    return [w for w in words if w not in _STOPWORDS]
+
+
+def _time_decay(created_at: str) -> float:
+    """Compute time decay factor: 2^(-age_days / half_life)."""
+    try:
+        signal_dt = datetime.fromisoformat(created_at.replace("Z", "+00:00"))
+    except (ValueError, AttributeError):
+        return 0.5
+    now = datetime.now(timezone.utc)
+    age_days = max(0, (now - signal_dt).total_seconds() / 86400)
+    return math.pow(2, -age_days / HALF_LIFE_DAYS)
+
+
+# ---------------------------------------------------------------------------
+# Preference computation
+# ---------------------------------------------------------------------------
+
+def compute_preferences() -> dict[str, float]:
+    """Compute user preference profile from all signals.
+
+    Returns the preference dict (also saved to DB).
+    """
+    signals = get_all_signals_with_papers()
+    if not signals:
+        save_preferences({})
+        return {}
+
+    # Accumulate raw scores per preference key
+    raw: dict[str, float] = defaultdict(float)
+    counts: dict[str, int] = defaultdict(int)
+
+    # For axis preferences: track domain means
+    axis_sums: dict[str, list[float]] = defaultdict(list)
+
+    for sig in signals:
+        base_weight = SIGNAL_WEIGHTS.get(sig["action"], 0)
+        decay = _time_decay(sig["created_at"])
+        weight = base_weight * decay
+
+        # Topics
+        topics = sig.get("topics") or []
+        if topics:
+            per_topic = weight / len(topics)
+            for t in topics:
+                key = f"topic:{t}"
+                raw[key] += per_topic
+                counts[key] += 1
+
+        # Categories
+        categories = sig.get("categories") or []
+        if categories:
+            per_cat = weight / len(categories)
+            for c in categories:
+                key = f"category:{c}"
+                raw[key] += per_cat
+                counts[key] += 1
+
+        # Keywords from title
+        keywords = _extract_keywords(sig.get("title", ""))
+        if keywords:
+            per_kw = weight / len(keywords)
+            for kw in keywords:
+                key = f"keyword:{kw}"
+                raw[key] += per_kw
+                counts[key] += 1
+
+        # Authors (first 3 only)
+        authors = sig.get("authors") or []
+        if isinstance(authors, str):
+            authors = [authors]
+        for author in authors[:3]:
+            name = author if isinstance(author, str) else str(author)
+            key = f"author:{name}"
+            raw[key] += weight * 0.5  # reduced weight for authors
+            counts[key] += 1
+
+        # Axis preferences (track which axes are high on liked papers)
+        domain = sig.get("domain", "")
+        for i in range(1, 4):
+            axis_val = sig.get(f"score_axis_{i}")
+            if axis_val is not None:
+                axis_sums[f"{domain}:axis{i}"].append(axis_val)
+
+    # Compute axis preferences relative to domain mean
+    for sig in signals:
+        base_weight = SIGNAL_WEIGHTS.get(sig["action"], 0)
+        if base_weight <= 0:
+            continue  # Only positive signals inform axis preferences
+        decay = _time_decay(sig["created_at"])
+        weight = base_weight * decay
+        domain = sig.get("domain", "")
+
+        for i in range(1, 4):
+            axis_val = sig.get(f"score_axis_{i}")
+            mean_key = f"{domain}:axis{i}"
+            if axis_val is not None and axis_sums.get(mean_key):
+                mean = sum(axis_sums[mean_key]) / len(axis_sums[mean_key])
+                deviation = axis_val - mean
+                key = f"axis_pref:{domain}:axis{i}"
+                raw[key] += deviation * weight * 0.1
+                counts[key] += 1
+
+    # Normalize via tanh
+    prefs: dict[str, tuple[float, int]] = {}
+    for key, value in raw.items():
+        prefix = key.split(":")[0]
+        scale = SCALING_FACTORS.get(prefix, 5.0)
+        normalized = math.tanh(value / scale)
+        # Clamp to [-1, 1]
+        normalized = max(-1.0, min(1.0, normalized))
+        prefs[key] = (round(normalized, 4), counts[key])
+
+    save_preferences(prefs)
+    return {k: v for k, (v, _) in prefs.items()}
+
+
+# ---------------------------------------------------------------------------
+# Paper boost computation
+# ---------------------------------------------------------------------------
+
+def compute_paper_boost(paper: dict, preferences: dict[str, float]) -> tuple[float, list[str]]:
+    """Compute preference boost for a single paper.
+
+    Returns (boost_value, list_of_reasons).
+    Boost is clamped to [-1.0, +1.5].
+    Papers with composite >= 8 are never penalized (boost >= 0).
+    """
+    if not preferences:
+        return 0.0, []
+
+    scores: dict[str, float] = {}
+    reasons: list[str] = []
+
+    # Topic match
+    topics = paper.get("topics") or []
+    if topics:
+        topic_scores = []
+        for t in topics:
+            key = f"topic:{t}"
+            if key in preferences:
+                topic_scores.append((t, preferences[key]))
+        if topic_scores:
+            scores["topic"] = sum(v for _, v in topic_scores) / len(topic_scores)
+            for name, val in sorted(topic_scores, key=lambda x: abs(x[1]), reverse=True)[:2]:
+                if abs(val) > 0.05:
+                    reasons.append(f"Topic: {name} {val:+.2f}")
+
+    # Category match
+    categories = paper.get("categories") or []
+    if categories:
+        cat_scores = []
+        for c in categories:
+            key = f"category:{c}"
+            if key in preferences:
+                cat_scores.append((c, preferences[key]))
+        if cat_scores:
+            scores["category"] = sum(v for _, v in cat_scores) / len(cat_scores)
+            for name, val in sorted(cat_scores, key=lambda x: abs(x[1]), reverse=True)[:1]:
+                if abs(val) > 0.05:
+                    reasons.append(f"Category: {name} {val:+.2f}")
+
+    # Keyword match
+    keywords = _extract_keywords(paper.get("title", ""))
+    if keywords:
+        kw_scores = []
+        for kw in keywords:
+            key = f"keyword:{kw}"
+            if key in preferences:
+                kw_scores.append((kw, preferences[key]))
+        if kw_scores:
+            scores["keyword"] = sum(v for _, v in kw_scores) / len(kw_scores)
+            for name, val in sorted(kw_scores, key=lambda x: abs(x[1]), reverse=True)[:1]:
+                if abs(val) > 0.1:
+                    reasons.append(f"Keyword: {name} {val:+.2f}")
+
+    # Axis alignment
+    domain = paper.get("domain", "")
+    axis_scores = []
+    for i in range(1, 4):
+        key = f"axis_pref:{domain}:axis{i}"
+        if key in preferences:
+            axis_val = paper.get(f"score_axis_{i}")
+            if axis_val is not None:
+                # Higher axis value * positive preference = boost
+                axis_scores.append(preferences[key] * (axis_val / 10.0))
+    if axis_scores:
+        scores["axis"] = sum(axis_scores) / len(axis_scores)
+
+    # Author match
+    authors = paper.get("authors") or []
+    if isinstance(authors, str):
+        authors = [authors]
+    author_scores = []
+    for author in authors[:5]:
+        name = author if isinstance(author, str) else str(author)
+        key = f"author:{name}"
+        if key in preferences:
+            author_scores.append((name.split()[-1] if " " in name else name, preferences[key]))
+    if author_scores:
+        scores["author"] = max(v for _, v in author_scores)  # Best author match
+        for name, val in sorted(author_scores, key=lambda x: abs(x[1]), reverse=True)[:1]:
+            if abs(val) > 0.1:
+                reasons.append(f"Author: {name} {val:+.2f}")
+
+    # Weighted combine
+    if not scores:
+        return 0.0, []
+
+    boost = 0.0
+    total_weight = 0.0
+    for dim, dim_score in scores.items():
+        w = DIMENSION_WEIGHTS.get(dim, 0.1)
+        boost += dim_score * w
+        total_weight += w
+
+    if total_weight > 0:
+        boost = boost / total_weight  # Normalize by actual weight used
+
+    # Scale to boost range: preferences are [-1, 1], we want [-1, 1.5]
+    boost = boost * 1.5
+
+    # Clamp
+    boost = max(-1.0, min(1.5, boost))
+
+    # Safety net: high-scoring papers never penalized
+    composite = paper.get("composite") or 0
+    if composite >= 8 and boost < 0:
+        boost = 0.0
+
+    return round(boost, 2), reasons
+
+
+def is_discovery(paper: dict, boost: float) -> bool:
+    """Paper is 'discovery' if composite >= 6 AND boost <= 0."""
+    composite = paper.get("composite") or 0
+    return composite >= 6 and boost <= 0
+
+
+def enrich_papers_with_preferences(
+    papers: list[dict],
+    preferences: dict[str, float] | None = None,
+    sort_adjusted: bool = False,
+) -> list[dict]:
+    """Add preference fields to each paper dict.
+
+    Adds: adjusted_score, preference_boost, boost_reasons, is_discovery, user_signal.
+    """
+    if preferences is None:
+        preferences = load_preferences()
+
+    # Batch fetch user signals
+    paper_ids = [p["id"] for p in papers if "id" in p]
+    signals_map = get_paper_signals_batch(paper_ids) if paper_ids else {}
+
+    has_prefs = bool(preferences)
+
+    for p in papers:
+        pid = p.get("id")
+        composite = p.get("composite") or 0
+
+        if has_prefs:
+            boost, reasons = compute_paper_boost(p, preferences)
+        else:
+            boost, reasons = 0.0, []
+
+        p["preference_boost"] = boost
+        p["adjusted_score"] = round(composite + boost, 2)
+        p["boost_reasons"] = reasons
+        p["is_discovery"] = is_discovery(p, boost) if has_prefs else False
+        p["user_signal"] = signals_map.get(pid)
+
+    if sort_adjusted and has_prefs:
+        papers.sort(key=lambda p: p.get("adjusted_score", 0), reverse=True)
+
+    return papers

src/scheduler.py

@@ -0,0 +1,66 @@
+"""APScheduler — weekly pipeline trigger running inside the web process."""
+
+import logging
+
+from apscheduler.schedulers.background import BackgroundScheduler
+from apscheduler.triggers.cron import CronTrigger
+
+log = logging.getLogger(__name__)
+scheduler = BackgroundScheduler()
+
+
+def weekly_run():
+    """Run all pipelines: aiml → security → events → reports."""
+    log.info("Starting weekly run ...")
+
+    try:
+        from src.pipelines.aiml import run_aiml_pipeline
+        from src.scoring import score_run
+
+        aiml_run_id = run_aiml_pipeline()
+        score_run(aiml_run_id, "aiml")
+
+        from src.web.app import _generate_report
+        _generate_report(aiml_run_id, "aiml")
+    except Exception:
+        log.exception("AI/ML pipeline failed")
+
+    try:
+        from src.pipelines.security import run_security_pipeline
+        from src.scoring import score_run
+
+        sec_run_id = run_security_pipeline()
+        score_run(sec_run_id, "security")
+
+        from src.web.app import _generate_report
+        _generate_report(sec_run_id, "security")
+    except Exception:
+        log.exception("Security pipeline failed")
+
+    try:
+        from src.pipelines.github import run_github_pipeline
+        run_github_pipeline()
+    except Exception:
+        log.exception("GitHub pipeline failed")
+
+    try:
+        from src.pipelines.events import run_events_pipeline
+        run_events_pipeline()
+    except Exception:
+        log.exception("Events pipeline failed")
+
+    log.info("Weekly run complete")
+
+
+def start_scheduler():
+    """Start the background scheduler with weekly job."""
+    # Sunday 22:00 UTC — dashboard ready Monday morning
+    scheduler.add_job(
+        weekly_run,
+        trigger=CronTrigger(day_of_week="sun", hour=22, minute=0),
+        id="weekly_run",
+        name="Weekly research pipeline",
+        replace_existing=True,
+    )
+    scheduler.start()
+    log.info("Started — weekly run scheduled for Sunday 22:00 UTC")

src/scoring.py

@@ -0,0 +1,186 @@
+"""Unified Claude API scoring for both AI/ML and security domains."""
+
+import json
+import logging
+import re
+import time
+
+import anthropic
+
+log = logging.getLogger(__name__)
+
+from src.config import (
+    ANTHROPIC_API_KEY,
+    BATCH_SIZE,
+    CLAUDE_MODEL,
+    MAX_ABSTRACT_CHARS_AIML,
+    MAX_ABSTRACT_CHARS_SECURITY,
+    SCORING_CONFIGS,
+    SECURITY_LLM_RE,
+)
+from src.db import get_unscored_papers, update_paper_scores
+
+
+def score_run(run_id: int, domain: str) -> int:
+    """Score all unscored papers in a run. Returns count of scored papers."""
+    if not ANTHROPIC_API_KEY:
+        log.warning("ANTHROPIC_API_KEY not set — skipping scoring")
+        return 0
+
+    config = SCORING_CONFIGS[domain]
+    papers = get_unscored_papers(run_id)
+
+    if not papers:
+        log.info("No unscored papers for run %d", run_id)
+        return 0
+
+    log.info("Scoring %d %s papers ...", len(papers), domain)
+
+    client = anthropic.Anthropic(timeout=120.0)
+    max_chars = MAX_ABSTRACT_CHARS_AIML if domain == "aiml" else MAX_ABSTRACT_CHARS_SECURITY
+    scored_count = 0
+
+    for i in range(0, len(papers), BATCH_SIZE):
+        batch = papers[i : i + BATCH_SIZE]
+        batch_num = i // BATCH_SIZE + 1
+        total_batches = (len(papers) + BATCH_SIZE - 1) // BATCH_SIZE
+        log.info("Batch %d/%d (%d papers) ...", batch_num, total_batches, len(batch))
+
+        # Build user content
+        user_content = _build_batch_content(batch, domain, max_chars)
+
+        # Call Claude
+        scores = _call_claude(client, config["prompt"], user_content)
+        if not scores:
+            continue
+
+        # Map scores back to papers and update DB
+        scored_count += _apply_scores(batch, scores, domain, config)
+
+    log.info("Scored %d/%d papers", scored_count, len(papers))
+    return scored_count
+
+
+def _build_batch_content(papers: list[dict], domain: str, max_chars: int) -> str:
+    """Build the user content string for a batch of papers."""
+    lines = []
+    for p in papers:
+        abstract = (p.get("abstract") or "")[:max_chars]
+        id_field = p.get("entry_id") or p.get("arxiv_url") or p.get("arxiv_id", "")
+
+        lines.append("---")
+
+        if domain == "security":
+            lines.append(f"entry_id: {id_field}")
+        else:
+            lines.append(f"arxiv_id: {p.get('arxiv_id', '')}")
+
+        authors_list = p.get("authors", [])
+        if isinstance(authors_list, str):
+            authors_str = authors_list
+        else:
+            authors_str = ", ".join(authors_list[:5])
+
+        cats = p.get("categories", [])
+        if isinstance(cats, str):
+            cats_str = cats
+        else:
+            cats_str = ", ".join(cats)
+
+        lines.append(f"title: {p.get('title', '')}")
+        lines.append(f"authors: {authors_str}")
+        lines.append(f"categories: {cats_str}")
+
+        code_url = p.get("github_repo") or p.get("code_url") or "none found"
+        lines.append(f"code_url_found: {code_url}")
+
+        if domain == "security":
+            if "llm_adjacent" not in p:
+                text = f"{p.get('title', '')} {p.get('abstract', '')}"
+                p["llm_adjacent"] = bool(SECURITY_LLM_RE.search(text))
+            lines.append(f"llm_adjacent: {str(p['llm_adjacent']).lower()}")
+
+        if domain == "aiml":
+            lines.append(f"hf_upvotes: {p.get('hf_upvotes', 0)}")
+            hf_models = p.get("hf_models", [])
+            if hf_models:
+                model_ids = [m["id"] if isinstance(m, dict) else str(m) for m in hf_models[:3]]
+                lines.append(f"hf_models: {', '.join(model_ids)}")
+            hf_spaces = p.get("hf_spaces", [])
+            if hf_spaces:
+                space_ids = [s["id"] if isinstance(s, dict) else str(s) for s in hf_spaces[:3]]
+                lines.append(f"hf_spaces: {', '.join(space_ids)}")
+            lines.append(f"source: {p.get('source', 'unknown')}")
+
+        lines.append(f"abstract: {abstract}")
+        lines.append(f"comment: {p.get('comment', 'N/A')}")
+        lines.append("")
+
+    return "\n".join(lines)
+
+
+def _call_claude(client: anthropic.Anthropic, system_prompt: str, user_content: str) -> list[dict]:
+    """Call Claude API and extract JSON response."""
+    for attempt in range(3):
+        try:
+            response = client.messages.create(
+                model=CLAUDE_MODEL,
+                max_tokens=4096,
+                system=system_prompt,
+                messages=[{"role": "user", "content": user_content}],
+            )
+            text = response.content[0].text
+            json_match = re.search(r"\[.*\]", text, re.DOTALL)
+            if json_match:
+                return json.loads(json_match.group())
+            log.warning("No JSON array in response (attempt %d)", attempt + 1)
+        except (anthropic.APIError, json.JSONDecodeError) as e:
+            log.error("Scoring API error (attempt %d): %s", attempt + 1, e)
+            if attempt < 2:
+                time.sleep(2 ** (attempt + 1))
+            else:
+                log.error("Skipping batch after 3 failures")
+    return []
+
+
+def _apply_scores(papers: list[dict], scores: list[dict], domain: str, config: dict) -> int:
+    """Apply scores from Claude response to papers in DB. Returns count applied."""
+    axes = config["axes"]
+    weights = config["weights"]
+    weight_values = list(weights.values())
+
+    # Build lookup by ID
+    if domain == "security":
+        score_map = {s.get("entry_id", ""): s for s in scores}
+    else:
+        score_map = {s.get("arxiv_id", ""): s for s in scores}
+
+    applied = 0
+    for paper in papers:
+        if domain == "security":
+            key = paper.get("entry_id") or paper.get("arxiv_url") or ""
+        else:
+            key = paper.get("arxiv_id", "")
+
+        score = score_map.get(key)
+        if not score:
+            continue
+
+        # Extract axis scores
+        axis_scores = [score.get(ax, 0) for ax in axes]
+
+        # Compute composite
+        composite = sum(s * w for s, w in zip(axis_scores, weight_values))
+
+        update_paper_scores(paper["id"], {
+            "score_axis_1": axis_scores[0] if len(axis_scores) > 0 else None,
+            "score_axis_2": axis_scores[1] if len(axis_scores) > 1 else None,
+            "score_axis_3": axis_scores[2] if len(axis_scores) > 2 else None,
+            "composite": round(composite, 2),
+            "summary": score.get("summary", ""),
+            "reasoning": score.get("reasoning", ""),
+            "code_url": score.get("code_url"),
+        })
+        applied += 1
+
+    return applied

src/web/app.py

@@ -0,0 +1,983 @@
+"""FastAPI web application — Research Intelligence Dashboard."""
+
+import json
+import logging
+import os
+import threading
+from collections import defaultdict
+from datetime import datetime, timezone
+from pathlib import Path
+
+from fastapi import FastAPI, Request
+from fastapi.responses import HTMLResponse, JSONResponse, RedirectResponse
+from fastapi.staticfiles import StaticFiles
+from fastapi.templating import Jinja2Templates
+
+log = logging.getLogger(__name__)
+
+from starlette.middleware.base import BaseHTTPMiddleware
+
+from src.config import SCORING_CONFIGS
+from src.db import (
+    clear_preferences,
+    count_events,
+    count_github_projects,
+    count_papers,
+    delete_signal,
+    get_all_runs,
+    get_available_topics,
+    get_events,
+    get_github_languages,
+    get_github_projects_page,
+    get_latest_run,
+    get_paper,
+    get_paper_connections,
+    get_paper_signal,
+    get_papers_page,
+    get_preferences_detail,
+    get_preferences_updated_at,
+    get_signal_counts,
+    get_top_github_projects,
+    get_top_papers,
+    init_db,
+    insert_signal,
+    load_preferences,
+)
+from src.preferences import compute_preferences, enrich_papers_with_preferences
+
+app = FastAPI(title="Research Intelligence")
+
+# Static files & templates
+STATIC_DIR = Path(__file__).parent / "static"
+TEMPLATE_DIR = Path(__file__).parent / "templates"
+DATA_DIR = Path("data")
+
+app.mount("/static", StaticFiles(directory=str(STATIC_DIR)), name="static")
+templates = Jinja2Templates(directory=str(TEMPLATE_DIR))
+
+
+# ---------------------------------------------------------------------------
+# First-run redirect middleware
+# ---------------------------------------------------------------------------
+
+
+class FirstRunMiddleware(BaseHTTPMiddleware):
+    """Redirect all non-setup requests to /setup when config.yaml is missing."""
+
+    _ALLOWED_PREFIXES = ("/setup", "/static", "/api/setup", "/sw.js")
+
+    async def dispatch(self, request: Request, call_next):
+        from src.config import FIRST_RUN
+        if FIRST_RUN:
+            path = request.url.path
+            if not any(path.startswith(p) for p in self._ALLOWED_PREFIXES):
+                return RedirectResponse("/setup", status_code=302)
+        return await call_next(request)
+
+
+app.add_middleware(FirstRunMiddleware)
+
+
+@app.get("/sw.js")
+async def service_worker():
+    """Serve SW from root scope for PWA."""
+    from fastapi.responses import FileResponse
+    return FileResponse(
+        STATIC_DIR / "sw.js",
+        media_type="application/javascript",
+        headers={"Service-Worker-Allowed": "/"},
+    )
+
+
+def score_bar(value, max_val=10):
+    """Render a visual score bar."""
+    if value is None or max_val == 0:
+        return "░" * 10
+    filled = round(float(value) * 10 / max_val)
+    filled = max(0, min(10, filled))
+    return "█" * filled + "░" * (10 - filled)
+
+
+def format_date(value, fmt="short"):
+    """Format dates from various input formats (ISO, RFC 2822, etc.)."""
+    if not value:
+        return ""
+    from email.utils import parsedate_to_datetime
+    dt = None
+    # Try ISO format first
+    for pattern in ("%Y-%m-%dT%H:%M:%S%z", "%Y-%m-%dT%H:%M:%S", "%Y-%m-%d"):
+        try:
+            dt = datetime.strptime(value[:26], pattern)
+            break
+        except (ValueError, TypeError):
+            continue
+    # Try RFC 2822 (RSS dates like "Wed, 18 Feb 2026 21:00:00 GMT")
+    if dt is None:
+        try:
+            dt = parsedate_to_datetime(value)
+        except (ValueError, TypeError):
+            return value[:10] if len(value) >= 10 else value
+    if fmt == "short":
+        return dt.strftime("%Y-%m-%d")
+    elif fmt == "medium":
+        return dt.strftime("%b %d, %Y")
+    elif fmt == "long":
+        return dt.strftime("%a, %b %d %Y")
+    return dt.strftime("%Y-%m-%d")
+
+
+def abbreviate_label(label):
+    """Abbreviate axis labels for table headers."""
+    abbrevs = {
+        "Code & Weights": "Code/Wt",
+        "Novelty": "Novel",
+        "Practical Applicability": "Practical",
+        "Has Code/PoC": "Code/PoC",
+        "Novel Attack Surface": "Attack",
+        "Real-World Impact": "Impact",
+    }
+    return abbrevs.get(label, label[:10])
+
+
+# Register as Jinja2 globals/filters
+templates.env.globals["score_bar"] = score_bar
+templates.env.globals["abbreviate_label"] = abbreviate_label
+templates.env.filters["format_date"] = format_date
+
+
+@app.on_event("startup")
+def startup():
+    from src.config import validate_env
+    validate_env()
+    init_db()
+    from src.scheduler import start_scheduler
+    start_scheduler()
+    log.info("Research Intelligence started")
+
+
+@app.on_event("shutdown")
+def shutdown():
+    from src.scheduler import scheduler
+    scheduler.shutdown(wait=False)
+    # Wait for running pipeline threads (up to 30s each)
+    for t in _pipeline_threads:
+        if t.is_alive():
+            log.info("Waiting for %s to finish ...", t.name)
+            t.join(timeout=30)
+    log.info("Research Intelligence stopped")
+
+
+# ---------------------------------------------------------------------------
+# Dashboard
+# ---------------------------------------------------------------------------
+
+
+@app.get("/", response_class=HTMLResponse)
+async def dashboard(request: Request):
+    now = datetime.now(timezone.utc)
+    week_label = now.strftime("%b %d, %Y")
+
+    aiml_top = get_top_papers("aiml", limit=5)
+    security_top = get_top_papers("security", limit=5)
+
+    # Enrich dashboard cards with preference data
+    preferences = load_preferences()
+    if preferences:
+        aiml_top = enrich_papers_with_preferences(aiml_top, preferences)
+        security_top = enrich_papers_with_preferences(security_top, preferences)
+
+    aiml_run = get_latest_run("aiml")
+    security_run = get_latest_run("security")
+
+    last_run = None
+    for r in [aiml_run, security_run]:
+        if r and r.get("finished_at"):
+            ts = r["finished_at"][:16]
+            if last_run is None or ts > last_run:
+                last_run = ts
+
+    events = get_events(limit=50)
+    today = now.strftime("%Y-%m-%d")
+    # Deduplicate + filter past conference deadlines
+    events_grouped = defaultdict(list)
+    seen: dict[str, set] = defaultdict(set)
+    for e in events:
+        cat = e.get("category", "other")
+        title = e.get("title", "")
+        if title in seen[cat]:
+            continue
+        # Skip past conference deadlines
+        if cat == "conference" and (e.get("event_date") or "") < today:
+            continue
+        seen[cat].add(title)
+        events_grouped[cat].append(e)
+
+    with _pipeline_lock:
+        running = list(_running_pipelines)
+
+    # Show seed banner if few signals exist
+    signal_counts = get_signal_counts()
+    total_signals = sum(v for k, v in signal_counts.items() if k != "view")
+    show_seed_banner = total_signals < 5
+
+    return templates.TemplateResponse("dashboard.html", {
+        "request": request,
+        "active": "dashboard",
+        "week_label": week_label,
+        "aiml_count": count_papers("aiml", scored_only=True),
+        "security_count": count_papers("security", scored_only=True),
+        "github_count": count_github_projects(),
+        "event_count": count_events(),
+        "last_run": last_run,
+        "aiml_top": aiml_top,
+        "security_top": security_top,
+        "events": events,
+        "events_grouped": dict(events_grouped),
+        "running_pipelines": running,
+        "show_seed_banner": show_seed_banner,
+    })
+
+
+# ---------------------------------------------------------------------------
+# Papers list
+# ---------------------------------------------------------------------------
+
+
+@app.get("/papers/{domain}", response_class=HTMLResponse)
+async def papers_list(
+    request: Request,
+    domain: str,
+    offset: int = 0,
+    limit: int = 50,
+    search: str | None = None,
+    min_score: float | None = None,
+    has_code: bool = False,
+    topic: str | None = None,
+    sort: str | None = None,
+):
+    if domain not in ("aiml", "security"):
+        return RedirectResponse("/")
+
+    config = SCORING_CONFIGS[domain]
+    run = get_latest_run(domain) or {}
+
+    # Load preferences to determine if personalized sort is available
+    preferences = load_preferences()
+    has_preferences = bool(preferences)
+
+    # Default to personalized sort when preferences exist
+    effective_sort = sort
+    if sort == "adjusted" and not has_preferences:
+        effective_sort = "score"
+
+    papers, total = get_papers_page(
+        domain, run_id=run.get("id"),
+        offset=offset, limit=limit,
+        min_score=min_score,
+        has_code=has_code if has_code else None,
+        search=search,
+        topic=topic,
+        sort=effective_sort if effective_sort != "adjusted" else "score",
+    )
+
+    # Enrich with preferences
+    sort_adjusted = (sort == "adjusted") and has_preferences
+    papers = enrich_papers_with_preferences(papers, preferences, sort_adjusted=sort_adjusted)
+
+    # Get available topics for the filter dropdown
+    available_topics = get_available_topics(domain, run.get("id", 0)) if run else []
+
+    domain_label = "AI/ML" if domain == "aiml" else "Security"
+
+    context = {
+        "request": request,
+        "active": domain,
+        "domain": domain,
+        "domain_label": domain_label,
+        "papers": papers,
+        "total": total,
+        "offset": offset,
+        "limit": limit,
+        "search": search,
+        "min_score": min_score,
+        "has_code": has_code,
+        "topic": topic,
+        "sort": sort,
+        "available_topics": available_topics,
+        "run": run,
+        "axis_labels": config["axis_labels"],
+        "has_preferences": has_preferences,
+    }
+
+    # Return partial for HTMX requests (filter / pagination)
+    if request.headers.get("HX-Request"):
+        return templates.TemplateResponse("partials/papers_results.html", context)
+
+    return templates.TemplateResponse("papers.html", context)
+
+
+# ---------------------------------------------------------------------------
+# Paper detail
+# ---------------------------------------------------------------------------
+
+
+@app.get("/papers/{domain}/{paper_id}", response_class=HTMLResponse)
+async def paper_detail(request: Request, domain: str, paper_id: int):
+    paper = get_paper(paper_id)
+    if not paper:
+        return RedirectResponse(f"/papers/{domain}")
+
+    config = SCORING_CONFIGS.get(domain, SCORING_CONFIGS["aiml"])
+    domain_label = "AI/ML" if domain == "aiml" else "Security"
+
+    connections = get_paper_connections(paper_id)
+
+    # Record view signal (deduped by 5-min window)
+    insert_signal(paper_id, "view")
+
+    # Preference boost info
+    preferences = load_preferences()
+    papers_enriched = enrich_papers_with_preferences([paper], preferences)
+    paper = papers_enriched[0]
+
+    return templates.TemplateResponse("paper_detail.html", {
+        "request": request,
+        "active": domain,
+        "domain": domain,
+        "domain_label": domain_label,
+        "paper": paper,
+        "axis_labels": config["axis_labels"],
+        "score_bar": score_bar,
+        "connections": connections,
+    })
+
+
+# ---------------------------------------------------------------------------
+# Events
+# ---------------------------------------------------------------------------
+
+
+@app.get("/events", response_class=HTMLResponse)
+async def events_page(request: Request):
+    deadlines_raw = get_events(category="conference", limit=50)
+    releases = get_events(category="release", limit=20)
+    news_raw = get_events(category="news", limit=40)
+
+    # Filter out past deadlines
+    today = datetime.now(timezone.utc).strftime("%Y-%m-%d")
+    deadlines = [d for d in deadlines_raw if (d.get("event_date") or "") >= today]
+
+    # Deduplicate news by title and sort by date (RFC 2822 dates don't sort lexicographically)
+    from email.utils import parsedate_to_datetime as _parse_rfc
+    seen_titles: set[str] = set()
+    news: list[dict] = []
+    for n in news_raw:
+        t = n.get("title", "")
+        if t not in seen_titles:
+            seen_titles.add(t)
+            news.append(n)
+
+    def _news_sort_key(item):
+        d = item.get("event_date", "")
+        try:
+            return _parse_rfc(d)
+        except (ValueError, TypeError):
+            try:
+                return datetime.fromisoformat(d[:19])
+            except (ValueError, TypeError):
+                return datetime.min
+
+    news.sort(key=_news_sort_key, reverse=True)
+    news = news[:20]
+
+    return templates.TemplateResponse("events.html", {
+        "request": request,
+        "active": "events",
+        "total": count_events(),
+        "deadlines": deadlines,
+        "releases": releases,
+        "news": news,
+    })
+
+
+# ---------------------------------------------------------------------------
+# GitHub Projects
+# ---------------------------------------------------------------------------
+
+
+@app.get("/github", response_class=HTMLResponse)
+async def github_page(
+    request: Request,
+    offset: int = 0,
+    limit: int = 50,
+    search: str | None = None,
+    language: str | None = None,
+    domain: str | None = None,
+    sort: str | None = None,
+):
+    run = get_latest_run("github") or {}
+
+    projects, total = get_github_projects_page(
+        run_id=run.get("id"),
+        offset=offset,
+        limit=limit,
+        search=search,
+        language=language,
+        domain=domain,
+        sort=sort,
+    )
+
+    available_languages = get_github_languages(run["id"]) if run else []
+
+    context = {
+        "request": request,
+        "active": "github",
+        "projects": projects,
+        "total": total,
+        "offset": offset,
+        "limit": limit,
+        "search": search,
+        "language": language,
+        "domain_filter": domain,
+        "sort": sort,
+        "available_languages": available_languages,
+        "run": run,
+    }
+
+    if request.headers.get("HX-Request"):
+        return templates.TemplateResponse("partials/github_results.html", context)
+
+    return templates.TemplateResponse("github.html", context)
+
+
+# ---------------------------------------------------------------------------
+# Archive
+# ---------------------------------------------------------------------------
+
+
+@app.get("/weeks", response_class=HTMLResponse)
+async def weeks_page(request: Request):
+    weeks_dir = DATA_DIR / "weeks"
+    archives = []
+    if weeks_dir.exists():
+        for f in sorted(weeks_dir.glob("*.md"), reverse=True):
+            parts = f.stem.rsplit("-", 1)
+            domain = parts[-1] if len(parts) > 1 and parts[-1] in ("aiml", "security") else "unknown"
+            date = parts[0] if len(parts) > 1 else f.stem
+            archives.append({"filename": f.name, "date": date, "domain": domain})
+
+    runs = get_all_runs(limit=20)
+
+    return templates.TemplateResponse("weeks.html", {
+        "request": request,
+        "active": "weeks",
+        "archives": archives,
+        "runs": runs,
+    })
+
+
+@app.get("/weeks/{filename}", response_class=HTMLResponse)
+async def weeks_file(filename: str):
+    import html as html_mod
+    filepath = (DATA_DIR / "weeks" / filename).resolve()
+    weeks_root = (DATA_DIR / "weeks").resolve()
+    if not filepath.is_relative_to(weeks_root) or not filepath.exists() or not filepath.suffix == ".md":
+        return RedirectResponse("/weeks")
+    content = html_mod.escape(filepath.read_text())
+    safe_name = html_mod.escape(filename)
+    page = f"""<!DOCTYPE html><html><head><title>{safe_name}</title>
+    <link rel="stylesheet" href="/static/style.css">
+    <style>body {{ padding: 2rem; max-width: 900px; margin: 0 auto; }}
+    pre, code {{ font-family: var(--font-mono); }} table {{ border-collapse: collapse; width: 100%; }}
+    th, td {{ border: 1px solid var(--border); padding: 0.5rem; text-align: left; }}</style>
+    </head><body><a href="/weeks">&larr; Back to archive</a>
+    <pre style="white-space:pre-wrap; line-height:1.7">{content}</pre></body></html>"""
+    return HTMLResponse(content=page)
+
+
+# ---------------------------------------------------------------------------
+# Pipeline triggers
+# ---------------------------------------------------------------------------
+
+
+_running_pipelines: set[str] = set()
+_pipeline_lock = threading.Lock()
+_pipeline_threads: list[threading.Thread] = []
+
+
+def _enrich_s2(run_id: int, domain: str):
+    """Run S2 enrichment (best-effort, failures don't break pipeline)."""
+    try:
+        from src.pipelines.semantic_scholar import enrich_run
+        enrich_run(run_id, domain)
+    except Exception as e:
+        log.warning("S2 enrichment for %s run %d failed: %s", domain, run_id, e)
+
+
+def _run_pipeline_bg(domain: str):
+    """Run a pipeline in a background thread."""
+    try:
+        if domain == "aiml":
+            from src.pipelines.aiml import run_aiml_pipeline
+            from src.scoring import score_run
+            run_id = run_aiml_pipeline()
+            score_run(run_id, "aiml")
+            _enrich_s2(run_id, "aiml")
+            _generate_report(run_id, "aiml")
+        elif domain == "security":
+            from src.pipelines.security import run_security_pipeline
+            from src.scoring import score_run
+            run_id = run_security_pipeline()
+            score_run(run_id, "security")
+            _enrich_s2(run_id, "security")
+            _generate_report(run_id, "security")
+        elif domain == "github":
+            from src.pipelines.github import run_github_pipeline
+            run_github_pipeline()
+        elif domain == "events":
+            from src.pipelines.events import run_events_pipeline
+            run_events_pipeline()
+    except Exception as e:
+        log.exception("Background pipeline %s failed", domain)
+    finally:
+        with _pipeline_lock:
+            _running_pipelines.discard(domain)
+
+
+@app.post("/run/{domain}")
+async def trigger_run(domain: str):
+    if domain not in ("aiml", "security", "github", "events"):
+        return RedirectResponse("/", status_code=303)
+    with _pipeline_lock:
+        if domain in _running_pipelines:
+            return RedirectResponse("/", status_code=303)
+        _running_pipelines.add(domain)
+        thread = threading.Thread(target=_run_pipeline_bg, args=(domain,), name=f"pipeline-{domain}")
+        thread.start()
+        _pipeline_threads.append(thread)
+    return RedirectResponse("/", status_code=303)
+
+
+# ---------------------------------------------------------------------------
+# API status
+# ---------------------------------------------------------------------------
+
+
+@app.get("/api/status")
+async def api_status():
+    aiml_run = get_latest_run("aiml")
+    security_run = get_latest_run("security")
+    github_run = get_latest_run("github")
+    with _pipeline_lock:
+        running = list(_running_pipelines)
+    return {
+        "aiml": aiml_run,
+        "security": security_run,
+        "github": github_run,
+        "github_count": count_github_projects(),
+        "events_count": count_events(),
+        "running_pipelines": running,
+    }
+
+
+# ---------------------------------------------------------------------------
+# Preference signals
+# ---------------------------------------------------------------------------
+
+
+def _maybe_recompute_preferences():
+    """Recompute preferences if stale (>1 hour since last update)."""
+    updated_at = get_preferences_updated_at()
+    if updated_at:
+        try:
+            last = datetime.fromisoformat(updated_at.replace("Z", "+00:00"))
+            age_hours = (datetime.now(timezone.utc) - last).total_seconds() / 3600
+            if age_hours < 1:
+                return
+        except (ValueError, AttributeError):
+            pass
+    # Recompute in background thread
+    thread = threading.Thread(target=compute_preferences, name="pref-recompute")
+    thread.start()
+
+
+@app.post("/api/signal/{paper_id}/{action}", response_class=HTMLResponse)
+async def record_signal(request: Request, paper_id: int, action: str):
+    """Record a user signal. Returns HTMX partial with updated button state."""
+    if action not in ("save", "upvote", "downvote", "dismiss"):
+        return HTMLResponse("Invalid action", status_code=400)
+
+    paper = get_paper(paper_id)
+    if not paper:
+        return HTMLResponse("Paper not found", status_code=404)
+
+    # Toggle: if same signal exists, remove it
+    current = get_paper_signal(paper_id)
+    if current == action:
+        delete_signal(paper_id, action)
+        _maybe_recompute_preferences()
+        return templates.TemplateResponse("partials/signal_buttons.html", {
+            "request": request,
+            "paper_id": paper_id,
+            "user_signal": None,
+        })
+
+    # Remove conflicting signals (e.g., remove upvote if downvoting)
+    for conflicting in ("upvote", "downvote", "dismiss"):
+        if conflicting != action:
+            delete_signal(paper_id, conflicting)
+
+    insert_signal(paper_id, action)
+    _maybe_recompute_preferences()
+
+    return templates.TemplateResponse("partials/signal_buttons.html", {
+        "request": request,
+        "paper_id": paper_id,
+        "user_signal": action,
+    })
+
+
+@app.get("/api/preferences")
+async def api_preferences():
+    """Return preference profile as JSON."""
+    prefs = load_preferences()
+    counts = get_signal_counts()
+    return {"preferences": prefs, "signal_counts": counts}
+
+
+@app.post("/api/preferences/recompute")
+async def api_recompute_preferences():
+    """Force recompute preferences."""
+    prefs = compute_preferences()
+    return {"status": "ok", "preference_count": len(prefs)}
+
+
+@app.post("/api/preferences/reset")
+async def api_reset_preferences():
+    """Clear all signals and preferences."""
+    clear_preferences()
+    return {"status": "ok"}
+
+
+@app.get("/preferences", response_class=HTMLResponse)
+async def preferences_page(request: Request):
+    """User preferences dashboard."""
+    prefs_detail = get_preferences_detail()
+    counts = get_signal_counts()
+    updated_at = get_preferences_updated_at()
+
+    # Group preferences by type
+    grouped: dict[str, list[dict]] = defaultdict(list)
+    for p in prefs_detail:
+        prefix = p["pref_key"].split(":")[0]
+        name = p["pref_key"].split(":", 1)[1] if ":" in p["pref_key"] else p["pref_key"]
+        grouped[prefix].append({
+            "name": name,
+            "value": p["pref_value"],
+            "count": p["signal_count"],
+        })
+
+    return templates.TemplateResponse("preferences.html", {
+        "request": request,
+        "active": "preferences",
+        "grouped": dict(grouped),
+        "signal_counts": counts,
+        "updated_at": updated_at,
+        "total_prefs": len(prefs_detail),
+    })
+
+
+# ---------------------------------------------------------------------------
+# S2 enrichment trigger
+# ---------------------------------------------------------------------------
+
+
+@app.post("/run/enrich/{domain}")
+async def trigger_enrich(domain: str):
+    """Trigger Semantic Scholar enrichment for the latest run."""
+    if domain not in ("aiml", "security"):
+        return RedirectResponse("/", status_code=303)
+
+    run = get_latest_run(domain)
+    if not run:
+        return RedirectResponse(f"/papers/{domain}", status_code=303)
+
+    with _pipeline_lock:
+        key = f"enrich-{domain}"
+        if key in _running_pipelines:
+            return RedirectResponse(f"/papers/{domain}", status_code=303)
+        _running_pipelines.add(key)
+
+        def _run():
+            try:
+                from src.pipelines.semantic_scholar import enrich_run
+                enrich_run(run["id"], domain)
+            except Exception as e:
+                log.warning("S2 enrichment for %s failed: %s", domain, e)
+            finally:
+                with _pipeline_lock:
+                    _running_pipelines.discard(key)
+
+        thread = threading.Thread(target=_run)
+        thread.start()
+    return RedirectResponse(f"/papers/{domain}", status_code=303)
+
+
+# ---------------------------------------------------------------------------
+# Setup wizard
+# ---------------------------------------------------------------------------
+
+
+@app.get("/setup", response_class=HTMLResponse)
+async def setup_page(request: Request):
+    """First-time setup wizard."""
+    return templates.TemplateResponse("setup.html", {"request": request})
+
+
+@app.post("/api/setup/validate-key")
+async def validate_api_key(request: Request):
+    """Validate an Anthropic API key with a test call."""
+    try:
+        body = await request.json()
+        key = body.get("api_key", "").strip()
+        if not key:
+            return JSONResponse({"valid": False, "error": "No key provided"})
+
+        import anthropic
+        client = anthropic.Anthropic(api_key=key, timeout=15.0)
+        client.messages.create(
+            model="claude-haiku-4-5-20251001",
+            max_tokens=10,
+            messages=[{"role": "user", "content": "Hi"}],
+        )
+        return JSONResponse({"valid": True})
+    except Exception as e:
+        return JSONResponse({"valid": False, "error": str(e)[:100]})
+
+
+@app.post("/api/setup/save")
+async def save_setup(request: Request):
+    """Save setup wizard config to config.yaml and .env."""
+    try:
+        body = await request.json()
+        api_key = body.get("api_key", "").strip()
+
+        # Write API key to .env (never in config.yaml)
+        if api_key:
+            env_path = Path(".env")
+            env_lines = []
+            if env_path.exists():
+                for line in env_path.read_text().splitlines():
+                    if not line.startswith("ANTHROPIC_API_KEY="):
+                        env_lines.append(line)
+            env_lines.append(f"ANTHROPIC_API_KEY={api_key}")
+            env_path.write_text("\n".join(env_lines) + "\n")
+            # Also set in current process
+            os.environ["ANTHROPIC_API_KEY"] = api_key
+            import src.config
+            src.config.ANTHROPIC_API_KEY = api_key
+
+        # Build config.yaml
+        domains_data = body.get("domains", {})
+        schedule_cron = body.get("schedule", "0 22 * * 0")
+
+        config_data = {
+            "domains": {
+                "aiml": {
+                    "enabled": domains_data.get("aiml", {}).get("enabled", True),
+                    "label": "AI / ML",
+                    "sources": ["huggingface", "arxiv"],
+                    "arxiv_categories": ["cs.CV", "cs.CL", "cs.LG"],
+                    "scoring_axes": _build_axes_config("aiml", domains_data),
+                    "include_patterns": [],
+                    "exclude_patterns": [],
+                    "preferences": {"boost_topics": [], "penalize_topics": []},
+                },
+                "security": {
+                    "enabled": domains_data.get("security", {}).get("enabled", True),
+                    "label": "Security",
+                    "sources": ["arxiv"],
+                    "arxiv_categories": ["cs.CR"],
+                    "scoring_axes": _build_axes_config("security", domains_data),
+                    "include_patterns": [],
+                    "exclude_patterns": [],
+                    "preferences": {"boost_topics": [], "penalize_topics": []},
+                },
+            },
+            "github": {"enabled": body.get("github", {}).get("enabled", True)},
+            "schedule": {"cron": schedule_cron} if schedule_cron else {"cron": ""},
+            "database": {"path": "data/researcher.db"},
+            "web": {"host": "0.0.0.0", "port": 8888},
+        }
+
+        from src.config import save_config
+        save_config(config_data)
+
+        return JSONResponse({"status": "ok"})
+    except Exception as e:
+        log.exception("Setup save failed")
+        return JSONResponse({"status": "error", "error": str(e)[:200]})
+
+
+def _build_axes_config(domain: str, domains_data: dict) -> list[dict]:
+    """Build scoring axes config from wizard form data."""
+    d = domains_data.get(domain, {})
+    weights = d.get("scoring_weights", [])
+
+    if domain == "aiml":
+        defaults = [
+            {"name": "Code & Weights", "weight": 0.30, "description": "Open weights on HF, code on GitHub"},
+            {"name": "Novelty", "weight": 0.35, "description": "Paradigm shifts over incremental"},
+            {"name": "Practical Applicability", "weight": 0.35, "description": "Usable by practitioners soon"},
+        ]
+    else:
+        defaults = [
+            {"name": "Has Code/PoC", "weight": 0.25, "description": "Working tools, repos, artifacts"},
+            {"name": "Novel Attack Surface", "weight": 0.40, "description": "First-of-kind research"},
+            {"name": "Real-World Impact", "weight": 0.35, "description": "Affects production systems"},
+        ]
+
+    for i, ax in enumerate(defaults):
+        if i < len(weights):
+            ax["weight"] = round(weights[i], 2)
+
+    return defaults
+
+
+# ---------------------------------------------------------------------------
+# Seed preferences
+# ---------------------------------------------------------------------------
+
+
+@app.get("/seed-preferences", response_class=HTMLResponse)
+async def seed_preferences_page(request: Request):
+    """Show seed papers for preference bootstrapping."""
+    seed_path = Path("data/seed_papers.json")
+    papers = []
+    if seed_path.exists():
+        papers = json.loads(seed_path.read_text())
+    return templates.TemplateResponse("seed_preferences.html", {
+        "request": request,
+        "active": "preferences",
+        "papers": papers,
+    })
+
+
+@app.post("/api/seed-preferences")
+async def save_seed_preferences(request: Request):
+    """Bulk-insert seed preference signals."""
+    body = await request.json()
+    ratings = body.get("ratings", {})
+
+    # Find papers in DB by arxiv_id
+    from src.db import get_conn
+    inserted = 0
+    with get_conn() as conn:
+        for arxiv_id, action in ratings.items():
+            if action not in ("upvote", "downvote"):
+                continue
+            row = conn.execute(
+                "SELECT id FROM papers WHERE arxiv_id=? LIMIT 1",
+                (arxiv_id,),
+            ).fetchone()
+            if row:
+                insert_signal(row["id"], action)
+                inserted += 1
+
+    if inserted > 0:
+        compute_preferences()
+
+    return JSONResponse({"status": "ok", "count": inserted})
+
+
+# ---------------------------------------------------------------------------
+# Report generation
+# ---------------------------------------------------------------------------
+
+
+def _generate_report(run_id: int, domain: str):
+    """Generate a markdown report and save to data/weeks/."""
+    from src.db import get_run
+    run = get_run(run_id)
+    if not run:
+        return
+
+    papers = get_top_papers(domain, run_id=run_id, limit=20)
+    if not papers:
+        return
+
+    config = SCORING_CONFIGS[domain]
+    axis_labels = config["axis_labels"]
+    date_start = run["date_start"]
+    date_end = run["date_end"]
+
+    if domain == "aiml":
+        title = f"AI/ML Research Weekly: {date_start} – {date_end}"
+    else:
+        title = f"Security Research Weekly: {date_start} – {date_end}"
+
+    lines = [f"# {title}\n\n"]
+    lines.append(f"> **{run.get('paper_count', len(papers))}** papers analyzed and scored.\n\n")
+
+    # Top 5
+    top5 = papers[:5]
+    honorable = papers[5:20]
+
+    lines.append("## Top Papers\n\n")
+    for i, p in enumerate(top5, 1):
+        authors = p.get("authors", [])
+        if isinstance(authors, str):
+            authors_str = authors
+        elif len(authors) > 3:
+            authors_str = ", ".join(authors[:3]) + " et al."
+        else:
+            authors_str = ", ".join(authors)
+
+        lines.append(f"### {i}. {p['title']}\n\n")
+        lines.append(f"**Authors:** {authors_str}\n")
+        arxiv_id = p.get("arxiv_id", "")
+        lines.append(f"**arXiv:** [{arxiv_id}](https://arxiv.org/abs/{arxiv_id})\n")
+        if p.get("code_url"):
+            lines.append(f"**Code:** [{p['code_url']}]({p['code_url']})\n")
+        lines.append("\n")
+
+        if p.get("summary"):
+            lines.append(f"> {p['summary']}\n\n")
+
+        lines.append("| Metric | Score | |\n|--------|-------|-|\n")
+        for j, label in enumerate(axis_labels):
+            val = p.get(f"score_axis_{j+1}", 0) or 0
+            bar = score_bar(val)
+            lines.append(f"| {label} | {val}/10 | `{bar}` |\n")
+        comp = p.get("composite", 0) or 0
+        lines.append(f"| **Composite** | **{comp}/10** | `{score_bar(comp)}` |\n\n")
+
+        if p.get("reasoning"):
+            lines.append(f"*{p['reasoning']}*\n\n")
+        lines.append("---\n\n")
+
+    # Honorable mentions
+    if honorable:
+        lines.append("## Honorable Mentions\n\n")
+        lines.append("| # | Paper | Score | Summary |\n")
+        lines.append("|---|-------|-------|---------|\n")
+        for i, p in enumerate(honorable, 6):
+            t = p["title"][:80].replace("|", "\\|")
+            if len(p["title"]) > 80:
+                t += "..."
+            s = (p.get("summary") or "")[:120].replace("|", "\\|")
+            if len(p.get("summary") or "") > 120:
+                s += "..."
+            aid = p.get("arxiv_id", "")
+            lines.append(f"| {i} | [{t}](https://arxiv.org/abs/{aid}) | {p.get('composite', 0)} | {s} |\n")
+        lines.append("\n")
+
+    lines.append("---\n*Generated by Research Intelligence*\n")
+
+    report = "".join(lines)
+
+    weeks_dir = DATA_DIR / "weeks"
+    weeks_dir.mkdir(parents=True, exist_ok=True)
+    filename = f"{date_start}-{domain}.md"
+    (weeks_dir / filename).write_text(report)
+    log.info("Report written to %s", weeks_dir / filename)

src/web/static/htmx.min.js

@@ -0,0 +1 @@
+var htmx=function(){"use strict";const Q={onLoad:null,process:null,on:null,off:null,trigger:null,ajax:null,find:null,findAll:null,closest:null,values:function(e,t){const n=cn(e,t||"post");return n.values},remove:null,addClass:null,removeClass:null,toggleClass:null,takeClass:null,swap:null,defineExtension:null,removeExtension:null,logAll:null,logNone:null,logger:null,config:{historyEnabled:true,historyCacheSize:10,refreshOnHistoryMiss:false,defaultSwapStyle:"innerHTML",defaultSwapDelay:0,defaultSettleDelay:20,includeIndicatorStyles:true,indicatorClass:"htmx-indicator",requestClass:"htmx-request",addedClass:"htmx-added",settlingClass:"htmx-settling",swappingClass:"htmx-swapping",allowEval:true,allowScriptTags:true,inlineScriptNonce:"",inlineStyleNonce:"",attributesToSettle:["class","style","width","height"],withCredentials:false,timeout:0,wsReconnectDelay:"full-jitter",wsBinaryType:"blob",disableSelector:"[hx-disable], [data-hx-disable]",scrollBehavior:"instant",defaultFocusScroll:false,getCacheBusterParam:false,globalViewTransitions:false,methodsThatUseUrlParams:["get","delete"],selfRequestsOnly:true,ignoreTitle:false,scrollIntoViewOnBoost:true,triggerSpecsCache:null,disableInheritance:false,responseHandling:[{code:"204",swap:false},{code:"[23]..",swap:true},{code:"[45]..",swap:false,error:true}],allowNestedOobSwaps:true},parseInterval:null,_:null,version:"2.0.4"};Q.onLoad=j;Q.process=kt;Q.on=ye;Q.off=be;Q.trigger=he;Q.ajax=Rn;Q.find=u;Q.findAll=x;Q.closest=g;Q.remove=z;Q.addClass=K;Q.removeClass=G;Q.toggleClass=W;Q.takeClass=Z;Q.swap=$e;Q.defineExtension=Fn;Q.removeExtension=Bn;Q.logAll=V;Q.logNone=_;Q.parseInterval=d;Q._=e;const n={addTriggerHandler:St,bodyContains:le,canAccessLocalStorage:B,findThisElement:Se,filterValues:hn,swap:$e,hasAttribute:s,getAttributeValue:te,getClosestAttributeValue:re,getClosestMatch:o,getExpressionVars:En,getHeaders:fn,getInputValues:cn,getInternalData:ie,getSwapSpecification:gn,getTriggerSpecs:st,getTarget:Ee,makeFragment:P,mergeObjects:ce,makeSettleInfo:xn,oobSwap:He,querySelectorExt:ae,settleImmediately:Kt,shouldCancel:ht,triggerEvent:he,triggerErrorEvent:fe,withExtensions:Ft};const r=["get","post","put","delete","patch"];const H=r.map(function(e){return"[hx-"+e+"], [data-hx-"+e+"]"}).join(", ");function d(e){if(e==undefined){return undefined}let t=NaN;if(e.slice(-2)=="ms"){t=parseFloat(e.slice(0,-2))}else if(e.slice(-1)=="s"){t=parseFloat(e.slice(0,-1))*1e3}else if(e.slice(-1)=="m"){t=parseFloat(e.slice(0,-1))*1e3*60}else{t=parseFloat(e)}return isNaN(t)?undefined:t}function ee(e,t){return e instanceof Element&&e.getAttribute(t)}function s(e,t){return!!e.hasAttribute&&(e.hasAttribute(t)||e.hasAttribute("data-"+t))}function te(e,t){return ee(e,t)||ee(e,"data-"+t)}function c(e){const t=e.parentElement;if(!t&&e.parentNode instanceof ShadowRoot)return e.parentNode;return t}function ne(){return document}function m(e,t){return e.getRootNode?e.getRootNode({composed:t}):ne()}function o(e,t){while(e&&!t(e)){e=c(e)}return e||null}function i(e,t,n){const r=te(t,n);const o=te(t,"hx-disinherit");var i=te(t,"hx-inherit");if(e!==t){if(Q.config.disableInheritance){if(i&&(i==="*"||i.split(" ").indexOf(n)>=0)){return r}else{return null}}if(o&&(o==="*"||o.split(" ").indexOf(n)>=0)){return"unset"}}return r}function re(t,n){let r=null;o(t,function(e){return!!(r=i(t,ue(e),n))});if(r!=="unset"){return r}}function h(e,t){const n=e instanceof Element&&(e.matches||e.matchesSelector||e.msMatchesSelector||e.mozMatchesSelector||e.webkitMatchesSelector||e.oMatchesSelector);return!!n&&n.call(e,t)}function T(e){const t=/<([a-z][^\/\0>\x20\t\r\n\f]*)/i;const n=t.exec(e);if(n){return n[1].toLowerCase()}else{return""}}function q(e){const t=new DOMParser;return t.parseFromString(e,"text/html")}function L(e,t){while(t.childNodes.length>0){e.append(t.childNodes[0])}}function A(e){const t=ne().createElement("script");se(e.attributes,function(e){t.setAttribute(e.name,e.value)});t.textContent=e.textContent;t.async=false;if(Q.config.inlineScriptNonce){t.nonce=Q.config.inlineScriptNonce}return t}function N(e){return e.matches("script")&&(e.type==="text/javascript"||e.type==="module"||e.type==="")}function I(e){Array.from(e.querySelectorAll("script")).forEach(e=>{if(N(e)){const t=A(e);const n=e.parentNode;try{n.insertBefore(t,e)}catch(e){O(e)}finally{e.remove()}}})}function P(e){const t=e.replace(/<head(\s[^>]*)?>[\s\S]*?<\/head>/i,"");const n=T(t);let r;if(n==="html"){r=new DocumentFragment;const i=q(e);L(r,i.body);r.title=i.title}else if(n==="body"){r=new DocumentFragment;const i=q(t);L(r,i.body);r.title=i.title}else{const i=q('<body><template class="internal-htmx-wrapper">'+t+"</template></body>");r=i.querySelector("template").content;r.title=i.title;var o=r.querySelector("title");if(o&&o.parentNode===r){o.remove();r.title=o.innerText}}if(r){if(Q.config.allowScriptTags){I(r)}else{r.querySelectorAll("script").forEach(e=>e.remove())}}return r}function oe(e){if(e){e()}}function t(e,t){return Object.prototype.toString.call(e)==="[object "+t+"]"}function k(e){return typeof e==="function"}function D(e){return t(e,"Object")}function ie(e){const t="htmx-internal-data";let n=e[t];if(!n){n=e[t]={}}return n}function M(t){const n=[];if(t){for(let e=0;e<t.length;e++){n.push(t[e])}}return n}function se(t,n){if(t){for(let e=0;e<t.length;e++){n(t[e])}}}function X(e){const t=e.getBoundingClientRect();const n=t.top;const r=t.bottom;return n<window.innerHeight&&r>=0}function le(e){return e.getRootNode({composed:true})===document}function F(e){return e.trim().split(/\s+/)}function ce(e,t){for(const n in t){if(t.hasOwnProperty(n)){e[n]=t[n]}}return e}function S(e){try{return JSON.parse(e)}catch(e){O(e);return null}}function B(){const e="htmx:localStorageTest";try{localStorage.setItem(e,e);localStorage.removeItem(e);return true}catch(e){return false}}function U(t){try{const e=new URL(t);if(e){t=e.pathname+e.search}if(!/^\/$/.test(t)){t=t.replace(/\/+$/,"")}return t}catch(e){return t}}function e(e){return vn(ne().body,function(){return eval(e)})}function j(t){const e=Q.on("htmx:load",function(e){t(e.detail.elt)});return e}function V(){Q.logger=function(e,t,n){if(console){console.log(t,e,n)}}}function _(){Q.logger=null}function u(e,t){if(typeof e!=="string"){return e.querySelector(t)}else{return u(ne(),e)}}function x(e,t){if(typeof e!=="string"){return e.querySelectorAll(t)}else{return x(ne(),e)}}function E(){return window}function z(e,t){e=y(e);if(t){E().setTimeout(function(){z(e);e=null},t)}else{c(e).removeChild(e)}}function ue(e){return e instanceof Element?e:null}function $(e){return e instanceof HTMLElement?e:null}function J(e){return typeof e==="string"?e:null}function f(e){return e instanceof Element||e instanceof Document||e instanceof DocumentFragment?e:null}function K(e,t,n){e=ue(y(e));if(!e){return}if(n){E().setTimeout(function(){K(e,t);e=null},n)}else{e.classList&&e.classList.add(t)}}function G(e,t,n){let r=ue(y(e));if(!r){return}if(n){E().setTimeout(function(){G(r,t);r=null},n)}else{if(r.classList){r.classList.remove(t);if(r.classList.length===0){r.removeAttribute("class")}}}}function W(e,t){e=y(e);e.classList.toggle(t)}function Z(e,t){e=y(e);se(e.parentElement.children,function(e){G(e,t)});K(ue(e),t)}function g(e,t){e=ue(y(e));if(e&&e.closest){return e.closest(t)}else{do{if(e==null||h(e,t)){return e}}while(e=e&&ue(c(e)));return null}}function l(e,t){return e.substring(0,t.length)===t}function Y(e,t){return e.substring(e.length-t.length)===t}function ge(e){const t=e.trim();if(l(t,"<")&&Y(t,"/>")){return t.substring(1,t.length-2)}else{return t}}function p(t,r,n){if(r.indexOf("global ")===0){return p(t,r.slice(7),true)}t=y(t);const o=[];{let t=0;let n=0;for(let e=0;e<r.length;e++){const l=r[e];if(l===","&&t===0){o.push(r.substring(n,e));n=e+1;continue}if(l==="<"){t++}else if(l==="/"&&e<r.length-1&&r[e+1]===">"){t--}}if(n<r.length){o.push(r.substring(n))}}const i=[];const s=[];while(o.length>0){const r=ge(o.shift());let e;if(r.indexOf("closest ")===0){e=g(ue(t),ge(r.substr(8)))}else if(r.indexOf("find ")===0){e=u(f(t),ge(r.substr(5)))}else if(r==="next"||r==="nextElementSibling"){e=ue(t).nextElementSibling}else if(r.indexOf("next ")===0){e=pe(t,ge(r.substr(5)),!!n)}else if(r==="previous"||r==="previousElementSibling"){e=ue(t).previousElementSibling}else if(r.indexOf("previous ")===0){e=me(t,ge(r.substr(9)),!!n)}else if(r==="document"){e=document}else if(r==="window"){e=window}else if(r==="body"){e=document.body}else if(r==="root"){e=m(t,!!n)}else if(r==="host"){e=t.getRootNode().host}else{s.push(r)}if(e){i.push(e)}}if(s.length>0){const e=s.join(",");const c=f(m(t,!!n));i.push(...M(c.querySelectorAll(e)))}return i}var pe=function(t,e,n){const r=f(m(t,n)).querySelectorAll(e);for(let e=0;e<r.length;e++){const o=r[e];if(o.compareDocumentPosition(t)===Node.DOCUMENT_POSITION_PRECEDING){return o}}};var me=function(t,e,n){const r=f(m(t,n)).querySelectorAll(e);for(let e=r.length-1;e>=0;e--){const o=r[e];if(o.compareDocumentPosition(t)===Node.DOCUMENT_POSITION_FOLLOWING){return o}}};function ae(e,t){if(typeof e!=="string"){return p(e,t)[0]}else{return p(ne().body,e)[0]}}function y(e,t){if(typeof e==="string"){return u(f(t)||document,e)}else{return e}}function xe(e,t,n,r){if(k(t)){return{target:ne().body,event:J(e),listener:t,options:n}}else{return{target:y(e),event:J(t),listener:n,options:r}}}function ye(t,n,r,o){Vn(function(){const e=xe(t,n,r,o);e.target.addEventListener(e.event,e.listener,e.options)});const e=k(n);return e?n:r}function be(t,n,r){Vn(function(){const e=xe(t,n,r);e.target.removeEventListener(e.event,e.listener)});return k(n)?n:r}const ve=ne().createElement("output");function we(e,t){const n=re(e,t);if(n){if(n==="this"){return[Se(e,t)]}else{const r=p(e,n);if(r.length===0){O('The selector "'+n+'" on '+t+" returned no matches!");return[ve]}else{return r}}}}function Se(e,t){return ue(o(e,function(e){return te(ue(e),t)!=null}))}function Ee(e){const t=re(e,"hx-target");if(t){if(t==="this"){return Se(e,"hx-target")}else{return ae(e,t)}}else{const n=ie(e);if(n.boosted){return ne().body}else{return e}}}function Ce(t){const n=Q.config.attributesToSettle;for(let e=0;e<n.length;e++){if(t===n[e]){return true}}return false}function Oe(t,n){se(t.attributes,function(e){if(!n.hasAttribute(e.name)&&Ce(e.name)){t.removeAttribute(e.name)}});se(n.attributes,function(e){if(Ce(e.name)){t.setAttribute(e.name,e.value)}})}function Re(t,e){const n=Un(e);for(let e=0;e<n.length;e++){const r=n[e];try{if(r.isInlineSwap(t)){return true}}catch(e){O(e)}}return t==="outerHTML"}function He(e,o,i,t){t=t||ne();let n="#"+ee(o,"id");let s="outerHTML";if(e==="true"){}else if(e.indexOf(":")>0){s=e.substring(0,e.indexOf(":"));n=e.substring(e.indexOf(":")+1)}else{s=e}o.removeAttribute("hx-swap-oob");o.removeAttribute("data-hx-swap-oob");const r=p(t,n,false);if(r){se(r,function(e){let t;const n=o.cloneNode(true);t=ne().createDocumentFragment();t.appendChild(n);if(!Re(s,e)){t=f(n)}const r={shouldSwap:true,target:e,fragment:t};if(!he(e,"htmx:oobBeforeSwap",r))return;e=r.target;if(r.shouldSwap){qe(t);_e(s,e,e,t,i);Te()}se(i.elts,function(e){he(e,"htmx:oobAfterSwap",r)})});o.parentNode.removeChild(o)}else{o.parentNode.removeChild(o);fe(ne().body,"htmx:oobErrorNoTarget",{content:o})}return e}function Te(){const e=u("#--htmx-preserve-pantry--");if(e){for(const t of[...e.children]){const n=u("#"+t.id);n.parentNode.moveBefore(t,n);n.remove()}e.remove()}}function qe(e){se(x(e,"[hx-preserve], [data-hx-preserve]"),function(e){const t=te(e,"id");const n=ne().getElementById(t);if(n!=null){if(e.moveBefore){let e=u("#--htmx-preserve-pantry--");if(e==null){ne().body.insertAdjacentHTML("afterend","<div id='--htmx-preserve-pantry--'></div>");e=u("#--htmx-preserve-pantry--")}e.moveBefore(n,null)}else{e.parentNode.replaceChild(n,e)}}})}function Le(l,e,c){se(e.querySelectorAll("[id]"),function(t){const n=ee(t,"id");if(n&&n.length>0){const r=n.replace("'","\\'");const o=t.tagName.replace(":","\\:");const e=f(l);const i=e&&e.querySelector(o+"[id='"+r+"']");if(i&&i!==e){const s=t.cloneNode();Oe(t,i);c.tasks.push(function(){Oe(t,s)})}}})}function Ae(e){return function(){G(e,Q.config.addedClass);kt(ue(e));Ne(f(e));he(e,"htmx:load")}}function Ne(e){const t="[autofocus]";const n=$(h(e,t)?e:e.querySelector(t));if(n!=null){n.focus()}}function a(e,t,n,r){Le(e,n,r);while(n.childNodes.length>0){const o=n.firstChild;K(ue(o),Q.config.addedClass);e.insertBefore(o,t);if(o.nodeType!==Node.TEXT_NODE&&o.nodeType!==Node.COMMENT_NODE){r.tasks.push(Ae(o))}}}function Ie(e,t){let n=0;while(n<e.length){t=(t<<5)-t+e.charCodeAt(n++)|0}return t}function Pe(t){let n=0;if(t.attributes){for(let e=0;e<t.attributes.length;e++){const r=t.attributes[e];if(r.value){n=Ie(r.name,n);n=Ie(r.value,n)}}}return n}function ke(t){const n=ie(t);if(n.onHandlers){for(let e=0;e<n.onHandlers.length;e++){const r=n.onHandlers[e];be(t,r.event,r.listener)}delete n.onHandlers}}function De(e){const t=ie(e);if(t.timeout){clearTimeout(t.timeout)}if(t.listenerInfos){se(t.listenerInfos,function(e){if(e.on){be(e.on,e.trigger,e.listener)}})}ke(e);se(Object.keys(t),function(e){if(e!=="firstInitCompleted")delete t[e]})}function b(e){he(e,"htmx:beforeCleanupElement");De(e);if(e.children){se(e.children,function(e){b(e)})}}function Me(t,e,n){if(t instanceof Element&&t.tagName==="BODY"){return Ve(t,e,n)}let r;const o=t.previousSibling;const i=c(t);if(!i){return}a(i,t,e,n);if(o==null){r=i.firstChild}else{r=o.nextSibling}n.elts=n.elts.filter(function(e){return e!==t});while(r&&r!==t){if(r instanceof Element){n.elts.push(r)}r=r.nextSibling}b(t);if(t instanceof Element){t.remove()}else{t.parentNode.removeChild(t)}}function Xe(e,t,n){return a(e,e.firstChild,t,n)}function Fe(e,t,n){return a(c(e),e,t,n)}function Be(e,t,n){return a(e,null,t,n)}function Ue(e,t,n){return a(c(e),e.nextSibling,t,n)}function je(e){b(e);const t=c(e);if(t){return t.removeChild(e)}}function Ve(e,t,n){const r=e.firstChild;a(e,r,t,n);if(r){while(r.nextSibling){b(r.nextSibling);e.removeChild(r.nextSibling)}b(r);e.removeChild(r)}}function _e(t,e,n,r,o){switch(t){case"none":return;case"outerHTML":Me(n,r,o);return;case"afterbegin":Xe(n,r,o);return;case"beforebegin":Fe(n,r,o);return;case"beforeend":Be(n,r,o);return;case"afterend":Ue(n,r,o);return;case"delete":je(n);return;default:var i=Un(e);for(let e=0;e<i.length;e++){const s=i[e];try{const l=s.handleSwap(t,n,r,o);if(l){if(Array.isArray(l)){for(let e=0;e<l.length;e++){const c=l[e];if(c.nodeType!==Node.TEXT_NODE&&c.nodeType!==Node.COMMENT_NODE){o.tasks.push(Ae(c))}}}return}}catch(e){O(e)}}if(t==="innerHTML"){Ve(n,r,o)}else{_e(Q.config.defaultSwapStyle,e,n,r,o)}}}function ze(e,n,r){var t=x(e,"[hx-swap-oob], [data-hx-swap-oob]");se(t,function(e){if(Q.config.allowNestedOobSwaps||e.parentElement===null){const t=te(e,"hx-swap-oob");if(t!=null){He(t,e,n,r)}}else{e.removeAttribute("hx-swap-oob");e.removeAttribute("data-hx-swap-oob")}});return t.length>0}function $e(e,t,r,o){if(!o){o={}}e=y(e);const i=o.contextElement?m(o.contextElement,false):ne();const n=document.activeElement;let s={};try{s={elt:n,start:n?n.selectionStart:null,end:n?n.selectionEnd:null}}catch(e){}const l=xn(e);if(r.swapStyle==="textContent"){e.textContent=t}else{let n=P(t);l.title=n.title;if(o.selectOOB){const u=o.selectOOB.split(",");for(let t=0;t<u.length;t++){const a=u[t].split(":",2);let e=a[0].trim();if(e.indexOf("#")===0){e=e.substring(1)}const f=a[1]||"true";const h=n.querySelector("#"+e);if(h){He(f,h,l,i)}}}ze(n,l,i);se(x(n,"template"),function(e){if(e.content&&ze(e.content,l,i)){e.remove()}});if(o.select){const d=ne().createDocumentFragment();se(n.querySelectorAll(o.select),function(e){d.appendChild(e)});n=d}qe(n);_e(r.swapStyle,o.contextElement,e,n,l);Te()}if(s.elt&&!le(s.elt)&&ee(s.elt,"id")){const g=document.getElementById(ee(s.elt,"id"));const p={preventScroll:r.focusScroll!==undefined?!r.focusScroll:!Q.config.defaultFocusScroll};if(g){if(s.start&&g.setSelectionRange){try{g.setSelectionRange(s.start,s.end)}catch(e){}}g.focus(p)}}e.classList.remove(Q.config.swappingClass);se(l.elts,function(e){if(e.classList){e.classList.add(Q.config.settlingClass)}he(e,"htmx:afterSwap",o.eventInfo)});if(o.afterSwapCallback){o.afterSwapCallback()}if(!r.ignoreTitle){kn(l.title)}const c=function(){se(l.tasks,function(e){e.call()});se(l.elts,function(e){if(e.classList){e.classList.remove(Q.config.settlingClass)}he(e,"htmx:afterSettle",o.eventInfo)});if(o.anchor){const e=ue(y("#"+o.anchor));if(e){e.scrollIntoView({block:"start",behavior:"auto"})}}yn(l.elts,r);if(o.afterSettleCallback){o.afterSettleCallback()}};if(r.settleDelay>0){E().setTimeout(c,r.settleDelay)}else{c()}}function Je(e,t,n){const r=e.getResponseHeader(t);if(r.indexOf("{")===0){const o=S(r);for(const i in o){if(o.hasOwnProperty(i)){let e=o[i];if(D(e)){n=e.target!==undefined?e.target:n}else{e={value:e}}he(n,i,e)}}}else{const s=r.split(",");for(let e=0;e<s.length;e++){he(n,s[e].trim(),[])}}}const Ke=/\s/;const v=/[\s,]/;const Ge=/[_$a-zA-Z]/;const We=/[_$a-zA-Z0-9]/;const Ze=['"',"'","/"];const w=/[^\s]/;const Ye=/[{(]/;const Qe=/[})]/;function et(e){const t=[];let n=0;while(n<e.length){if(Ge.exec(e.charAt(n))){var r=n;while(We.exec(e.charAt(n+1))){n++}t.push(e.substring(r,n+1))}else if(Ze.indexOf(e.charAt(n))!==-1){const o=e.charAt(n);var r=n;n++;while(n<e.length&&e.charAt(n)!==o){if(e.charAt(n)==="\\"){n++}n++}t.push(e.substring(r,n+1))}else{const i=e.charAt(n);t.push(i)}n++}return t}function tt(e,t,n){return Ge.exec(e.charAt(0))&&e!=="true"&&e!=="false"&&e!=="this"&&e!==n&&t!=="."}function nt(r,o,i){if(o[0]==="["){o.shift();let e=1;let t=" return (function("+i+"){ return (";let n=null;while(o.length>0){const s=o[0];if(s==="]"){e--;if(e===0){if(n===null){t=t+"true"}o.shift();t+=")})";try{const l=vn(r,function(){return Function(t)()},function(){return true});l.source=t;return l}catch(e){fe(ne().body,"htmx:syntax:error",{error:e,source:t});return null}}}else if(s==="["){e++}if(tt(s,n,i)){t+="(("+i+"."+s+") ? ("+i+"."+s+") : (window."+s+"))"}else{t=t+s}n=o.shift()}}}function C(e,t){let n="";while(e.length>0&&!t.test(e[0])){n+=e.shift()}return n}function rt(e){let t;if(e.length>0&&Ye.test(e[0])){e.shift();t=C(e,Qe).trim();e.shift()}else{t=C(e,v)}return t}const ot="input, textarea, select";function it(e,t,n){const r=[];const o=et(t);do{C(o,w);const l=o.length;const c=C(o,/[,\[\s]/);if(c!==""){if(c==="every"){const u={trigger:"every"};C(o,w);u.pollInterval=d(C(o,/[,\[\s]/));C(o,w);var i=nt(e,o,"event");if(i){u.eventFilter=i}r.push(u)}else{const a={trigger:c};var i=nt(e,o,"event");if(i){a.eventFilter=i}C(o,w);while(o.length>0&&o[0]!==","){const f=o.shift();if(f==="changed"){a.changed=true}else if(f==="once"){a.once=true}else if(f==="consume"){a.consume=true}else if(f==="delay"&&o[0]===":"){o.shift();a.delay=d(C(o,v))}else if(f==="from"&&o[0]===":"){o.shift();if(Ye.test(o[0])){var s=rt(o)}else{var s=C(o,v);if(s==="closest"||s==="find"||s==="next"||s==="previous"){o.shift();const h=rt(o);if(h.length>0){s+=" "+h}}}a.from=s}else if(f==="target"&&o[0]===":"){o.shift();a.target=rt(o)}else if(f==="throttle"&&o[0]===":"){o.shift();a.throttle=d(C(o,v))}else if(f==="queue"&&o[0]===":"){o.shift();a.queue=C(o,v)}else if(f==="root"&&o[0]===":"){o.shift();a[f]=rt(o)}else if(f==="threshold"&&o[0]===":"){o.shift();a[f]=C(o,v)}else{fe(e,"htmx:syntax:error",{token:o.shift()})}C(o,w)}r.push(a)}}if(o.length===l){fe(e,"htmx:syntax:error",{token:o.shift()})}C(o,w)}while(o[0]===","&&o.shift());if(n){n[t]=r}return r}function st(e){const t=te(e,"hx-trigger");let n=[];if(t){const r=Q.config.triggerSpecsCache;n=r&&r[t]||it(e,t,r)}if(n.length>0){return n}else if(h(e,"form")){return[{trigger:"submit"}]}else if(h(e,'input[type="button"], input[type="submit"]')){return[{trigger:"click"}]}else if(h(e,ot)){return[{trigger:"change"}]}else{return[{trigger:"click"}]}}function lt(e){ie(e).cancelled=true}function ct(e,t,n){const r=ie(e);r.timeout=E().setTimeout(function(){if(le(e)&&r.cancelled!==true){if(!gt(n,e,Mt("hx:poll:trigger",{triggerSpec:n,target:e}))){t(e)}ct(e,t,n)}},n.pollInterval)}function ut(e){return location.hostname===e.hostname&&ee(e,"href")&&ee(e,"href").indexOf("#")!==0}function at(e){return g(e,Q.config.disableSelector)}function ft(t,n,e){if(t instanceof HTMLAnchorElement&&ut(t)&&(t.target===""||t.target==="_self")||t.tagName==="FORM"&&String(ee(t,"method")).toLowerCase()!=="dialog"){n.boosted=true;let r,o;if(t.tagName==="A"){r="get";o=ee(t,"href")}else{const i=ee(t,"method");r=i?i.toLowerCase():"get";o=ee(t,"action");if(o==null||o===""){o=ne().location.href}if(r==="get"&&o.includes("?")){o=o.replace(/\?[^#]+/,"")}}e.forEach(function(e){pt(t,function(e,t){const n=ue(e);if(at(n)){b(n);return}de(r,o,n,t)},n,e,true)})}}function ht(e,t){const n=ue(t);if(!n){return false}if(e.type==="submit"||e.type==="click"){if(n.tagName==="FORM"){return true}if(h(n,'input[type="submit"], button')&&(h(n,"[form]")||g(n,"form")!==null)){return true}if(n instanceof HTMLAnchorElement&&n.href&&(n.getAttribute("href")==="#"||n.getAttribute("href").indexOf("#")!==0)){return true}}return false}function dt(e,t){return ie(e).boosted&&e instanceof HTMLAnchorElement&&t.type==="click"&&(t.ctrlKey||t.metaKey)}function gt(e,t,n){const r=e.eventFilter;if(r){try{return r.call(t,n)!==true}catch(e){const o=r.source;fe(ne().body,"htmx:eventFilter:error",{error:e,source:o});return true}}return false}function pt(l,c,e,u,a){const f=ie(l);let t;if(u.from){t=p(l,u.from)}else{t=[l]}if(u.changed){if(!("lastValue"in f)){f.lastValue=new WeakMap}t.forEach(function(e){if(!f.lastValue.has(u)){f.lastValue.set(u,new WeakMap)}f.lastValue.get(u).set(e,e.value)})}se(t,function(i){const s=function(e){if(!le(l)){i.removeEventListener(u.trigger,s);return}if(dt(l,e)){return}if(a||ht(e,l)){e.preventDefault()}if(gt(u,l,e)){return}const t=ie(e);t.triggerSpec=u;if(t.handledFor==null){t.handledFor=[]}if(t.handledFor.indexOf(l)<0){t.handledFor.push(l);if(u.consume){e.stopPropagation()}if(u.target&&e.target){if(!h(ue(e.target),u.target)){return}}if(u.once){if(f.triggeredOnce){return}else{f.triggeredOnce=true}}if(u.changed){const n=event.target;const r=n.value;const o=f.lastValue.get(u);if(o.has(n)&&o.get(n)===r){return}o.set(n,r)}if(f.delayed){clearTimeout(f.delayed)}if(f.throttle){return}if(u.throttle>0){if(!f.throttle){he(l,"htmx:trigger");c(l,e);f.throttle=E().setTimeout(function(){f.throttle=null},u.throttle)}}else if(u.delay>0){f.delayed=E().setTimeout(function(){he(l,"htmx:trigger");c(l,e)},u.delay)}else{he(l,"htmx:trigger");c(l,e)}}};if(e.listenerInfos==null){e.listenerInfos=[]}e.listenerInfos.push({trigger:u.trigger,listener:s,on:i});i.addEventListener(u.trigger,s)})}let mt=false;let xt=null;function yt(){if(!xt){xt=function(){mt=true};window.addEventListener("scroll",xt);window.addEventListener("resize",xt);setInterval(function(){if(mt){mt=false;se(ne().querySelectorAll("[hx-trigger*='revealed'],[data-hx-trigger*='revealed']"),function(e){bt(e)})}},200)}}function bt(e){if(!s(e,"data-hx-revealed")&&X(e)){e.setAttribute("data-hx-revealed","true");const t=ie(e);if(t.initHash){he(e,"revealed")}else{e.addEventListener("htmx:afterProcessNode",function(){he(e,"revealed")},{once:true})}}}function vt(e,t,n,r){const o=function(){if(!n.loaded){n.loaded=true;he(e,"htmx:trigger");t(e)}};if(r>0){E().setTimeout(o,r)}else{o()}}function wt(t,n,e){let i=false;se(r,function(r){if(s(t,"hx-"+r)){const o=te(t,"hx-"+r);i=true;n.path=o;n.verb=r;e.forEach(function(e){St(t,e,n,function(e,t){const n=ue(e);if(g(n,Q.config.disableSelector)){b(n);return}de(r,o,n,t)})})}});return i}function St(r,e,t,n){if(e.trigger==="revealed"){yt();pt(r,n,t,e);bt(ue(r))}else if(e.trigger==="intersect"){const o={};if(e.root){o.root=ae(r,e.root)}if(e.threshold){o.threshold=parseFloat(e.threshold)}const i=new IntersectionObserver(function(t){for(let e=0;e<t.length;e++){const n=t[e];if(n.isIntersecting){he(r,"intersect");break}}},o);i.observe(ue(r));pt(ue(r),n,t,e)}else if(!t.firstInitCompleted&&e.trigger==="load"){if(!gt(e,r,Mt("load",{elt:r}))){vt(ue(r),n,t,e.delay)}}else if(e.pollInterval>0){t.polling=true;ct(ue(r),n,e)}else{pt(r,n,t,e)}}function Et(e){const t=ue(e);if(!t){return false}const n=t.attributes;for(let e=0;e<n.length;e++){const r=n[e].name;if(l(r,"hx-on:")||l(r,"data-hx-on:")||l(r,"hx-on-")||l(r,"data-hx-on-")){return true}}return false}const Ct=(new XPathEvaluator).createExpression('.//*[@*[ starts-with(name(), "hx-on:") or starts-with(name(), "data-hx-on:") or'+' starts-with(name(), "hx-on-") or starts-with(name(), "data-hx-on-") ]]');function Ot(e,t){if(Et(e)){t.push(ue(e))}const n=Ct.evaluate(e);let r=null;while(r=n.iterateNext())t.push(ue(r))}function Rt(e){const t=[];if(e instanceof DocumentFragment){for(const n of e.childNodes){Ot(n,t)}}else{Ot(e,t)}return t}function Ht(e){if(e.querySelectorAll){const n=", [hx-boost] a, [data-hx-boost] a, a[hx-boost], a[data-hx-boost]";const r=[];for(const i in Mn){const s=Mn[i];if(s.getSelectors){var t=s.getSelectors();if(t){r.push(t)}}}const o=e.querySelectorAll(H+n+", form, [type='submit'],"+" [hx-ext], [data-hx-ext], [hx-trigger], [data-hx-trigger]"+r.flat().map(e=>", "+e).join(""));return o}else{return[]}}function Tt(e){const t=g(ue(e.target),"button, input[type='submit']");const n=Lt(e);if(n){n.lastButtonClicked=t}}function qt(e){const t=Lt(e);if(t){t.lastButtonClicked=null}}function Lt(e){const t=g(ue(e.target),"button, input[type='submit']");if(!t){return}const n=y("#"+ee(t,"form"),t.getRootNode())||g(t,"form");if(!n){return}return ie(n)}function At(e){e.addEventListener("click",Tt);e.addEventListener("focusin",Tt);e.addEventListener("focusout",qt)}function Nt(t,e,n){const r=ie(t);if(!Array.isArray(r.onHandlers)){r.onHandlers=[]}let o;const i=function(e){vn(t,function(){if(at(t)){return}if(!o){o=new Function("event",n)}o.call(t,e)})};t.addEventListener(e,i);r.onHandlers.push({event:e,listener:i})}function It(t){ke(t);for(let e=0;e<t.attributes.length;e++){const n=t.attributes[e].name;const r=t.attributes[e].value;if(l(n,"hx-on")||l(n,"data-hx-on")){const o=n.indexOf("-on")+3;const i=n.slice(o,o+1);if(i==="-"||i===":"){let e=n.slice(o+1);if(l(e,":")){e="htmx"+e}else if(l(e,"-")){e="htmx:"+e.slice(1)}else if(l(e,"htmx-")){e="htmx:"+e.slice(5)}Nt(t,e,r)}}}}function Pt(t){if(g(t,Q.config.disableSelector)){b(t);return}const n=ie(t);const e=Pe(t);if(n.initHash!==e){De(t);n.initHash=e;he(t,"htmx:beforeProcessNode");const r=st(t);const o=wt(t,n,r);if(!o){if(re(t,"hx-boost")==="true"){ft(t,n,r)}else if(s(t,"hx-trigger")){r.forEach(function(e){St(t,e,n,function(){})})}}if(t.tagName==="FORM"||ee(t,"type")==="submit"&&s(t,"form")){At(t)}n.firstInitCompleted=true;he(t,"htmx:afterProcessNode")}}function kt(e){e=y(e);if(g(e,Q.config.disableSelector)){b(e);return}Pt(e);se(Ht(e),function(e){Pt(e)});se(Rt(e),It)}function Dt(e){return e.replace(/([a-z0-9])([A-Z])/g,"$1-$2").toLowerCase()}function Mt(e,t){let n;if(window.CustomEvent&&typeof window.CustomEvent==="function"){n=new CustomEvent(e,{bubbles:true,cancelable:true,composed:true,detail:t})}else{n=ne().createEvent("CustomEvent");n.initCustomEvent(e,true,true,t)}return n}function fe(e,t,n){he(e,t,ce({error:t},n))}function Xt(e){return e==="htmx:afterProcessNode"}function Ft(e,t){se(Un(e),function(e){try{t(e)}catch(e){O(e)}})}function O(e){if(console.error){console.error(e)}else if(console.log){console.log("ERROR: ",e)}}function he(e,t,n){e=y(e);if(n==null){n={}}n.elt=e;const r=Mt(t,n);if(Q.logger&&!Xt(t)){Q.logger(e,t,n)}if(n.error){O(n.error);he(e,"htmx:error",{errorInfo:n})}let o=e.dispatchEvent(r);const i=Dt(t);if(o&&i!==t){const s=Mt(i,r.detail);o=o&&e.dispatchEvent(s)}Ft(ue(e),function(e){o=o&&(e.onEvent(t,r)!==false&&!r.defaultPrevented)});return o}let Bt=location.pathname+location.search;function Ut(){const e=ne().querySelector("[hx-history-elt],[data-hx-history-elt]");return e||ne().body}function jt(t,e){if(!B()){return}const n=_t(e);const r=ne().title;const o=window.scrollY;if(Q.config.historyCacheSize<=0){localStorage.removeItem("htmx-history-cache");return}t=U(t);const i=S(localStorage.getItem("htmx-history-cache"))||[];for(let e=0;e<i.length;e++){if(i[e].url===t){i.splice(e,1);break}}const s={url:t,content:n,title:r,scroll:o};he(ne().body,"htmx:historyItemCreated",{item:s,cache:i});i.push(s);while(i.length>Q.config.historyCacheSize){i.shift()}while(i.length>0){try{localStorage.setItem("htmx-history-cache",JSON.stringify(i));break}catch(e){fe(ne().body,"htmx:historyCacheError",{cause:e,cache:i});i.shift()}}}function Vt(t){if(!B()){return null}t=U(t);const n=S(localStorage.getItem("htmx-history-cache"))||[];for(let e=0;e<n.length;e++){if(n[e].url===t){return n[e]}}return null}function _t(e){const t=Q.config.requestClass;const n=e.cloneNode(true);se(x(n,"."+t),function(e){G(e,t)});se(x(n,"[data-disabled-by-htmx]"),function(e){e.removeAttribute("disabled")});return n.innerHTML}function zt(){const e=Ut();const t=Bt||location.pathname+location.search;let n;try{n=ne().querySelector('[hx-history="false" i],[data-hx-history="false" i]')}catch(e){n=ne().querySelector('[hx-history="false"],[data-hx-history="false"]')}if(!n){he(ne().body,"htmx:beforeHistorySave",{path:t,historyElt:e});jt(t,e)}if(Q.config.historyEnabled)history.replaceState({htmx:true},ne().title,window.location.href)}function $t(e){if(Q.config.getCacheBusterParam){e=e.replace(/org\.htmx\.cache-buster=[^&]*&?/,"");if(Y(e,"&")||Y(e,"?")){e=e.slice(0,-1)}}if(Q.config.historyEnabled){history.pushState({htmx:true},"",e)}Bt=e}function Jt(e){if(Q.config.historyEnabled)history.replaceState({htmx:true},"",e);Bt=e}function Kt(e){se(e,function(e){e.call(undefined)})}function Gt(o){const e=new XMLHttpRequest;const i={path:o,xhr:e};he(ne().body,"htmx:historyCacheMiss",i);e.open("GET",o,true);e.setRequestHeader("HX-Request","true");e.setRequestHeader("HX-History-Restore-Request","true");e.setRequestHeader("HX-Current-URL",ne().location.href);e.onload=function(){if(this.status>=200&&this.status<400){he(ne().body,"htmx:historyCacheMissLoad",i);const e=P(this.response);const t=e.querySelector("[hx-history-elt],[data-hx-history-elt]")||e;const n=Ut();const r=xn(n);kn(e.title);qe(e);Ve(n,t,r);Te();Kt(r.tasks);Bt=o;he(ne().body,"htmx:historyRestore",{path:o,cacheMiss:true,serverResponse:this.response})}else{fe(ne().body,"htmx:historyCacheMissLoadError",i)}};e.send()}function Wt(e){zt();e=e||location.pathname+location.search;const t=Vt(e);if(t){const n=P(t.content);const r=Ut();const o=xn(r);kn(t.title);qe(n);Ve(r,n,o);Te();Kt(o.tasks);E().setTimeout(function(){window.scrollTo(0,t.scroll)},0);Bt=e;he(ne().body,"htmx:historyRestore",{path:e,item:t})}else{if(Q.config.refreshOnHistoryMiss){window.location.reload(true)}else{Gt(e)}}}function Zt(e){let t=we(e,"hx-indicator");if(t==null){t=[e]}se(t,function(e){const t=ie(e);t.requestCount=(t.requestCount||0)+1;e.classList.add.call(e.classList,Q.config.requestClass)});return t}function Yt(e){let t=we(e,"hx-disabled-elt");if(t==null){t=[]}se(t,function(e){const t=ie(e);t.requestCount=(t.requestCount||0)+1;e.setAttribute("disabled","");e.setAttribute("data-disabled-by-htmx","")});return t}function Qt(e,t){se(e.concat(t),function(e){const t=ie(e);t.requestCount=(t.requestCount||1)-1});se(e,function(e){const t=ie(e);if(t.requestCount===0){e.classList.remove.call(e.classList,Q.config.requestClass)}});se(t,function(e){const t=ie(e);if(t.requestCount===0){e.removeAttribute("disabled");e.removeAttribute("data-disabled-by-htmx")}})}function en(t,n){for(let e=0;e<t.length;e++){const r=t[e];if(r.isSameNode(n)){return true}}return false}function tn(e){const t=e;if(t.name===""||t.name==null||t.disabled||g(t,"fieldset[disabled]")){return false}if(t.type==="button"||t.type==="submit"||t.tagName==="image"||t.tagName==="reset"||t.tagName==="file"){return false}if(t.type==="checkbox"||t.type==="radio"){return t.checked}return true}function nn(t,e,n){if(t!=null&&e!=null){if(Array.isArray(e)){e.forEach(function(e){n.append(t,e)})}else{n.append(t,e)}}}function rn(t,n,r){if(t!=null&&n!=null){let e=r.getAll(t);if(Array.isArray(n)){e=e.filter(e=>n.indexOf(e)<0)}else{e=e.filter(e=>e!==n)}r.delete(t);se(e,e=>r.append(t,e))}}function on(t,n,r,o,i){if(o==null||en(t,o)){return}else{t.push(o)}if(tn(o)){const s=ee(o,"name");let e=o.value;if(o instanceof HTMLSelectElement&&o.multiple){e=M(o.querySelectorAll("option:checked")).map(function(e){return e.value})}if(o instanceof HTMLInputElement&&o.files){e=M(o.files)}nn(s,e,n);if(i){sn(o,r)}}if(o instanceof HTMLFormElement){se(o.elements,function(e){if(t.indexOf(e)>=0){rn(e.name,e.value,n)}else{t.push(e)}if(i){sn(e,r)}});new FormData(o).forEach(function(e,t){if(e instanceof File&&e.name===""){return}nn(t,e,n)})}}function sn(e,t){const n=e;if(n.willValidate){he(n,"htmx:validation:validate");if(!n.checkValidity()){t.push({elt:n,message:n.validationMessage,validity:n.validity});he(n,"htmx:validation:failed",{message:n.validationMessage,validity:n.validity})}}}function ln(n,e){for(const t of e.keys()){n.delete(t)}e.forEach(function(e,t){n.append(t,e)});return n}function cn(e,t){const n=[];const r=new FormData;const o=new FormData;const i=[];const s=ie(e);if(s.lastButtonClicked&&!le(s.lastButtonClicked)){s.lastButtonClicked=null}let l=e instanceof HTMLFormElement&&e.noValidate!==true||te(e,"hx-validate")==="true";if(s.lastButtonClicked){l=l&&s.lastButtonClicked.formNoValidate!==true}if(t!=="get"){on(n,o,i,g(e,"form"),l)}on(n,r,i,e,l);if(s.lastButtonClicked||e.tagName==="BUTTON"||e.tagName==="INPUT"&&ee(e,"type")==="submit"){const u=s.lastButtonClicked||e;const a=ee(u,"name");nn(a,u.value,o)}const c=we(e,"hx-include");se(c,function(e){on(n,r,i,ue(e),l);if(!h(e,"form")){se(f(e).querySelectorAll(ot),function(e){on(n,r,i,e,l)})}});ln(r,o);return{errors:i,formData:r,values:An(r)}}function un(e,t,n){if(e!==""){e+="&"}if(String(n)==="[object Object]"){n=JSON.stringify(n)}const r=encodeURIComponent(n);e+=encodeURIComponent(t)+"="+r;return e}function an(e){e=qn(e);let n="";e.forEach(function(e,t){n=un(n,t,e)});return n}function fn(e,t,n){const r={"HX-Request":"true","HX-Trigger":ee(e,"id"),"HX-Trigger-Name":ee(e,"name"),"HX-Target":te(t,"id"),"HX-Current-URL":ne().location.href};bn(e,"hx-headers",false,r);if(n!==undefined){r["HX-Prompt"]=n}if(ie(e).boosted){r["HX-Boosted"]="true"}return r}function hn(n,e){const t=re(e,"hx-params");if(t){if(t==="none"){return new FormData}else if(t==="*"){return n}else if(t.indexOf("not ")===0){se(t.slice(4).split(","),function(e){e=e.trim();n.delete(e)});return n}else{const r=new FormData;se(t.split(","),function(t){t=t.trim();if(n.has(t)){n.getAll(t).forEach(function(e){r.append(t,e)})}});return r}}else{return n}}function dn(e){return!!ee(e,"href")&&ee(e,"href").indexOf("#")>=0}function gn(e,t){const n=t||re(e,"hx-swap");const r={swapStyle:ie(e).boosted?"innerHTML":Q.config.defaultSwapStyle,swapDelay:Q.config.defaultSwapDelay,settleDelay:Q.config.defaultSettleDelay};if(Q.config.scrollIntoViewOnBoost&&ie(e).boosted&&!dn(e)){r.show="top"}if(n){const s=F(n);if(s.length>0){for(let e=0;e<s.length;e++){const l=s[e];if(l.indexOf("swap:")===0){r.swapDelay=d(l.slice(5))}else if(l.indexOf("settle:")===0){r.settleDelay=d(l.slice(7))}else if(l.indexOf("transition:")===0){r.transition=l.slice(11)==="true"}else if(l.indexOf("ignoreTitle:")===0){r.ignoreTitle=l.slice(12)==="true"}else if(l.indexOf("scroll:")===0){const c=l.slice(7);var o=c.split(":");const u=o.pop();var i=o.length>0?o.join(":"):null;r.scroll=u;r.scrollTarget=i}else if(l.indexOf("show:")===0){const a=l.slice(5);var o=a.split(":");const f=o.pop();var i=o.length>0?o.join(":"):null;r.show=f;r.showTarget=i}else if(l.indexOf("focus-scroll:")===0){const h=l.slice("focus-scroll:".length);r.focusScroll=h=="true"}else if(e==0){r.swapStyle=l}else{O("Unknown modifier in hx-swap: "+l)}}}}return r}function pn(e){return re(e,"hx-encoding")==="multipart/form-data"||h(e,"form")&&ee(e,"enctype")==="multipart/form-data"}function mn(t,n,r){let o=null;Ft(n,function(e){if(o==null){o=e.encodeParameters(t,r,n)}});if(o!=null){return o}else{if(pn(n)){return ln(new FormData,qn(r))}else{return an(r)}}}function xn(e){return{tasks:[],elts:[e]}}function yn(e,t){const n=e[0];const r=e[e.length-1];if(t.scroll){var o=null;if(t.scrollTarget){o=ue(ae(n,t.scrollTarget))}if(t.scroll==="top"&&(n||o)){o=o||n;o.scrollTop=0}if(t.scroll==="bottom"&&(r||o)){o=o||r;o.scrollTop=o.scrollHeight}}if(t.show){var o=null;if(t.showTarget){let e=t.showTarget;if(t.showTarget==="window"){e="body"}o=ue(ae(n,e))}if(t.show==="top"&&(n||o)){o=o||n;o.scrollIntoView({block:"start",behavior:Q.config.scrollBehavior})}if(t.show==="bottom"&&(r||o)){o=o||r;o.scrollIntoView({block:"end",behavior:Q.config.scrollBehavior})}}}function bn(r,e,o,i){if(i==null){i={}}if(r==null){return i}const s=te(r,e);if(s){let e=s.trim();let t=o;if(e==="unset"){return null}if(e.indexOf("javascript:")===0){e=e.slice(11);t=true}else if(e.indexOf("js:")===0){e=e.slice(3);t=true}if(e.indexOf("{")!==0){e="{"+e+"}"}let n;if(t){n=vn(r,function(){return Function("return ("+e+")")()},{})}else{n=S(e)}for(const l in n){if(n.hasOwnProperty(l)){if(i[l]==null){i[l]=n[l]}}}}return bn(ue(c(r)),e,o,i)}function vn(e,t,n){if(Q.config.allowEval){return t()}else{fe(e,"htmx:evalDisallowedError");return n}}function wn(e,t){return bn(e,"hx-vars",true,t)}function Sn(e,t){return bn(e,"hx-vals",false,t)}function En(e){return ce(wn(e),Sn(e))}function Cn(t,n,r){if(r!==null){try{t.setRequestHeader(n,r)}catch(e){t.setRequestHeader(n,encodeURIComponent(r));t.setRequestHeader(n+"-URI-AutoEncoded","true")}}}function On(t){if(t.responseURL&&typeof URL!=="undefined"){try{const e=new URL(t.responseURL);return e.pathname+e.search}catch(e){fe(ne().body,"htmx:badResponseUrl",{url:t.responseURL})}}}function R(e,t){return t.test(e.getAllResponseHeaders())}function Rn(t,n,r){t=t.toLowerCase();if(r){if(r instanceof Element||typeof r==="string"){return de(t,n,null,null,{targetOverride:y(r)||ve,returnPromise:true})}else{let e=y(r.target);if(r.target&&!e||r.source&&!e&&!y(r.source)){e=ve}return de(t,n,y(r.source),r.event,{handler:r.handler,headers:r.headers,values:r.values,targetOverride:e,swapOverride:r.swap,select:r.select,returnPromise:true})}}else{return de(t,n,null,null,{returnPromise:true})}}function Hn(e){const t=[];while(e){t.push(e);e=e.parentElement}return t}function Tn(e,t,n){let r;let o;if(typeof URL==="function"){o=new URL(t,document.location.href);const i=document.location.origin;r=i===o.origin}else{o=t;r=l(t,document.location.origin)}if(Q.config.selfRequestsOnly){if(!r){return false}}return he(e,"htmx:validateUrl",ce({url:o,sameHost:r},n))}function qn(e){if(e instanceof FormData)return e;const t=new FormData;for(const n in e){if(e.hasOwnProperty(n)){if(e[n]&&typeof e[n].forEach==="function"){e[n].forEach(function(e){t.append(n,e)})}else if(typeof e[n]==="object"&&!(e[n]instanceof Blob)){t.append(n,JSON.stringify(e[n]))}else{t.append(n,e[n])}}}return t}function Ln(r,o,e){return new Proxy(e,{get:function(t,e){if(typeof e==="number")return t[e];if(e==="length")return t.length;if(e==="push"){return function(e){t.push(e);r.append(o,e)}}if(typeof t[e]==="function"){return function(){t[e].apply(t,arguments);r.delete(o);t.forEach(function(e){r.append(o,e)})}}if(t[e]&&t[e].length===1){return t[e][0]}else{return t[e]}},set:function(e,t,n){e[t]=n;r.delete(o);e.forEach(function(e){r.append(o,e)});return true}})}function An(o){return new Proxy(o,{get:function(e,t){if(typeof t==="symbol"){const r=Reflect.get(e,t);if(typeof r==="function"){return function(){return r.apply(o,arguments)}}else{return r}}if(t==="toJSON"){return()=>Object.fromEntries(o)}if(t in e){if(typeof e[t]==="function"){return function(){return o[t].apply(o,arguments)}}else{return e[t]}}const n=o.getAll(t);if(n.length===0){return undefined}else if(n.length===1){return n[0]}else{return Ln(e,t,n)}},set:function(t,n,e){if(typeof n!=="string"){return false}t.delete(n);if(e&&typeof e.forEach==="function"){e.forEach(function(e){t.append(n,e)})}else if(typeof e==="object"&&!(e instanceof Blob)){t.append(n,JSON.stringify(e))}else{t.append(n,e)}return true},deleteProperty:function(e,t){if(typeof t==="string"){e.delete(t)}return true},ownKeys:function(e){return Reflect.ownKeys(Object.fromEntries(e))},getOwnPropertyDescriptor:function(e,t){return Reflect.getOwnPropertyDescriptor(Object.fromEntries(e),t)}})}function de(t,n,r,o,i,D){let s=null;let l=null;i=i!=null?i:{};if(i.returnPromise&&typeof Promise!=="undefined"){var e=new Promise(function(e,t){s=e;l=t})}if(r==null){r=ne().body}const M=i.handler||Dn;const X=i.select||null;if(!le(r)){oe(s);return e}const c=i.targetOverride||ue(Ee(r));if(c==null||c==ve){fe(r,"htmx:targetError",{target:te(r,"hx-target")});oe(l);return e}let u=ie(r);const a=u.lastButtonClicked;if(a){const L=ee(a,"formaction");if(L!=null){n=L}const A=ee(a,"formmethod");if(A!=null){if(A.toLowerCase()!=="dialog"){t=A}}}const f=re(r,"hx-confirm");if(D===undefined){const K=function(e){return de(t,n,r,o,i,!!e)};const G={target:c,elt:r,path:n,verb:t,triggeringEvent:o,etc:i,issueRequest:K,question:f};if(he(r,"htmx:confirm",G)===false){oe(s);return e}}let h=r;let d=re(r,"hx-sync");let g=null;let F=false;if(d){const N=d.split(":");const I=N[0].trim();if(I==="this"){h=Se(r,"hx-sync")}else{h=ue(ae(r,I))}d=(N[1]||"drop").trim();u=ie(h);if(d==="drop"&&u.xhr&&u.abortable!==true){oe(s);return e}else if(d==="abort"){if(u.xhr){oe(s);return e}else{F=true}}else if(d==="replace"){he(h,"htmx:abort")}else if(d.indexOf("queue")===0){const W=d.split(" ");g=(W[1]||"last").trim()}}if(u.xhr){if(u.abortable){he(h,"htmx:abort")}else{if(g==null){if(o){const P=ie(o);if(P&&P.triggerSpec&&P.triggerSpec.queue){g=P.triggerSpec.queue}}if(g==null){g="last"}}if(u.queuedRequests==null){u.queuedRequests=[]}if(g==="first"&&u.queuedRequests.length===0){u.queuedRequests.push(function(){de(t,n,r,o,i)})}else if(g==="all"){u.queuedRequests.push(function(){de(t,n,r,o,i)})}else if(g==="last"){u.queuedRequests=[];u.queuedRequests.push(function(){de(t,n,r,o,i)})}oe(s);return e}}const p=new XMLHttpRequest;u.xhr=p;u.abortable=F;const m=function(){u.xhr=null;u.abortable=false;if(u.queuedRequests!=null&&u.queuedRequests.length>0){const e=u.queuedRequests.shift();e()}};const B=re(r,"hx-prompt");if(B){var x=prompt(B);if(x===null||!he(r,"htmx:prompt",{prompt:x,target:c})){oe(s);m();return e}}if(f&&!D){if(!confirm(f)){oe(s);m();return e}}let y=fn(r,c,x);if(t!=="get"&&!pn(r)){y["Content-Type"]="application/x-www-form-urlencoded"}if(i.headers){y=ce(y,i.headers)}const U=cn(r,t);let b=U.errors;const j=U.formData;if(i.values){ln(j,qn(i.values))}const V=qn(En(r));const v=ln(j,V);let w=hn(v,r);if(Q.config.getCacheBusterParam&&t==="get"){w.set("org.htmx.cache-buster",ee(c,"id")||"true")}if(n==null||n===""){n=ne().location.href}const S=bn(r,"hx-request");const _=ie(r).boosted;let E=Q.config.methodsThatUseUrlParams.indexOf(t)>=0;const C={boosted:_,useUrlParams:E,formData:w,parameters:An(w),unfilteredFormData:v,unfilteredParameters:An(v),headers:y,target:c,verb:t,errors:b,withCredentials:i.credentials||S.credentials||Q.config.withCredentials,timeout:i.timeout||S.timeout||Q.config.timeout,path:n,triggeringEvent:o};if(!he(r,"htmx:configRequest",C)){oe(s);m();return e}n=C.path;t=C.verb;y=C.headers;w=qn(C.parameters);b=C.errors;E=C.useUrlParams;if(b&&b.length>0){he(r,"htmx:validation:halted",C);oe(s);m();return e}const z=n.split("#");const $=z[0];const O=z[1];let R=n;if(E){R=$;const Z=!w.keys().next().done;if(Z){if(R.indexOf("?")<0){R+="?"}else{R+="&"}R+=an(w);if(O){R+="#"+O}}}if(!Tn(r,R,C)){fe(r,"htmx:invalidPath",C);oe(l);return e}p.open(t.toUpperCase(),R,true);p.overrideMimeType("text/html");p.withCredentials=C.withCredentials;p.timeout=C.timeout;if(S.noHeaders){}else{for(const k in y){if(y.hasOwnProperty(k)){const Y=y[k];Cn(p,k,Y)}}}const H={xhr:p,target:c,requestConfig:C,etc:i,boosted:_,select:X,pathInfo:{requestPath:n,finalRequestPath:R,responsePath:null,anchor:O}};p.onload=function(){try{const t=Hn(r);H.pathInfo.responsePath=On(p);M(r,H);if(H.keepIndicators!==true){Qt(T,q)}he(r,"htmx:afterRequest",H);he(r,"htmx:afterOnLoad",H);if(!le(r)){let e=null;while(t.length>0&&e==null){const n=t.shift();if(le(n)){e=n}}if(e){he(e,"htmx:afterRequest",H);he(e,"htmx:afterOnLoad",H)}}oe(s);m()}catch(e){fe(r,"htmx:onLoadError",ce({error:e},H));throw e}};p.onerror=function(){Qt(T,q);fe(r,"htmx:afterRequest",H);fe(r,"htmx:sendError",H);oe(l);m()};p.onabort=function(){Qt(T,q);fe(r,"htmx:afterRequest",H);fe(r,"htmx:sendAbort",H);oe(l);m()};p.ontimeout=function(){Qt(T,q);fe(r,"htmx:afterRequest",H);fe(r,"htmx:timeout",H);oe(l);m()};if(!he(r,"htmx:beforeRequest",H)){oe(s);m();return e}var T=Zt(r);var q=Yt(r);se(["loadstart","loadend","progress","abort"],function(t){se([p,p.upload],function(e){e.addEventListener(t,function(e){he(r,"htmx:xhr:"+t,{lengthComputable:e.lengthComputable,loaded:e.loaded,total:e.total})})})});he(r,"htmx:beforeSend",H);const J=E?null:mn(p,r,w);p.send(J);return e}function Nn(e,t){const n=t.xhr;let r=null;let o=null;if(R(n,/HX-Push:/i)){r=n.getResponseHeader("HX-Push");o="push"}else if(R(n,/HX-Push-Url:/i)){r=n.getResponseHeader("HX-Push-Url");o="push"}else if(R(n,/HX-Replace-Url:/i)){r=n.getResponseHeader("HX-Replace-Url");o="replace"}if(r){if(r==="false"){return{}}else{return{type:o,path:r}}}const i=t.pathInfo.finalRequestPath;const s=t.pathInfo.responsePath;const l=re(e,"hx-push-url");const c=re(e,"hx-replace-url");const u=ie(e).boosted;let a=null;let f=null;if(l){a="push";f=l}else if(c){a="replace";f=c}else if(u){a="push";f=s||i}if(f){if(f==="false"){return{}}if(f==="true"){f=s||i}if(t.pathInfo.anchor&&f.indexOf("#")===-1){f=f+"#"+t.pathInfo.anchor}return{type:a,path:f}}else{return{}}}function In(e,t){var n=new RegExp(e.code);return n.test(t.toString(10))}function Pn(e){for(var t=0;t<Q.config.responseHandling.length;t++){var n=Q.config.responseHandling[t];if(In(n,e.status)){return n}}return{swap:false}}function kn(e){if(e){const t=u("title");if(t){t.innerHTML=e}else{window.document.title=e}}}function Dn(o,i){const s=i.xhr;let l=i.target;const e=i.etc;const c=i.select;if(!he(o,"htmx:beforeOnLoad",i))return;if(R(s,/HX-Trigger:/i)){Je(s,"HX-Trigger",o)}if(R(s,/HX-Location:/i)){zt();let e=s.getResponseHeader("HX-Location");var t;if(e.indexOf("{")===0){t=S(e);e=t.path;delete t.path}Rn("get",e,t).then(function(){$t(e)});return}const n=R(s,/HX-Refresh:/i)&&s.getResponseHeader("HX-Refresh")==="true";if(R(s,/HX-Redirect:/i)){i.keepIndicators=true;location.href=s.getResponseHeader("HX-Redirect");n&&location.reload();return}if(n){i.keepIndicators=true;location.reload();return}if(R(s,/HX-Retarget:/i)){if(s.getResponseHeader("HX-Retarget")==="this"){i.target=o}else{i.target=ue(ae(o,s.getResponseHeader("HX-Retarget")))}}const u=Nn(o,i);const r=Pn(s);const a=r.swap;let f=!!r.error;let h=Q.config.ignoreTitle||r.ignoreTitle;let d=r.select;if(r.target){i.target=ue(ae(o,r.target))}var g=e.swapOverride;if(g==null&&r.swapOverride){g=r.swapOverride}if(R(s,/HX-Retarget:/i)){if(s.getResponseHeader("HX-Retarget")==="this"){i.target=o}else{i.target=ue(ae(o,s.getResponseHeader("HX-Retarget")))}}if(R(s,/HX-Reswap:/i)){g=s.getResponseHeader("HX-Reswap")}var p=s.response;var m=ce({shouldSwap:a,serverResponse:p,isError:f,ignoreTitle:h,selectOverride:d,swapOverride:g},i);if(r.event&&!he(l,r.event,m))return;if(!he(l,"htmx:beforeSwap",m))return;l=m.target;p=m.serverResponse;f=m.isError;h=m.ignoreTitle;d=m.selectOverride;g=m.swapOverride;i.target=l;i.failed=f;i.successful=!f;if(m.shouldSwap){if(s.status===286){lt(o)}Ft(o,function(e){p=e.transformResponse(p,s,o)});if(u.type){zt()}var x=gn(o,g);if(!x.hasOwnProperty("ignoreTitle")){x.ignoreTitle=h}l.classList.add(Q.config.swappingClass);let n=null;let r=null;if(c){d=c}if(R(s,/HX-Reselect:/i)){d=s.getResponseHeader("HX-Reselect")}const y=re(o,"hx-select-oob");const b=re(o,"hx-select");let e=function(){try{if(u.type){he(ne().body,"htmx:beforeHistoryUpdate",ce({history:u},i));if(u.type==="push"){$t(u.path);he(ne().body,"htmx:pushedIntoHistory",{path:u.path})}else{Jt(u.path);he(ne().body,"htmx:replacedInHistory",{path:u.path})}}$e(l,p,x,{select:d||b,selectOOB:y,eventInfo:i,anchor:i.pathInfo.anchor,contextElement:o,afterSwapCallback:function(){if(R(s,/HX-Trigger-After-Swap:/i)){let e=o;if(!le(o)){e=ne().body}Je(s,"HX-Trigger-After-Swap",e)}},afterSettleCallback:function(){if(R(s,/HX-Trigger-After-Settle:/i)){let e=o;if(!le(o)){e=ne().body}Je(s,"HX-Trigger-After-Settle",e)}oe(n)}})}catch(e){fe(o,"htmx:swapError",i);oe(r);throw e}};let t=Q.config.globalViewTransitions;if(x.hasOwnProperty("transition")){t=x.transition}if(t&&he(o,"htmx:beforeTransition",i)&&typeof Promise!=="undefined"&&document.startViewTransition){const v=new Promise(function(e,t){n=e;r=t});const w=e;e=function(){document.startViewTransition(function(){w();return v})}}if(x.swapDelay>0){E().setTimeout(e,x.swapDelay)}else{e()}}if(f){fe(o,"htmx:responseError",ce({error:"Response Status Error Code "+s.status+" from "+i.pathInfo.requestPath},i))}}const Mn={};function Xn(){return{init:function(e){return null},getSelectors:function(){return null},onEvent:function(e,t){return true},transformResponse:function(e,t,n){return e},isInlineSwap:function(e){return false},handleSwap:function(e,t,n,r){return false},encodeParameters:function(e,t,n){return null}}}function Fn(e,t){if(t.init){t.init(n)}Mn[e]=ce(Xn(),t)}function Bn(e){delete Mn[e]}function Un(e,n,r){if(n==undefined){n=[]}if(e==undefined){return n}if(r==undefined){r=[]}const t=te(e,"hx-ext");if(t){se(t.split(","),function(e){e=e.replace(/ /g,"");if(e.slice(0,7)=="ignore:"){r.push(e.slice(7));return}if(r.indexOf(e)<0){const t=Mn[e];if(t&&n.indexOf(t)<0){n.push(t)}}})}return Un(ue(c(e)),n,r)}var jn=false;ne().addEventListener("DOMContentLoaded",function(){jn=true});function Vn(e){if(jn||ne().readyState==="complete"){e()}else{ne().addEventListener("DOMContentLoaded",e)}}function _n(){if(Q.config.includeIndicatorStyles!==false){const e=Q.config.inlineStyleNonce?` nonce="${Q.config.inlineStyleNonce}"`:"";ne().head.insertAdjacentHTML("beforeend","<style"+e+">      ."+Q.config.indicatorClass+"{opacity:0}      ."+Q.config.requestClass+" ."+Q.config.indicatorClass+"{opacity:1; transition: opacity 200ms ease-in;}      ."+Q.config.requestClass+"."+Q.config.indicatorClass+"{opacity:1; transition: opacity 200ms ease-in;}      </style>")}}function zn(){const e=ne().querySelector('meta[name="htmx-config"]');if(e){return S(e.content)}else{return null}}function $n(){const e=zn();if(e){Q.config=ce(Q.config,e)}}Vn(function(){$n();_n();let e=ne().body;kt(e);const t=ne().querySelectorAll("[hx-trigger='restored'],[data-hx-trigger='restored']");e.addEventListener("htmx:abort",function(e){const t=e.target;const n=ie(t);if(n&&n.xhr){n.xhr.abort()}});const n=window.onpopstate?window.onpopstate.bind(window):null;window.onpopstate=function(e){if(e.state&&e.state.htmx){Wt();se(t,function(e){he(e,"htmx:restored",{document:ne(),triggerEvent:he})})}else{if(n){n(e)}}};E().setTimeout(function(){he(e,"htmx:load",{});e=null},0)});return Q}();

src/web/static/manifest.json

@@ -0,0 +1,32 @@
+{
+  "name": "Research Intelligence",
+  "short_name": "Research",
+  "description": "AI/ML and Security paper triage dashboard",
+  "start_url": "/",
+  "scope": "/",
+  "display": "standalone",
+  "background_color": "#060a13",
+  "theme_color": "#0b1121",
+  "orientation": "any",
+  "categories": ["productivity", "utilities"],
+  "icons": [
+    {
+      "src": "/static/favicon.svg",
+      "type": "image/svg+xml",
+      "sizes": "any",
+      "purpose": "any"
+    },
+    {
+      "src": "/static/favicon-192.png",
+      "type": "image/png",
+      "sizes": "192x192",
+      "purpose": "any"
+    },
+    {
+      "src": "/static/favicon-512.png",
+      "type": "image/png",
+      "sizes": "512x512",
+      "purpose": "any"
+    }
+  ]
+}

src/web/static/style.css

@@ -0,0 +1,1701 @@
+/* ============================================
+   Research Intelligence — Observatory Theme
+
+   Deep navy dark theme with luminous score
+   indicators and editorial typography.
+   ============================================ */
+
+/* ─── Custom Properties ─── */
+:root {
+    --bg-deep: #060a13;
+    --bg: #0b1121;
+    --bg-card: #0f172a;
+    --bg-surface: #1e293b;
+    --bg-hover: #172554;
+
+    --border: rgba(148, 163, 184, 0.08);
+    --border-strong: rgba(148, 163, 184, 0.15);
+    --border-accent: rgba(59, 130, 246, 0.4);
+
+    --text: #f1f5f9;
+    --text-secondary: #cbd5e1;
+    --text-muted: #64748b;
+    --text-dim: #334155;
+
+    --accent: #3b82f6;
+    --accent-hover: #60a5fa;
+    --accent-muted: rgba(59, 130, 246, 0.12);
+
+    --emerald: #10b981;
+    --emerald-glow: rgba(16, 185, 129, 0.2);
+    --amber: #f59e0b;
+    --amber-glow: rgba(245, 158, 11, 0.2);
+    --red: #ef4444;
+    --red-glow: rgba(239, 68, 68, 0.2);
+    --purple: #a78bfa;
+    --purple-glow: rgba(167, 139, 250, 0.15);
+
+    --font-display: system-ui, -apple-system, 'Segoe UI', sans-serif;
+    --font-body: system-ui, -apple-system, 'Segoe UI', Roboto, sans-serif;
+    --font-mono: ui-monospace, 'SF Mono', 'Cascadia Code', Consolas, monospace;
+
+    --radius: 8px;
+    --radius-lg: 12px;
+    --radius-xl: 16px;
+    --radius-full: 9999px;
+
+    --shadow-sm: 0 1px 3px rgba(0, 0, 0, 0.4);
+    --shadow-md: 0 4px 16px rgba(0, 0, 0, 0.4);
+    --shadow-lg: 0 8px 32px rgba(0, 0, 0, 0.5);
+    --shadow-glow: 0 0 20px rgba(59, 130, 246, 0.1);
+
+    --nav-height: 56px;
+}
+
+/* ─── Reset ─── */
+*, *::before, *::after {
+    box-sizing: border-box;
+    margin: 0;
+    padding: 0;
+}
+
+* {
+    scrollbar-width: thin;
+    scrollbar-color: var(--bg-surface) transparent;
+}
+
+::-webkit-scrollbar { width: 6px; height: 6px; }
+::-webkit-scrollbar-track { background: transparent; }
+::-webkit-scrollbar-thumb { background: var(--bg-surface); border-radius: 3px; }
+::-webkit-scrollbar-thumb:hover { background: var(--text-dim); }
+
+::selection {
+    background: rgba(59, 130, 246, 0.3);
+    color: var(--text);
+}
+
+/* ─── Base ─── */
+body {
+    font-family: var(--font-body);
+    background: var(--bg-deep);
+    background-image:
+        radial-gradient(ellipse 80% 60% at 15% -10%, rgba(59, 130, 246, 0.08) 0%, transparent 60%),
+        radial-gradient(ellipse 60% 50% at 85% 110%, rgba(16, 185, 129, 0.04) 0%, transparent 60%);
+    color: var(--text);
+    line-height: 1.6;
+    min-height: 100vh;
+    font-size: 15px;
+    -webkit-font-smoothing: antialiased;
+    -moz-osx-font-smoothing: grayscale;
+}
+
+a {
+    color: var(--accent);
+    text-decoration: none;
+    transition: color 0.15s;
+}
+
+a:hover {
+    color: var(--accent-hover);
+}
+
+/* ─── Page Loader (HTMX) ─── */
+.page-loader {
+    position: fixed;
+    top: 0;
+    left: 0;
+    width: 100%;
+    height: 2px;
+    z-index: 10000;
+    overflow: hidden;
+    opacity: 0;
+    transition: opacity 0.15s;
+}
+
+.page-loader::after {
+    content: '';
+    position: absolute;
+    inset: 0;
+    background: linear-gradient(90deg, transparent, var(--accent), var(--accent-hover), transparent);
+    transform: translateX(-100%);
+}
+
+.htmx-request.page-loader,
+.htmx-request .page-loader {
+    opacity: 1;
+}
+
+.htmx-request.page-loader::after,
+.htmx-request .page-loader::after {
+    animation: loadSlide 1.2s ease-in-out infinite;
+}
+
+/* ─── Navigation ─── */
+nav {
+    background: rgba(11, 17, 33, 0.82);
+    backdrop-filter: blur(24px) saturate(180%);
+    -webkit-backdrop-filter: blur(24px) saturate(180%);
+    border-bottom: 1px solid var(--border);
+    padding: 0 2rem;
+    display: flex;
+    align-items: center;
+    gap: 2.5rem;
+    position: sticky;
+    top: 0;
+    z-index: 1000;
+    height: var(--nav-height);
+}
+
+.logo {
+    font-family: var(--font-display);
+    font-size: 1.15rem;
+    font-weight: 700;
+    color: var(--text);
+    white-space: nowrap;
+    display: flex;
+    align-items: center;
+    gap: 8px;
+    letter-spacing: -0.02em;
+}
+
+.logo-dot {
+    width: 7px;
+    height: 7px;
+    background: var(--accent);
+    border-radius: 50%;
+    flex-shrink: 0;
+    box-shadow: 0 0 6px var(--accent), 0 0 14px rgba(59, 130, 246, 0.3);
+    animation: pulse 3s ease-in-out infinite;
+}
+
+.nav-links {
+    display: flex;
+    gap: 0.25rem;
+    align-items: center;
+}
+
+.nav-links a {
+    color: var(--text-muted);
+    font-size: 0.85rem;
+    font-weight: 500;
+    padding: 0.35rem 0.75rem;
+    border-radius: var(--radius);
+    transition: color 0.15s, background 0.15s;
+    position: relative;
+}
+
+.nav-links a:hover {
+    color: var(--text-secondary);
+    background: var(--accent-muted);
+    text-decoration: none;
+}
+
+.nav-links a.active {
+    color: var(--text);
+    background: var(--accent-muted);
+}
+
+.nav-links a.active::after {
+    content: '';
+    position: absolute;
+    bottom: -1px;
+    left: 0.75rem;
+    right: 0.75rem;
+    height: 2px;
+    background: var(--accent);
+    border-radius: 1px 1px 0 0;
+}
+
+/* ─── Layout ─── */
+.container {
+    max-width: 1280px;
+    margin: 0 auto;
+    padding: 2rem;
+}
+
+.page-header {
+    margin-bottom: 2rem;
+}
+
+.page-header h1 {
+    font-family: var(--font-display);
+    font-size: 1.75rem;
+    font-weight: 700;
+    letter-spacing: -0.03em;
+    line-height: 1.25;
+}
+
+.page-header .subtitle {
+    color: var(--text-muted);
+    font-size: 0.875rem;
+    margin-top: 0.35rem;
+}
+
+/* ─── Stats Grid (Dashboard) ─── */
+.stats-grid {
+    display: grid;
+    grid-template-columns: repeat(4, 1fr);
+    gap: 1rem;
+    margin-bottom: 2.5rem;
+}
+
+.stat-card {
+    border: 1px solid var(--border);
+    border-radius: var(--radius-lg);
+    padding: 1.25rem 1.5rem;
+    position: relative;
+    overflow: hidden;
+    animation: fadeSlideUp 0.45s ease-out both;
+}
+
+.stat-card::before {
+    content: '';
+    position: absolute;
+    inset: 0;
+    opacity: 0.5;
+    border-radius: inherit;
+    pointer-events: none;
+}
+
+.stat-card--blue { background: linear-gradient(145deg, rgba(59,130,246,0.08) 0%, var(--bg-card) 70%); }
+.stat-card--red { background: linear-gradient(145deg, rgba(239,68,68,0.08) 0%, var(--bg-card) 70%); }
+.stat-card--purple { background: linear-gradient(145deg, rgba(167,139,250,0.08) 0%, var(--bg-card) 70%); }
+.stat-card--green { background: linear-gradient(145deg, rgba(16,185,129,0.08) 0%, var(--bg-card) 70%); }
+
+.stat-card:nth-child(1) { animation-delay: 0s; }
+.stat-card:nth-child(2) { animation-delay: 0.06s; }
+.stat-card:nth-child(3) { animation-delay: 0.12s; }
+.stat-card:nth-child(4) { animation-delay: 0.18s; }
+
+.stat-card .label {
+    color: var(--text-muted);
+    font-size: 0.75rem;
+    font-weight: 600;
+    text-transform: uppercase;
+    letter-spacing: 0.06em;
+}
+
+.stat-card .value {
+    font-family: var(--font-mono);
+    font-size: 1.75rem;
+    font-weight: 700;
+    margin-top: 0.5rem;
+    letter-spacing: -0.02em;
+}
+
+.stat-card .value--small {
+    font-size: 0.95rem;
+    font-weight: 500;
+    color: var(--text-secondary);
+}
+
+/* ─── Section Headers ─── */
+.section-header {
+    display: flex;
+    align-items: center;
+    gap: 0.75rem;
+    margin-bottom: 1rem;
+    padding-bottom: 0.75rem;
+    border-bottom: 1px solid var(--border);
+}
+
+.section-header h2 {
+    font-family: var(--font-display);
+    font-size: 1.15rem;
+    font-weight: 600;
+    letter-spacing: -0.02em;
+}
+
+.section-title {
+    font-family: var(--font-display);
+    font-size: 1.05rem;
+    font-weight: 600;
+    letter-spacing: -0.01em;
+    margin-bottom: 0.75rem;
+    display: flex;
+    align-items: center;
+    gap: 0.5rem;
+}
+
+/* ─── Badges ─── */
+.badge {
+    display: inline-flex;
+    align-items: center;
+    font-size: 0.65rem;
+    font-weight: 700;
+    padding: 0.15rem 0.55rem;
+    border-radius: var(--radius-full);
+    letter-spacing: 0.04em;
+    text-transform: uppercase;
+    white-space: nowrap;
+}
+
+.badge--accent { background: var(--accent-muted); color: var(--accent-hover); }
+.badge--red { background: var(--red-glow); color: var(--red); }
+.badge--emerald { background: var(--emerald-glow); color: var(--emerald); }
+.badge--amber { background: var(--amber-glow); color: var(--amber); }
+.badge--purple { background: var(--purple-glow); color: var(--purple); }
+
+.badge-code {
+    font-size: 0.65rem;
+    font-weight: 700;
+    padding: 0.12rem 0.45rem;
+    border-radius: var(--radius-full);
+    background: var(--emerald-glow);
+    color: var(--emerald);
+    letter-spacing: 0.03em;
+}
+
+.badge-hf {
+    font-size: 0.65rem;
+    font-weight: 700;
+    padding: 0.12rem 0.45rem;
+    border-radius: var(--radius-full);
+    background: var(--amber-glow);
+    color: var(--amber);
+    letter-spacing: 0.03em;
+}
+
+.badge-source {
+    font-size: 0.65rem;
+    font-weight: 700;
+    padding: 0.12rem 0.45rem;
+    border-radius: var(--radius-full);
+    background: var(--accent-muted);
+    color: var(--accent);
+    letter-spacing: 0.03em;
+}
+
+/* ─── Two-Column Grid ─── */
+.two-col {
+    display: grid;
+    grid-template-columns: 1fr 1fr;
+    gap: 2rem;
+    margin-bottom: 2rem;
+}
+
+.three-col {
+    display: grid;
+    grid-template-columns: repeat(3, 1fr);
+    gap: 1.5rem;
+    margin-bottom: 2rem;
+}
+
+.events-auto-grid {
+    display: grid;
+    grid-template-columns: repeat(auto-fit, minmax(280px, 1fr));
+    gap: 1.5rem;
+    margin-bottom: 2rem;
+}
+
+/* ─── Paper Cards ─── */
+.paper-card {
+    background: var(--bg-card);
+    border: 1px solid var(--border);
+    border-radius: var(--radius-lg);
+    padding: 1rem 1.25rem;
+    margin-bottom: 0.625rem;
+    transition: border-color 0.2s, box-shadow 0.2s, transform 0.2s;
+    animation: fadeSlideUp 0.4s ease-out both;
+}
+
+.paper-card:nth-child(1) { animation-delay: 0.05s; }
+.paper-card:nth-child(2) { animation-delay: 0.1s; }
+.paper-card:nth-child(3) { animation-delay: 0.15s; }
+.paper-card:nth-child(4) { animation-delay: 0.2s; }
+.paper-card:nth-child(5) { animation-delay: 0.25s; }
+
+.paper-card:hover {
+    border-color: var(--border-accent);
+    box-shadow: var(--shadow-glow);
+    transform: translateY(-1px);
+}
+
+.paper-card .card-top {
+    display: flex;
+    justify-content: space-between;
+    align-items: flex-start;
+    gap: 1rem;
+}
+
+.paper-card .rank {
+    color: var(--text-dim);
+    font-family: var(--font-mono);
+    font-size: 0.75rem;
+    font-weight: 600;
+}
+
+.paper-card .title {
+    font-weight: 600;
+    font-size: 0.9rem;
+    margin: 0.2rem 0 0.35rem;
+    line-height: 1.45;
+}
+
+.paper-card .title a {
+    color: var(--text);
+    transition: color 0.15s;
+}
+
+.paper-card .title a:hover {
+    color: var(--accent-hover);
+}
+
+.paper-card .meta {
+    display: flex;
+    align-items: center;
+    gap: 0.5rem;
+    font-size: 0.75rem;
+    color: var(--text-muted);
+    flex-wrap: wrap;
+}
+
+.paper-card .score-badge {
+    font-family: var(--font-mono);
+    font-size: 1.1rem;
+    font-weight: 700;
+    min-width: 2.5rem;
+    text-align: right;
+    line-height: 1;
+}
+
+.paper-card .score-badge.high { color: var(--emerald); text-shadow: 0 0 12px var(--emerald-glow); }
+.paper-card .score-badge.mid { color: var(--amber); text-shadow: 0 0 12px var(--amber-glow); }
+.paper-card .score-badge.low { color: var(--red); text-shadow: 0 0 12px var(--red-glow); }
+
+.paper-card .summary-text {
+    font-size: 0.82rem;
+    color: var(--text-muted);
+    margin-top: 0.5rem;
+    line-height: 1.55;
+    display: -webkit-box;
+    -webkit-line-clamp: 3;
+    -webkit-box-orient: vertical;
+    overflow: hidden;
+}
+
+.paper-card .score-mini-track {
+    height: 3px;
+    background: var(--bg-deep);
+    border-radius: 2px;
+    margin-top: 0.75rem;
+    overflow: hidden;
+}
+
+.paper-card .score-mini-fill {
+    height: 100%;
+    border-radius: 2px;
+    transition: width 0.6s cubic-bezier(0.16, 1, 0.3, 1);
+}
+
+.paper-card .score-mini-fill.high { background: linear-gradient(90deg, #059669, #10b981); }
+.paper-card .score-mini-fill.mid { background: linear-gradient(90deg, #d97706, #f59e0b); }
+.paper-card .score-mini-fill.low { background: linear-gradient(90deg, #dc2626, #ef4444); }
+
+/* ─── Paper Table ─── */
+.paper-table {
+    width: 100%;
+    border-collapse: separate;
+    border-spacing: 0;
+    font-size: 0.84rem;
+}
+
+.paper-table thead th {
+    text-align: left;
+    padding: 0.625rem 0.75rem;
+    color: var(--text-muted);
+    font-weight: 600;
+    font-size: 0.7rem;
+    text-transform: uppercase;
+    letter-spacing: 0.06em;
+    border-bottom: 1px solid var(--border-strong);
+    white-space: nowrap;
+    position: sticky;
+    top: var(--nav-height);
+    background: var(--bg-deep);
+    z-index: 10;
+}
+
+.paper-table tbody tr {
+    transition: background 0.1s;
+}
+
+.paper-table tbody tr:hover td {
+    background: rgba(59, 130, 246, 0.04);
+}
+
+.paper-table td {
+    padding: 0.6rem 0.75rem;
+    border-bottom: 1px solid var(--border);
+    vertical-align: middle;
+}
+
+.paper-table .col-rank {
+    width: 2.5rem;
+    text-align: center;
+    color: var(--text-dim);
+    font-family: var(--font-mono);
+    font-size: 0.75rem;
+}
+
+.paper-table .col-score {
+    width: 4.5rem;
+    text-align: center;
+    font-family: var(--font-mono);
+    font-weight: 600;
+    cursor: help;
+}
+
+.paper-table .col-score.composite {
+    font-weight: 700;
+}
+
+.paper-table .col-code {
+    width: 2.5rem;
+    text-align: center;
+}
+
+.paper-table .col-summary {
+    color: var(--text-muted);
+    font-size: 0.8rem;
+    max-width: 300px;
+}
+
+.paper-table .paper-title-link {
+    color: var(--text);
+    font-weight: 500;
+    transition: color 0.15s;
+}
+
+.paper-table .paper-title-link:hover {
+    color: var(--accent-hover);
+}
+
+/* Score colors in table */
+.score-high { color: var(--emerald); }
+.score-mid { color: var(--amber); }
+.score-low { color: var(--red); }
+
+/* ─── Score Visualization ─── */
+.score-track {
+    height: 6px;
+    background: var(--bg-deep);
+    border-radius: 3px;
+    overflow: hidden;
+    position: relative;
+}
+
+.score-track--sm { height: 4px; }
+
+.score-track--lg { height: 8px; }
+
+.score-fill {
+    height: 100%;
+    border-radius: 3px;
+    min-width: 2px;
+    animation: fillBar 0.8s cubic-bezier(0.16, 1, 0.3, 1) both;
+}
+
+.score-fill.high {
+    background: linear-gradient(90deg, #059669, #34d399);
+    box-shadow: 0 0 10px var(--emerald-glow);
+}
+
+.score-fill.mid {
+    background: linear-gradient(90deg, #d97706, #fbbf24);
+    box-shadow: 0 0 10px var(--amber-glow);
+}
+
+.score-fill.low {
+    background: linear-gradient(90deg, #dc2626, #f87171);
+    box-shadow: 0 0 10px var(--red-glow);
+}
+
+/* ─── Score Detail (Paper Detail Page) ─── */
+.score-grid {
+    display: grid;
+    grid-template-columns: repeat(auto-fit, minmax(220px, 1fr));
+    gap: 1rem;
+    margin: 1.5rem 0;
+}
+
+.score-item {
+    background: var(--bg-card);
+    border: 1px solid var(--border);
+    padding: 1rem 1.25rem;
+    border-radius: var(--radius-lg);
+}
+
+.score-item--composite {
+    border-color: var(--border-accent);
+    background: linear-gradient(145deg, rgba(59,130,246,0.06) 0%, var(--bg-card) 60%);
+}
+
+.score-item .label {
+    font-size: 0.75rem;
+    font-weight: 600;
+    color: var(--text-muted);
+    text-transform: uppercase;
+    letter-spacing: 0.04em;
+}
+
+.score-item .score-value {
+    font-family: var(--font-mono);
+    font-size: 1.5rem;
+    font-weight: 700;
+    margin: 0.4rem 0;
+    display: flex;
+    align-items: baseline;
+    gap: 0.25rem;
+}
+
+.score-item .score-value .max {
+    font-size: 0.85rem;
+    color: var(--text-dim);
+    font-weight: 400;
+}
+
+.score-item .score-track {
+    margin-top: 0.5rem;
+}
+
+/* ─── Paper Detail ─── */
+.paper-detail {
+    background: var(--bg-card);
+    border: 1px solid var(--border);
+    border-radius: var(--radius-xl);
+    padding: 2rem;
+    animation: fadeSlideUp 0.4s ease-out both;
+}
+
+.paper-detail .back-link {
+    font-size: 0.82rem;
+    color: var(--text-muted);
+    display: inline-flex;
+    align-items: center;
+    gap: 0.35rem;
+    transition: color 0.15s;
+}
+
+.paper-detail .back-link:hover {
+    color: var(--accent);
+}
+
+.paper-detail h1 {
+    font-family: var(--font-display);
+    font-size: 1.5rem;
+    font-weight: 700;
+    letter-spacing: -0.03em;
+    margin-top: 0.75rem;
+    margin-bottom: 0.5rem;
+    line-height: 1.3;
+}
+
+.paper-detail .authors {
+    color: var(--text-muted);
+    font-size: 0.875rem;
+    margin-bottom: 1.25rem;
+}
+
+.paper-summary {
+    margin: 1.25rem 0;
+    padding: 1rem 1.25rem;
+    background: var(--bg);
+    border-radius: var(--radius);
+    border-left: 3px solid var(--accent);
+    font-size: 0.9rem;
+    line-height: 1.7;
+    color: var(--text-secondary);
+}
+
+.paper-reasoning {
+    color: var(--text-muted);
+    font-style: italic;
+    font-size: 0.875rem;
+    margin: 0.75rem 0;
+    line-height: 1.6;
+}
+
+.paper-links {
+    display: flex;
+    gap: 0.5rem;
+    margin: 1.25rem 0;
+    flex-wrap: wrap;
+}
+
+.paper-links a {
+    padding: 0.4rem 0.85rem;
+    background: var(--bg);
+    border: 1px solid var(--border-strong);
+    border-radius: var(--radius);
+    font-size: 0.82rem;
+    font-weight: 500;
+    color: var(--text-secondary);
+    transition: border-color 0.15s, color 0.15s, background 0.15s;
+}
+
+.paper-links a:hover {
+    border-color: var(--accent);
+    color: var(--accent);
+    background: var(--accent-muted);
+    text-decoration: none;
+}
+
+.paper-abstract {
+    line-height: 1.75;
+    margin: 1.25rem 0;
+    padding: 1.25rem;
+    background: var(--bg);
+    border-radius: var(--radius-lg);
+    font-size: 0.9rem;
+    color: var(--text-secondary);
+}
+
+.paper-abstract strong {
+    color: var(--text);
+    font-size: 0.8rem;
+    text-transform: uppercase;
+    letter-spacing: 0.04em;
+}
+
+.paper-meta {
+    margin-top: 1rem;
+    font-size: 0.82rem;
+    color: var(--text-muted);
+}
+
+.paper-meta strong {
+    color: var(--text-secondary);
+}
+
+.context-block {
+    margin-top: 2rem;
+    padding: 1.25rem;
+    background: var(--bg);
+    border-radius: var(--radius-lg);
+    border: 1px solid var(--border);
+}
+
+.context-block .context-label {
+    font-size: 0.75rem;
+    font-weight: 600;
+    color: var(--text-muted);
+    text-transform: uppercase;
+    letter-spacing: 0.04em;
+    margin-bottom: 0.5rem;
+}
+
+.context-block pre {
+    font-family: var(--font-mono);
+    font-size: 0.78rem;
+    color: var(--text-muted);
+    white-space: pre-wrap;
+    line-height: 1.65;
+}
+
+/* ─── Filter Bar ─── */
+.filter-bar {
+    background: var(--bg-card);
+    border: 1px solid var(--border);
+    border-radius: var(--radius-lg);
+    padding: 0.75rem 1rem;
+    margin-bottom: 1.25rem;
+}
+
+.filter-bar form {
+    display: flex;
+    gap: 0.75rem;
+    align-items: center;
+    flex-wrap: wrap;
+    width: 100%;
+}
+
+.filter-bar input[type="search"],
+.filter-bar input[type="number"],
+.filter-bar select {
+    background: var(--bg);
+    border: 1px solid var(--border-strong);
+    border-radius: var(--radius);
+    color: var(--text);
+    padding: 0.45rem 0.75rem;
+    font-size: 0.84rem;
+    font-family: var(--font-body);
+    transition: border-color 0.15s, box-shadow 0.15s;
+}
+
+.filter-bar input:focus,
+.filter-bar select:focus {
+    outline: none;
+    border-color: var(--accent);
+    box-shadow: 0 0 0 2px var(--accent-muted);
+}
+
+.filter-bar input[type="search"] {
+    flex: 1;
+    min-width: 200px;
+}
+
+.filter-bar input[type="number"] {
+    width: 5rem;
+}
+
+.filter-bar label {
+    font-size: 0.8rem;
+    color: var(--text-muted);
+    display: flex;
+    align-items: center;
+    gap: 0.35rem;
+    cursor: pointer;
+    white-space: nowrap;
+}
+
+.filter-bar input[type="checkbox"] {
+    appearance: none;
+    width: 16px;
+    height: 16px;
+    border: 1.5px solid var(--border-strong);
+    border-radius: 4px;
+    background: var(--bg);
+    cursor: pointer;
+    position: relative;
+    transition: background 0.15s, border-color 0.15s;
+}
+
+.filter-bar input[type="checkbox"]:checked {
+    background: var(--accent);
+    border-color: var(--accent);
+}
+
+.filter-bar input[type="checkbox"]:checked::after {
+    content: '';
+    position: absolute;
+    left: 4px;
+    top: 1px;
+    width: 5px;
+    height: 9px;
+    border: solid var(--bg-deep);
+    border-width: 0 2px 2px 0;
+    transform: rotate(45deg);
+}
+
+.filter-bar input[type="search"]::placeholder {
+    color: var(--text-dim);
+}
+
+/* ─── Events ─── */
+.event-section {
+    margin-bottom: 2rem;
+}
+
+.event-section .section-title {
+    font-size: 1rem;
+}
+
+.event-card {
+    background: var(--bg-card);
+    border: 1px solid var(--border);
+    border-radius: var(--radius);
+    padding: 0.85rem 1rem;
+    margin-bottom: 0.5rem;
+    border-left: 3px solid transparent;
+    transition: border-color 0.15s, transform 0.15s;
+}
+
+.event-card:hover {
+    transform: translateX(2px);
+}
+
+.event-card--conference { border-left-color: var(--purple); }
+.event-card--release { border-left-color: var(--emerald); }
+.event-card--news { border-left-color: var(--accent); }
+
+.event-card .event-title {
+    font-weight: 600;
+    font-size: 0.875rem;
+}
+
+.event-card .event-title a {
+    color: var(--text);
+    transition: color 0.15s;
+}
+
+.event-card .event-title a:hover {
+    color: var(--accent-hover);
+}
+
+.event-card .event-meta {
+    font-size: 0.78rem;
+    color: var(--text-muted);
+    margin-top: 0.2rem;
+}
+
+.event-card .event-desc {
+    font-size: 0.82rem;
+    color: var(--text-muted);
+    margin-top: 0.35rem;
+    line-height: 1.5;
+}
+
+/* ─── Buttons ─── */
+.btn {
+    display: inline-flex;
+    align-items: center;
+    justify-content: center;
+    gap: 0.4rem;
+    padding: 0.5rem 1rem;
+    border-radius: var(--radius);
+    font-size: 0.84rem;
+    font-weight: 600;
+    font-family: var(--font-body);
+    cursor: pointer;
+    border: 1px solid var(--border-strong);
+    background: var(--bg-card);
+    color: var(--text-secondary);
+    transition: all 0.15s;
+    white-space: nowrap;
+}
+
+.btn:hover {
+    background: var(--bg-surface);
+    border-color: var(--text-dim);
+    color: var(--text);
+    text-decoration: none;
+}
+
+.btn-primary {
+    background: var(--accent);
+    color: var(--bg-deep);
+    border-color: var(--accent);
+    font-weight: 700;
+}
+
+.btn-primary:hover {
+    background: var(--accent-hover);
+    border-color: var(--accent-hover);
+    color: var(--bg-deep);
+    box-shadow: 0 0 16px rgba(59, 130, 246, 0.25);
+}
+
+.btn-sm {
+    padding: 0.35rem 0.7rem;
+    font-size: 0.78rem;
+}
+
+.btn-ghost {
+    background: transparent;
+    border-color: transparent;
+    color: var(--text-muted);
+}
+
+.btn-ghost:hover {
+    background: var(--accent-muted);
+    color: var(--accent);
+    border-color: transparent;
+}
+
+/* ─── Action Row ─── */
+.action-row {
+    display: flex;
+    gap: 0.75rem;
+    margin-top: 1.5rem;
+    flex-wrap: wrap;
+}
+
+/* ─── Pagination ─── */
+.pagination {
+    display: flex;
+    gap: 0.5rem;
+    margin-top: 1.5rem;
+    justify-content: center;
+    align-items: center;
+}
+
+.pagination .page-info {
+    color: var(--text-muted);
+    font-size: 0.82rem;
+    font-family: var(--font-mono);
+    padding: 0 0.75rem;
+}
+
+/* ─── Empty State ─── */
+.empty-state {
+    text-align: center;
+    padding: 4rem 2rem;
+    color: var(--text-muted);
+}
+
+.empty-state h2 {
+    font-family: var(--font-display);
+    color: var(--text-secondary);
+    margin-bottom: 0.5rem;
+    font-weight: 600;
+}
+
+.empty-state p {
+    max-width: 400px;
+    margin: 0 auto;
+    font-size: 0.9rem;
+}
+
+/* ─── Status Pills ─── */
+.status-running {
+    color: var(--amber);
+    position: relative;
+    padding-left: 14px;
+}
+
+.status-running::before {
+    content: '';
+    position: absolute;
+    left: 0;
+    top: 50%;
+    transform: translateY(-50%);
+    width: 6px;
+    height: 6px;
+    background: var(--amber);
+    border-radius: 50%;
+    animation: pulse 1.5s ease-in-out infinite;
+}
+
+.status-completed {
+    color: var(--emerald);
+    padding-left: 14px;
+    position: relative;
+}
+
+.status-completed::before {
+    content: '';
+    position: absolute;
+    left: 0;
+    top: 50%;
+    transform: translateY(-50%);
+    width: 6px;
+    height: 6px;
+    background: var(--emerald);
+    border-radius: 50%;
+}
+
+.status-failed {
+    color: var(--red);
+    padding-left: 14px;
+    position: relative;
+}
+
+.status-failed::before {
+    content: '';
+    position: absolute;
+    left: 0;
+    top: 50%;
+    transform: translateY(-50%);
+    width: 6px;
+    height: 6px;
+    background: var(--red);
+    border-radius: 50%;
+}
+
+/* ─── HTMX ─── */
+.htmx-indicator {
+    display: none;
+}
+
+.htmx-request .htmx-indicator,
+.htmx-request.htmx-indicator {
+    display: block;
+}
+
+.spinner {
+    width: 14px;
+    height: 14px;
+    border: 2px solid var(--border-strong);
+    border-top-color: var(--accent);
+    border-radius: 50%;
+    animation: spin 0.6s linear infinite;
+}
+
+/* ─── Code check in table ─── */
+.code-check {
+    color: var(--emerald);
+    font-size: 0.9rem;
+}
+
+.code-dash {
+    color: var(--text-dim);
+}
+
+/* ─── Animations ─── */
+@keyframes fadeSlideUp {
+    from {
+        opacity: 0;
+        transform: translateY(12px);
+    }
+    to {
+        opacity: 1;
+        transform: translateY(0);
+    }
+}
+
+@keyframes fillBar {
+    from { width: 0 !important; }
+}
+
+@keyframes pulse {
+    0%, 100% { opacity: 1; }
+    50% { opacity: 0.4; }
+}
+
+@keyframes spin {
+    to { transform: rotate(360deg); }
+}
+
+@keyframes loadSlide {
+    0% { transform: translateX(-100%); }
+    100% { transform: translateX(200%); }
+}
+
+@keyframes fadeIn {
+    from { opacity: 0; }
+    to { opacity: 1; }
+}
+
+/* ─── Connected Papers ─── */
+.connected-papers {
+    margin-top: 1rem;
+}
+
+.connection-group {
+    margin-bottom: 1.5rem;
+}
+
+.connection-group__label {
+    font-size: 0.8rem;
+    font-weight: 600;
+    color: var(--text-muted);
+    text-transform: uppercase;
+    letter-spacing: 0.04em;
+    margin-bottom: 0.5rem;
+    display: flex;
+    align-items: center;
+    gap: 0.5rem;
+}
+
+.connection-list {
+    background: var(--bg);
+    border-radius: var(--radius-lg);
+    border: 1px solid var(--border);
+    overflow: hidden;
+}
+
+.connection-item {
+    display: flex;
+    align-items: center;
+    gap: 0.75rem;
+    padding: 0.5rem 1rem;
+    border-bottom: 1px solid var(--border);
+    font-size: 0.82rem;
+    transition: background 0.1s;
+}
+
+.connection-item:last-child {
+    border-bottom: none;
+}
+
+.connection-item:hover {
+    background: rgba(59, 130, 246, 0.04);
+}
+
+.connection-item--in-db {
+    background: rgba(16, 185, 129, 0.03);
+}
+
+.connection-item--in-db:hover {
+    background: rgba(16, 185, 129, 0.06);
+}
+
+.connection-title {
+    flex: 1;
+    min-width: 0;
+    overflow: hidden;
+    text-overflow: ellipsis;
+    white-space: nowrap;
+}
+
+.connection-title a {
+    color: var(--text-secondary);
+}
+
+.connection-title a:hover {
+    color: var(--accent-hover);
+}
+
+.connection-year {
+    color: var(--text-dim);
+    font-family: var(--font-mono);
+    font-size: 0.75rem;
+    flex-shrink: 0;
+}
+
+/* ─── Filter select ─── */
+.filter-bar select {
+    min-width: 120px;
+}
+
+/* ─── Focus States ─── */
+a:focus-visible,
+button:focus-visible,
+input:focus-visible,
+select:focus-visible {
+    outline: 2px solid var(--accent);
+    outline-offset: 2px;
+}
+
+/* ─── Responsive ─── */
+@media (max-width: 1024px) {
+    .stats-grid {
+        grid-template-columns: repeat(2, 1fr);
+    }
+
+    .three-col {
+        grid-template-columns: 1fr 1fr;
+    }
+}
+
+@media (max-width: 768px) {
+    :root {
+        --nav-height: 50px;
+    }
+
+    nav {
+        padding: 0 1rem;
+        gap: 1rem;
+        height: var(--nav-height);
+    }
+
+    .logo {
+        font-size: 1rem;
+    }
+
+    .nav-links a {
+        font-size: 0.8rem;
+        padding: 0.3rem 0.5rem;
+    }
+
+    .container {
+        padding: 1.25rem 1rem;
+    }
+
+    .page-header h1 {
+        font-size: 1.35rem;
+    }
+
+    .stats-grid {
+        grid-template-columns: repeat(2, 1fr);
+        gap: 0.75rem;
+    }
+
+    .two-col {
+        grid-template-columns: 1fr;
+        gap: 1.5rem;
+    }
+
+    .three-col {
+        grid-template-columns: 1fr;
+    }
+
+    .paper-table .col-summary { display: none; }
+
+    .filter-bar form {
+        gap: 0.5rem;
+    }
+
+    .filter-bar input[type="search"] {
+        min-width: 150px;
+    }
+
+    .paper-detail {
+        padding: 1.25rem;
+    }
+
+    .score-grid {
+        grid-template-columns: repeat(2, 1fr);
+    }
+}
+
+@media (max-width: 480px) {
+    :root {
+        --nav-height: auto;
+    }
+
+    nav {
+        flex-wrap: wrap;
+        height: auto;
+        padding: 0.75rem 1rem;
+    }
+
+    .nav-links {
+        width: 100%;
+        overflow-x: auto;
+        -webkit-overflow-scrolling: touch;
+        padding-bottom: 0.25rem;
+    }
+
+    .stats-grid {
+        grid-template-columns: 1fr 1fr;
+    }
+
+    .stat-card {
+        padding: 0.85rem 1rem;
+    }
+
+    .stat-card .value {
+        font-size: 1.35rem;
+    }
+
+    .paper-table th:nth-child(n+4):nth-child(-n+6),
+    .paper-table td:nth-child(n+4):nth-child(-n+6) {
+        display: none;
+    }
+
+    .score-grid {
+        grid-template-columns: 1fr;
+    }
+
+    .paper-links {
+        flex-direction: column;
+    }
+
+    .paper-links a {
+        text-align: center;
+    }
+
+    .action-row {
+        flex-direction: column;
+    }
+
+    .action-row .btn {
+        width: 100%;
+    }
+}
+
+/* ─── Toast notifications ─── */
+.toast-container {
+    position: fixed;
+    bottom: 1.5rem;
+    right: 1.5rem;
+    z-index: 9999;
+    display: flex;
+    flex-direction: column;
+    gap: 0.5rem;
+}
+
+.toast {
+    background: var(--bg-card);
+    border: 1px solid var(--border-strong);
+    border-left: 3px solid var(--accent);
+    border-radius: 6px;
+    padding: 0.75rem 1rem;
+    font-size: 0.85rem;
+    color: var(--text);
+    box-shadow: 0 8px 24px rgba(0, 0, 0, 0.5);
+    animation: toast-in 0.3s ease-out, toast-out 0.3s ease-in 3.7s forwards;
+    max-width: 340px;
+}
+
+.toast--success { border-left-color: var(--emerald); }
+.toast--warning { border-left-color: var(--amber); }
+.toast--error { border-left-color: var(--red); }
+
+@keyframes toast-in {
+    from { opacity: 0; transform: translateY(1rem); }
+    to { opacity: 1; transform: translateY(0); }
+}
+
+@keyframes toast-out {
+    from { opacity: 1; }
+    to { opacity: 0; transform: translateY(-0.5rem); }
+}
+
+/* ─── Signal Buttons ─── */
+.signal-buttons {
+    display: inline-flex;
+    gap: 2px;
+    align-items: center;
+}
+
+.signal-btn {
+    background: transparent;
+    border: 1px solid transparent;
+    border-radius: 4px;
+    color: var(--text-dim);
+    cursor: pointer;
+    font-size: 0.7rem;
+    padding: 2px 4px;
+    line-height: 1;
+    transition: color 0.15s, background 0.15s, border-color 0.15s;
+}
+
+.signal-btn:hover {
+    background: var(--bg-surface);
+    border-color: var(--border-strong);
+}
+
+.signal-btn--up:hover,
+.signal-btn--up.active {
+    color: var(--emerald);
+    background: var(--emerald-glow);
+    border-color: rgba(16, 185, 129, 0.3);
+}
+
+.signal-btn--down:hover,
+.signal-btn--down.active {
+    color: var(--red);
+    background: var(--red-glow);
+    border-color: rgba(239, 68, 68, 0.3);
+}
+
+.signal-btn--dismiss:hover,
+.signal-btn--dismiss.active {
+    color: var(--text-muted);
+    background: var(--bg-surface);
+    border-color: var(--border-strong);
+}
+
+/* Show signal buttons on row hover (desktop) */
+.paper-table .col-signals {
+    width: 5rem;
+    text-align: center;
+}
+
+.paper-table .col-signals-header {
+    width: 5rem;
+    text-align: center;
+}
+
+.paper-table tbody tr .signal-buttons {
+    opacity: 0.3;
+    transition: opacity 0.15s;
+}
+
+.paper-table tbody tr:hover .signal-buttons {
+    opacity: 1;
+}
+
+/* Always show if a signal is active */
+.paper-table tbody tr .signal-buttons:has(.active) {
+    opacity: 1;
+}
+
+/* ─── Boost Indicators ─── */
+.boost-arrow {
+    font-size: 0.6rem;
+    margin-left: 2px;
+    vertical-align: middle;
+}
+
+.boost-up {
+    color: var(--emerald);
+}
+
+.boost-down {
+    color: var(--red);
+}
+
+.score-raw {
+    font-size: 0.6rem;
+    color: var(--text-dim);
+    font-weight: 400;
+    margin-top: 1px;
+}
+
+.boost-pip {
+    font-size: 0.55rem;
+    line-height: 1;
+}
+
+.boost-pip--up {
+    color: var(--emerald);
+}
+
+.boost-pip--down {
+    color: var(--red);
+}
+
+/* Boost detail in score item */
+.boost-detail {
+    margin-top: 0.5rem;
+    font-size: 0.75rem;
+    display: flex;
+    align-items: center;
+    gap: 0.35rem;
+    flex-wrap: wrap;
+}
+
+.boost-label {
+    color: var(--text-muted);
+}
+
+.boost-value {
+    font-family: var(--font-mono);
+    font-weight: 600;
+    font-size: 0.8rem;
+}
+
+/* ─── Discovery Badge ─── */
+.badge--discover {
+    background: var(--purple-glow);
+    color: var(--purple);
+    font-size: 0.55rem;
+    padding: 0.1rem 0.4rem;
+    margin-left: 0.35rem;
+    vertical-align: middle;
+}
+
+/* ─── Preference Explanation (Paper Detail) ─── */
+.pref-explanation {
+    background: var(--bg);
+    border: 1px solid var(--border);
+    border-left: 3px solid var(--purple);
+    border-radius: var(--radius);
+    padding: 0.75rem 1rem;
+    margin: 0.75rem 0;
+}
+
+.pref-explanation__label {
+    font-size: 0.7rem;
+    font-weight: 600;
+    text-transform: uppercase;
+    letter-spacing: 0.04em;
+    color: var(--purple);
+    margin-bottom: 0.4rem;
+}
+
+.pref-explanation__reasons {
+    display: flex;
+    gap: 0.5rem;
+    flex-wrap: wrap;
+}
+
+.pref-reason {
+    font-family: var(--font-mono);
+    font-size: 0.75rem;
+    color: var(--text-secondary);
+    background: var(--bg-card);
+    padding: 0.2rem 0.5rem;
+    border-radius: 4px;
+    border: 1px solid var(--border);
+}
+
+/* ─── Cold Start Hint ─── */
+.cold-start-hint {
+    text-align: center;
+    padding: 0.75rem;
+    color: var(--text-dim);
+    font-size: 0.82rem;
+    border-top: 1px solid var(--border);
+    margin-top: 0.5rem;
+}
+
+/* ─── Preferences Page ─── */
+.pref-groups {
+    display: grid;
+    gap: 1.5rem;
+}
+
+.pref-group {
+    background: var(--bg-card);
+    border: 1px solid var(--border);
+    border-radius: var(--radius-lg);
+    padding: 1.25rem;
+}
+
+.pref-group .section-header {
+    margin-bottom: 0.75rem;
+    padding-bottom: 0.5rem;
+}
+
+.pref-list {
+    display: flex;
+    flex-direction: column;
+    gap: 0.35rem;
+}
+
+.pref-item {
+    display: flex;
+    align-items: center;
+    gap: 0.75rem;
+    padding: 0.4rem 0.5rem;
+    border-radius: 4px;
+    transition: background 0.1s;
+}
+
+.pref-item:hover {
+    background: rgba(59, 130, 246, 0.04);
+}
+
+.pref-item__name {
+    font-size: 0.84rem;
+    color: var(--text);
+    min-width: 160px;
+    overflow: hidden;
+    text-overflow: ellipsis;
+    white-space: nowrap;
+}
+
+.pref-item__count {
+    font-family: var(--font-mono);
+    font-size: 0.7rem;
+    color: var(--text-dim);
+    min-width: 2.5rem;
+    text-align: right;
+}
+
+.pref-bar-container {
+    flex: 1;
+    height: 6px;
+    background: var(--bg-deep);
+    border-radius: 3px;
+    overflow: hidden;
+    min-width: 60px;
+}
+
+.pref-bar {
+    height: 100%;
+    border-radius: 3px;
+    min-width: 2px;
+    transition: width 0.4s ease;
+}
+
+.pref-bar--positive {
+    background: linear-gradient(90deg, #059669, #34d399);
+}
+
+.pref-bar--negative {
+    background: linear-gradient(90deg, #dc2626, #f87171);
+    float: right;
+}
+
+.pref-item__value {
+    font-family: var(--font-mono);
+    font-size: 0.78rem;
+    font-weight: 600;
+    min-width: 3.5rem;
+    text-align: right;
+}
+
+.pref-positive {
+    color: var(--emerald);
+}
+
+.pref-negative {
+    color: var(--red);
+}
+
+/* Responsive: preferences page */
+@media (max-width: 768px) {
+    .pref-item__name {
+        min-width: 100px;
+    }
+
+    .pref-bar-container {
+        min-width: 40px;
+    }
+}

src/web/static/sw.js

@@ -0,0 +1,79 @@
+const CACHE_NAME = 'ri-v4';
+const STATIC_ASSETS = [
+  '/static/style.css',
+  '/static/htmx.min.js',
+  '/static/favicon.svg',
+  '/static/manifest.json',
+];
+
+// Install: pre-cache static assets
+self.addEventListener('install', (event) => {
+  event.waitUntil(
+    caches.open(CACHE_NAME).then((cache) => cache.addAll(STATIC_ASSETS))
+  );
+  self.skipWaiting();
+});
+
+// Activate: clean up old caches
+self.addEventListener('activate', (event) => {
+  event.waitUntil(
+    caches.keys().then((keys) =>
+      Promise.all(keys.filter((k) => k !== CACHE_NAME).map((k) => caches.delete(k)))
+    )
+  );
+  self.clients.claim();
+});
+
+// Fetch: network-first for pages, cache-first for static assets
+self.addEventListener('fetch', (event) => {
+  const url = new URL(event.request.url);
+
+  // Static assets: cache-first
+  if (url.pathname.startsWith('/static/')) {
+    event.respondWith(
+      caches.match(event.request).then((cached) => {
+        if (cached) return cached;
+        return fetch(event.request).then((response) => {
+          const clone = response.clone();
+          caches.open(CACHE_NAME).then((cache) => cache.put(event.request, clone));
+          return response;
+        });
+      })
+    );
+    return;
+  }
+
+  // Google Fonts: cache-first
+  if (url.hostname.includes('fonts.googleapis.com') || url.hostname.includes('fonts.gstatic.com')) {
+    event.respondWith(
+      caches.match(event.request).then((cached) => {
+        if (cached) return cached;
+        return fetch(event.request).then((response) => {
+          const clone = response.clone();
+          caches.open(CACHE_NAME).then((cache) => cache.put(event.request, clone));
+          return response;
+        });
+      })
+    );
+    return;
+  }
+
+  // HTML pages: network-first with cache fallback
+  if (event.request.mode === 'navigate' || event.request.headers.get('accept')?.includes('text/html')) {
+    event.respondWith(
+      fetch(event.request)
+        .then((response) => {
+          const clone = response.clone();
+          caches.open(CACHE_NAME).then((cache) => cache.put(event.request, clone));
+          return response;
+        })
+        .catch(() => caches.match(event.request).then((cached) => cached || caches.match('/')))
+    );
+    return;
+  }
+
+  // Everything else: network with cache fallback
+  event.respondWith(
+    fetch(event.request).catch(() => caches.match(event.request))
+  );
+});

src/web/templates/base.html

@@ -0,0 +1,60 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+    <meta charset="utf-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1">
+    <title>{% block title %}Research Intelligence{% endblock %}</title>
+    <meta name="description" content="Research paper triage — AI/ML and Security">
+    <meta name="theme-color" content="#0b1121">
+    <meta name="apple-mobile-web-app-capable" content="yes">
+    <meta name="apple-mobile-web-app-status-bar-style" content="black-translucent">
+    <meta name="apple-mobile-web-app-title" content="Research">
+    <link rel="manifest" href="/static/manifest.json">
+    <link rel="icon" href="/static/favicon.svg" type="image/svg+xml">
+    <link rel="apple-touch-icon" href="/static/favicon-192.png">
+    <link rel="stylesheet" href="/static/style.css">
+    <script src="/static/htmx.min.js"></script>
+</head>
+<body>
+    <div class="page-loader htmx-indicator" id="page-loader"></div>
+    <nav>
+        <a href="/" class="logo" style="text-decoration:none">
+            <span class="logo-dot"></span>
+            Research Intelligence
+        </a>
+        <div class="nav-links">
+            <a href="/" class="{% if active == 'dashboard' %}active{% endif %}">Dashboard</a>
+            <a href="/papers/aiml" class="{% if active == 'aiml' %}active{% endif %}">AI / ML</a>
+            <a href="/papers/security" class="{% if active == 'security' %}active{% endif %}">Security</a>
+            <a href="/github" class="{% if active == 'github' %}active{% endif %}">GitHub</a>
+            <a href="/events" class="{% if active == 'events' %}active{% endif %}">Events</a>
+            <a href="/weeks" class="{% if active == 'weeks' %}active{% endif %}">Archive</a>
+            <a href="/preferences" class="{% if active == 'preferences' %}active{% endif %}" title="Preferences">&#9881;</a>
+        </div>
+    </nav>
+    <div class="container">
+        {% block content %}{% endblock %}
+    </div>
+    <div class="toast-container" id="toasts"></div>
+    <script>
+    if ('serviceWorker' in navigator) {
+        navigator.serviceWorker.register('/sw.js');
+    }
+    // Toast helper
+    function showToast(msg, type) {
+        var c = document.getElementById('toasts');
+        var t = document.createElement('div');
+        t.className = 'toast' + (type ? ' toast--' + type : '');
+        t.textContent = msg;
+        c.appendChild(t);
+        setTimeout(function() { t.remove(); }, 4000);
+    }
+    // Scroll to top on HTMX content swap (pagination)
+    document.body.addEventListener('htmx:afterSwap', function(e) {
+        if (e.detail.target.id === 'paper-results' || e.detail.target.id === 'gh-results') {
+            e.detail.target.scrollIntoView({behavior: 'smooth', block: 'start'});
+        }
+    });
+    </script>
+</body>
+</html>

src/web/templates/dashboard.html

@@ -0,0 +1,135 @@
+{% extends "base.html" %}
+{% block title %}Dashboard — Research Intelligence{% endblock %}
+{% block content %}
+<div class="page-header">
+    <h1>Week of {{ week_label }}</h1>
+    <div class="subtitle">Research triage overview</div>
+</div>
+
+{% if show_seed_banner is defined and show_seed_banner %}
+<div style="background:linear-gradient(135deg, rgba(167,139,250,0.08), rgba(59,130,246,0.06)); border:1px solid var(--border); border-left:3px solid var(--purple); border-radius:var(--radius-lg); padding:1rem 1.25rem; margin-bottom:1.5rem; display:flex; justify-content:space-between; align-items:center; flex-wrap:wrap; gap:0.75rem">
+    <div>
+        <div style="font-weight:600; font-size:0.9rem">New here?</div>
+        <div style="font-size:0.82rem; color:var(--text-muted)">Pick some papers you like to personalize your feed.</div>
+    </div>
+    <a href="/seed-preferences" class="btn btn-sm" style="border-color:var(--purple); color:var(--purple)">Pick Papers</a>
+</div>
+{% endif %}
+
+<div class="stats-grid">
+    <div class="stat-card stat-card--blue">
+        <div class="label">AI/ML Papers</div>
+        <div class="value">{{ aiml_count }}</div>
+    </div>
+    <div class="stat-card stat-card--red">
+        <div class="label">Security Papers</div>
+        <div class="value">{{ security_count }}</div>
+    </div>
+    <div class="stat-card stat-card--green">
+        <div class="label">GitHub Projects</div>
+        <div class="value">{{ github_count }}</div>
+    </div>
+    <div class="stat-card stat-card--purple">
+        <div class="label">Events Tracked</div>
+        <div class="value">{{ event_count }}</div>
+    </div>
+    <div class="stat-card stat-card--purple" style="opacity:0.8">
+        <div class="label">Last Run</div>
+        <div class="value value--small">{{ (last_run or "") | replace("T", " ") or "Never" }}</div>
+    </div>
+</div>
+
+<div class="two-col">
+    <div>
+        <div class="section-header">
+            <h2>AI/ML Top 5</h2>
+            <span class="badge badge--accent">AI/ML</span>
+        </div>
+        {% if aiml_top %}
+            {% for p in aiml_top %}
+            {% set rank = loop.index %}
+            {% include "partials/paper_card.html" %}
+            {% endfor %}
+        {% else %}
+        <div class="empty-state" style="padding:2rem">
+            <p>No AI/ML papers scored yet.</p>
+        </div>
+        {% endif %}
+    </div>
+    <div>
+        <div class="section-header">
+            <h2>Security Top 5</h2>
+            <span class="badge badge--red">Security</span>
+        </div>
+        {% if security_top %}
+            {% for p in security_top %}
+            {% set rank = loop.index %}
+            {% include "partials/paper_card.html" %}
+            {% endfor %}
+        {% else %}
+        <div class="empty-state" style="padding:2rem">
+            <p>No security papers scored yet.</p>
+        </div>
+        {% endif %}
+    </div>
+</div>
+
+{% if events_grouped %}
+<div class="section-header" style="margin-top:0.5rem">
+    <h2>Events This Week</h2>
+</div>
+<div class="events-auto-grid">
+    {% for cat, cat_events in events_grouped.items() %}
+    <div class="event-section">
+        <div class="section-title" style="font-size:0.95rem">{{ cat | title }}</div>
+        {% for e in cat_events[:5] %}
+        <div class="event-card event-card--{{ cat }}">
+            <div class="event-title">
+                {% if e.url %}<a href="{{ e.url }}">{{ e.title }}</a>{% else %}{{ e.title }}{% endif %}
+            </div>
+            <div class="event-meta">{{ e.source }}{% if e.event_date %} · {% if cat == 'conference' %}<span style="color:var(--amber)">{{ e.event_date | format_date('medium') }}</span>{% else %}{{ e.event_date | format_date('medium') }}{% endif %}{% endif %}</div>
+        </div>
+        {% endfor %}
+    </div>
+    {% endfor %}
+</div>
+{% endif %}
+
+<div class="action-row">
+    <button type="button" class="btn btn-primary" id="btn-aiml"
+        {% if 'aiml' in running_pipelines %}disabled style="opacity:0.6"{% endif %}
+        onclick="triggerPipeline('aiml', this)">
+        {% if 'aiml' in running_pipelines %}Running...{% else %}Run AI/ML Pipeline{% endif %}
+    </button>
+    <button type="button" class="btn btn-primary" id="btn-security"
+        {% if 'security' in running_pipelines %}disabled style="opacity:0.6"{% endif %}
+        onclick="triggerPipeline('security', this)">
+        {% if 'security' in running_pipelines %}Running...{% else %}Run Security Pipeline{% endif %}
+    </button>
+    <button type="button" class="btn" id="btn-github"
+        {% if 'github' in running_pipelines %}disabled style="opacity:0.6"{% endif %}
+        onclick="triggerPipeline('github', this)">
+        {% if 'github' in running_pipelines %}Running...{% else %}Run GitHub{% endif %}
+    </button>
+    <button type="button" class="btn" id="btn-events"
+        {% if 'events' in running_pipelines %}disabled style="opacity:0.6"{% endif %}
+        onclick="triggerPipeline('events', this)">
+        {% if 'events' in running_pipelines %}Running...{% else %}Run Events{% endif %}
+    </button>
+</div>
+<script>
+function triggerPipeline(domain, btn) {
+    btn.disabled = true;
+    btn.textContent = 'Starting...';
+    fetch('/run/' + domain, {method: 'POST'}).then(function() {
+        showToast(domain.charAt(0).toUpperCase() + domain.slice(1) + ' pipeline started', 'success');
+        btn.textContent = 'Running...';
+        btn.style.opacity = '0.6';
+    }).catch(function() {
+        showToast('Failed to start pipeline', 'error');
+        btn.disabled = false;
+        btn.textContent = 'Run ' + domain + ' Pipeline';
+    });
+}
+</script>
+{% endblock %}

src/web/templates/events.html

@@ -0,0 +1,91 @@
+{% extends "base.html" %}
+{% block title %}Events — Research Intelligence{% endblock %}
+{% block content %}
+<div class="page-header">
+    <h1>Events</h1>
+    <div class="subtitle">{{ total }} events tracked</div>
+</div>
+
+{% if deadlines %}
+<div class="event-section">
+    <div class="section-header">
+        <h2>Upcoming Deadlines</h2>
+        <span class="badge badge--purple">{{ deadlines | length }}</span>
+    </div>
+    {% for e in deadlines %}
+    <div class="event-card event-card--conference">
+        <div style="display:flex; justify-content:space-between; align-items:flex-start; gap:1rem">
+            <div style="min-width:0">
+                <div class="event-title">
+                    {% if e.url %}<a href="{{ e.url }}">{{ e.title }}</a>{% else %}{{ e.title }}{% endif %}
+                </div>
+                <div class="event-meta">
+                    {{ e.source }}
+                    {% if e.event_date %}· <strong style="color:var(--amber)">Deadline: {{ e.event_date | format_date('medium') }}</strong>{% endif %}
+                </div>
+                {% if e.description %}<div class="event-desc">{{ e.description[:250] }}{% if e.description | length > 250 %}&hellip;{% endif %}</div>{% endif %}
+            </div>
+            {% if e.event_date %}
+            <div style="flex-shrink:0; text-align:right; font-family:var(--font-mono); font-size:0.8rem; color:var(--text-muted); white-space:nowrap">
+                {{ e.event_date | format_date }}
+            </div>
+            {% endif %}
+        </div>
+    </div>
+    {% endfor %}
+</div>
+{% endif %}
+
+{% if releases %}
+<div class="event-section">
+    <div class="section-header">
+        <h2>Notable Releases</h2>
+        <span class="badge badge--emerald">{{ releases | length }}</span>
+    </div>
+    {% for e in releases %}
+    <div class="event-card event-card--release">
+        <div class="event-title">
+            {% if e.url %}<a href="{{ e.url }}">{{ e.title }}</a>{% else %}{{ e.title }}{% endif %}
+        </div>
+        <div class="event-meta">
+            {{ e.source }}
+            {% if e.event_date %}· {{ e.event_date | format_date }}{% endif %}
+            {% if e.relevance_score %}· Relevance: {{ e.relevance_score }}{% endif %}
+        </div>
+        {% if e.description %}<div class="event-desc">{{ e.description[:200] }}{% if e.description | length > 200 %}&hellip;{% endif %}</div>{% endif %}
+    </div>
+    {% endfor %}
+</div>
+{% endif %}
+
+{% if news %}
+<div class="event-section">
+    <div class="section-header">
+        <h2>News</h2>
+        <span class="badge badge--accent">{{ news | length }}</span>
+    </div>
+    {% for e in news %}
+    <div class="event-card event-card--news">
+        <div class="event-title">
+            {% if e.url %}<a href="{{ e.url }}">{{ e.title }}</a>{% else %}{{ e.title }}{% endif %}
+        </div>
+        <div class="event-meta">
+            {{ e.source }}
+            {% if e.event_date %}· {{ e.event_date | format_date('medium') }}{% endif %}
+        </div>
+        {% if e.description %}<div class="event-desc">{{ e.description[:200] }}{% if e.description | length > 200 %}&hellip;{% endif %}</div>{% endif %}
+    </div>
+    {% endfor %}
+</div>
+{% endif %}
+
+{% if not deadlines and not releases and not news %}
+<div class="empty-state">
+    <h2>No events yet</h2>
+    <p>Run the events pipeline to populate this page.</p>
+    <form method="post" action="/run/events" style="margin-top:1rem">
+        <button type="submit" class="btn btn-primary">Run Events Pipeline</button>
+    </form>
+</div>
+{% endif %}
+{% endblock %}

src/web/templates/github.html

@@ -0,0 +1,43 @@
+{% extends "base.html" %}
+{% block title %}GitHub Projects — Research Intelligence{% endblock %}
+{% block content %}
+<div class="page-header">
+    <div style="display:flex; justify-content:space-between; align-items:flex-start; flex-wrap:wrap; gap:0.5rem">
+        <div>
+            <h1>GitHub Projects</h1>
+            <div class="subtitle">{{ total }} trending projects{% if run.date_start %} · {{ run.date_start }} to {{ run.date_end }}{% endif %}</div>
+        </div>
+        <button type="button" class="btn btn-sm btn-ghost" onclick="this.disabled=true;this.textContent='Running...';fetch('/run/github',{method:'POST'}).then(function(){showToast('GitHub pipeline started','success')}).catch(function(){showToast('Pipeline failed','error')})">Refresh Projects</button>
+    </div>
+</div>
+
+<div class="filter-bar">
+    <form hx-get="/github" hx-target="#gh-results" hx-push-url="true" hx-indicator="#page-loader">
+        <input type="search" name="search" value="{{ search or '' }}" placeholder="Search repos...">
+        {% if available_languages %}
+        <select name="language">
+            <option value="">All languages</option>
+            {% for lang in available_languages %}
+            <option value="{{ lang }}" {% if language == lang %}selected{% endif %}>{{ lang }}</option>
+            {% endfor %}
+        </select>
+        {% endif %}
+        <select name="domain">
+            <option value="">All domains</option>
+            <option value="aiml" {% if domain_filter == 'aiml' %}selected{% endif %}>AI/ML</option>
+            <option value="security" {% if domain_filter == 'security' %}selected{% endif %}>Security</option>
+        </select>
+        <select name="sort">
+            <option value="score" {% if not sort or sort == 'score' %}selected{% endif %}>Sort: Score</option>
+            <option value="stars" {% if sort == 'stars' %}selected{% endif %}>Sort: Stars</option>
+            <option value="forks" {% if sort == 'forks' %}selected{% endif %}>Sort: Forks</option>
+            <option value="name" {% if sort == 'name' %}selected{% endif %}>Sort: Name</option>
+        </select>
+        <button type="submit" class="btn btn-primary btn-sm">Filter</button>
+    </form>
+</div>
+
+<div id="gh-results">
+    {% include "partials/github_results.html" %}
+</div>
+{% endblock %}

src/web/templates/paper_detail.html

@@ -0,0 +1,205 @@
+{% extends "base.html" %}
+{% block title %}{{ paper.title }} — Research Intelligence{% endblock %}
+{% block content %}
+<div class="paper-detail">
+    <div style="display:flex; justify-content:space-between; align-items:center; flex-wrap:wrap; gap:0.5rem">
+        <a href="/papers/{{ domain }}" class="back-link">&larr; Back to {{ domain_label }} papers</a>
+        <div style="display:flex; gap:0.5rem; align-items:center">
+            {% set paper_id = paper.id %}
+            {% set user_signal = paper.user_signal if paper.user_signal is defined else None %}
+            {% include "partials/signal_buttons.html" %}
+        </div>
+    </div>
+
+    <h1>{{ paper.title }}</h1>
+
+    <div class="authors">
+        {% if paper.authors is string %}{{ paper.authors }}{% else %}{{ paper.authors | join(", ") }}{% endif %}
+    </div>
+
+    {% if paper.topics is iterable and paper.topics is not string and paper.topics | length > 0 %}
+    <div style="display:flex; gap:0.35rem; margin-bottom:1rem; flex-wrap:wrap">
+        {% for t in paper.topics %}
+        <span class="badge badge--accent">{{ t }}</span>
+        {% endfor %}
+        {% if paper.is_discovery is defined and paper.is_discovery %}
+        <span class="badge badge--discover">DISCOVER</span>
+        {% endif %}
+    </div>
+    {% endif %}
+
+    {% if paper.composite is not none %}
+    <div class="score-grid">
+        {% set axes = [
+            (axis_labels[0], paper.score_axis_1),
+            (axis_labels[1], paper.score_axis_2),
+            (axis_labels[2], paper.score_axis_3)
+        ] %}
+        {% for label, val in axes %}
+        {% set pct = ((val or 0) | float / 10 * 100) | round(0) | int %}
+        {% set level = 'high' if pct >= 65 else ('mid' if pct >= 40 else 'low') %}
+        <div class="score-item">
+            <div class="label">{{ label }}</div>
+            <div class="score-value score-{{ level }}">{{ val | default("&mdash;") }}<span class="max">/10</span></div>
+            <div class="score-track score-track--lg">
+                <div class="score-fill {{ level }}" style="width:{{ pct }}%"></div>
+            </div>
+        </div>
+        {% endfor %}
+        {% set comp_pct = ((paper.composite or 0) | float / 10 * 100) | round(0) | int %}
+        {% set comp_level = 'high' if comp_pct >= 65 else ('mid' if comp_pct >= 40 else 'low') %}
+        <div class="score-item score-item--composite">
+            <div class="label">Composite</div>
+            <div class="score-value score-{{ comp_level }}">{{ paper.composite }}<span class="max">/10</span></div>
+            <div class="score-track score-track--lg">
+                <div class="score-fill {{ comp_level }}" style="width:{{ comp_pct }}%"></div>
+            </div>
+            {% if paper.preference_boost is defined and paper.preference_boost != 0 %}
+            <div class="boost-detail">
+                <span class="boost-label">Preference boost:</span>
+                <span class="boost-value {% if paper.preference_boost > 0 %}boost-up{% else %}boost-down{% endif %}">{{ '%+.2f'|format(paper.preference_boost) }}</span>
+                <span class="boost-label">&rarr; Adjusted:</span>
+                <span class="boost-value">{{ paper.adjusted_score }}</span>
+            </div>
+            {% endif %}
+        </div>
+    </div>
+
+    {% if paper.boost_reasons is defined and paper.boost_reasons | length > 0 %}
+    <div class="pref-explanation">
+        <div class="pref-explanation__label">Preference Signals</div>
+        <div class="pref-explanation__reasons">
+            {% for reason in paper.boost_reasons %}
+            <span class="pref-reason">{{ reason }}</span>
+            {% endfor %}
+        </div>
+    </div>
+    {% endif %}
+    {% endif %}
+
+    {% if paper.summary %}
+    <div class="paper-summary">{{ paper.summary }}</div>
+    {% endif %}
+
+    {% if paper.s2_tldr %}
+    <div style="margin:0.75rem 0; padding:0.75rem 1rem; background:var(--bg); border-radius:var(--radius); border-left:3px solid var(--purple); font-size:0.88rem; color:var(--text-secondary)">
+        <span style="font-size:0.7rem; font-weight:600; text-transform:uppercase; letter-spacing:0.04em; color:var(--purple)">S2 TL;DR</span><br>
+        {{ paper.s2_tldr }}
+    </div>
+    {% endif %}
+
+    {% if paper.reasoning %}
+    <p class="paper-reasoning">{{ paper.reasoning }}</p>
+    {% endif %}
+
+    <div class="paper-links">
+        {% if paper.arxiv_url %}<a href="{{ paper.arxiv_url }}">arXiv</a>{% endif %}
+        {% if paper.pdf_url %}<a href="{{ paper.pdf_url }}">PDF</a>{% endif %}
+        {% if paper.code_url %}<a href="{{ paper.code_url }}">Code</a>{% endif %}
+        {% if paper.github_repo and paper.github_repo != paper.code_url %}<a href="{{ paper.github_repo }}">GitHub</a>{% endif %}
+        {% if paper.hf_models %}
+            {% for m in paper.hf_models[:3] %}
+            <a href="https://huggingface.co/{{ m.id if m is mapping else m }}">Model: {{ (m.id if m is mapping else m)[:30] }}</a>
+            {% endfor %}
+        {% endif %}
+        {% if paper.hf_datasets %}
+            {% for d in paper.hf_datasets[:2] %}
+            <a href="https://huggingface.co/datasets/{{ d.id if d is mapping else d }}">Dataset: {{ (d.id if d is mapping else d)[:30] }}</a>
+            {% endfor %}
+        {% endif %}
+        {% if paper.hf_spaces %}
+            {% for s in paper.hf_spaces[:2] %}
+            <a href="https://huggingface.co/spaces/{{ s.id if s is mapping else s }}">Space: {{ (s.id if s is mapping else s)[:30] }}</a>
+            {% endfor %}
+        {% endif %}
+    </div>
+
+    <div class="paper-abstract">
+        <strong>Abstract</strong><br><br>
+        {{ paper.abstract }}
+    </div>
+
+    {% if paper.categories %}
+    <div class="paper-meta">
+        <strong>Categories:</strong>
+        {% if paper.categories is string %}{{ paper.categories }}{% else %}{{ paper.categories | join(", ") }}{% endif %}
+    </div>
+    {% endif %}
+
+    {% if paper.comment %}
+    <div class="paper-meta" style="margin-top:0.4rem">
+        <strong>Comment:</strong> {{ paper.comment }}
+    </div>
+    {% endif %}
+
+    {# ── Connected Papers ── #}
+    {% if connections and (connections.references or connections.recommendations) %}
+    <div class="connected-papers">
+        <div class="section-header" style="margin-top:2rem">
+            <h2>Connected Papers</h2>
+        </div>
+
+        {% if connections.references %}
+        <div class="connection-group">
+            <div class="connection-group__label">References <span class="badge badge--accent">{{ connections.references | length }}</span></div>
+            <div class="connection-list">
+                {% for c in connections.references[:20] %}
+                <div class="connection-item{% if c.in_db_paper_id %} connection-item--in-db{% endif %}">
+                    <span class="connection-title">
+                        {% if c.in_db_paper_id %}
+                        <a href="/papers/{{ domain }}/{{ c.in_db_paper_id }}">{{ c.connected_title }}</a>
+                        {% elif c.connected_arxiv_id %}
+                        <a href="https://arxiv.org/abs/{{ c.connected_arxiv_id }}">{{ c.connected_title }}</a>
+                        {% elif c.connected_s2_id %}
+                        <a href="https://api.semanticscholar.org/{{ c.connected_s2_id }}">{{ c.connected_title }}</a>
+                        {% else %}
+                        {{ c.connected_title }}
+                        {% endif %}
+                    </span>
+                    {% if c.connected_year %}<span class="connection-year">{{ c.connected_year }}</span>{% endif %}
+                    {% if c.in_db_paper_id %}<span class="badge badge--emerald" style="font-size:0.6rem">IN DB</span>{% endif %}
+                </div>
+                {% endfor %}
+            </div>
+        </div>
+        {% endif %}
+
+        {% if connections.recommendations %}
+        <div class="connection-group">
+            <div class="connection-group__label">Similar Papers <span class="badge badge--purple">{{ connections.recommendations | length }}</span></div>
+            <div class="connection-list">
+                {% for c in connections.recommendations[:15] %}
+                <div class="connection-item{% if c.in_db_paper_id %} connection-item--in-db{% endif %}">
+                    <span class="connection-title">
+                        {% if c.in_db_paper_id %}
+                        <a href="/papers/{{ domain }}/{{ c.in_db_paper_id }}">{{ c.connected_title }}</a>
+                        {% elif c.connected_arxiv_id %}
+                        <a href="https://arxiv.org/abs/{{ c.connected_arxiv_id }}">{{ c.connected_title }}</a>
+                        {% elif c.connected_s2_id %}
+                        <a href="https://api.semanticscholar.org/{{ c.connected_s2_id }}">{{ c.connected_title }}</a>
+                        {% else %}
+                        {{ c.connected_title }}
+                        {% endif %}
+                    </span>
+                    {% if c.connected_year %}<span class="connection-year">{{ c.connected_year }}</span>{% endif %}
+                    {% if c.in_db_paper_id %}<span class="badge badge--emerald" style="font-size:0.6rem">IN DB</span>{% endif %}
+                </div>
+                {% endfor %}
+            </div>
+        </div>
+        {% endif %}
+    </div>
+    {% endif %}
+
+    <div class="context-block">
+        <div class="context-label">Context for Claude Code</div>
+        <pre>Paper: {{ paper.title }}
+arXiv: {{ paper.arxiv_id }}
+Score: {{ paper.composite }}/10
+Summary: {{ paper.summary }}
+{% if paper.code_url %}Code: {{ paper.code_url }}{% endif %}
+
+Tell me more about this paper's approach and results.</pre>
+    </div>
+</div>
+{% endblock %}

src/web/templates/papers.html

@@ -0,0 +1,49 @@
+{% extends "base.html" %}
+{% block title %}{{ domain_label }} Papers — Research Intelligence{% endblock %}
+{% block content %}
+<div class="page-header">
+    <div style="display:flex; justify-content:space-between; align-items:flex-start; flex-wrap:wrap; gap:0.5rem">
+        <div>
+            <h1>{{ domain_label }} Papers</h1>
+            <div class="subtitle">{{ total }} papers scored{% if run.date_start %} · {{ run.date_start }} to {{ run.date_end }}{% endif %}</div>
+        </div>
+        <button type="button" class="btn btn-sm btn-ghost" onclick="this.disabled=true;this.textContent='Enriching...';fetch('/run/enrich/{{ domain }}',{method:'POST'}).then(function(){showToast('S2 enrichment started','success')}).catch(function(){showToast('Enrichment failed','error')})">Enrich with S2</button>
+    </div>
+</div>
+
+<div class="filter-bar">
+    <form hx-get="/papers/{{ domain }}" hx-target="#paper-results" hx-push-url="true" hx-indicator="#page-loader">
+        <input type="search" name="search" value="{{ search or '' }}" placeholder="Search papers...">
+        <label>
+            Min score
+            <input type="number" name="min_score" value="{{ min_score or '' }}" min="0" max="10" step="0.5">
+        </label>
+        <label>
+            <input type="checkbox" name="has_code" value="1" {% if has_code %}checked{% endif %}>
+            Has code
+        </label>
+        {% if available_topics %}
+        <select name="topic">
+            <option value="">All topics</option>
+            {% for t in available_topics %}
+            <option value="{{ t }}" {% if topic == t %}selected{% endif %}>{{ t }}</option>
+            {% endfor %}
+        </select>
+        {% endif %}
+        <select name="sort">
+            <option value="adjusted" {% if sort == 'adjusted' or (not sort and has_preferences) %}selected{% endif %}>Sort: Personalized</option>
+            <option value="score" {% if sort == 'score' or (not sort and not has_preferences) %}selected{% endif %}>Sort: Score</option>
+            <option value="date" {% if sort == 'date' %}selected{% endif %}>Sort: Date</option>
+            <option value="axis1" {% if sort == 'axis1' %}selected{% endif %}>Sort: {{ axis_labels[0] }}</option>
+            <option value="axis2" {% if sort == 'axis2' %}selected{% endif %}>Sort: {{ axis_labels[1] }}</option>
+            <option value="axis3" {% if sort == 'axis3' %}selected{% endif %}>Sort: {{ axis_labels[2] }}</option>
+            <option value="title" {% if sort == 'title' %}selected{% endif %}>Sort: Title</option>
+        </select>
+        <button type="submit" class="btn btn-primary btn-sm">Filter</button>
+    </form>
+</div>
+
+<div id="paper-results">
+    {% include "partials/papers_results.html" %}
+</div>
+{% endblock %}

src/web/templates/partials/github_results.html

@@ -0,0 +1,83 @@
+{% if projects %}
+<table class="paper-table">
+    <thead>
+        <tr>
+            <th class="col-rank">#</th>
+            <th>Repository</th>
+            <th class="col-score">Stars</th>
+            <th class="col-score">Forks</th>
+            <th class="col-score">PRs</th>
+            <th class="col-code">Lang</th>
+            <th class="col-code">Domain</th>
+        </tr>
+    </thead>
+    <tbody>
+        {% for p in projects %}
+        {% set rank = offset + loop.index %}
+        <tr>
+            <td class="col-rank">{{ rank }}</td>
+            <td>
+                <div class="paper-title">
+                    <a href="{{ p.url }}" target="_blank" rel="noopener">{{ p.repo_name }}</a>
+                </div>
+                {% if p.description %}
+                <div class="paper-summary">{{ p.description[:200] }}{% if p.description | length > 200 %}&hellip;{% endif %}</div>
+                {% endif %}
+                {% if p.collection_names %}
+                <div style="margin-top:0.25rem">
+                    {% for tag in p.collection_names.split(',') %}
+                    {% if tag.strip() %}
+                    <span class="badge badge--accent" style="font-size:0.7rem">{{ tag.strip() }}</span>
+                    {% endif %}
+                    {% endfor %}
+                </div>
+                {% endif %}
+            </td>
+            <td class="col-score">
+                <span style="color:var(--amber); font-weight:600">{{ p.stars }}</span>
+            </td>
+            <td class="col-score">{{ p.forks }}</td>
+            <td class="col-score">{{ p.pull_requests }}</td>
+            <td class="col-code">
+                {% if p.language %}
+                <span class="badge" style="font-size:0.7rem">{{ p.language }}</span>
+                {% endif %}
+            </td>
+            <td class="col-code">
+                {% if p.domain == 'aiml' %}
+                <span class="badge badge--accent" style="font-size:0.7rem">AI/ML</span>
+                {% elif p.domain == 'security' %}
+                <span class="badge badge--red" style="font-size:0.7rem">Security</span>
+                {% endif %}
+            </td>
+        </tr>
+        {% endfor %}
+    </tbody>
+</table>
+
+{% set filter_qs %}{% if search %}&search={{ search | urlencode }}{% endif %}{% if language %}&language={{ language | urlencode }}{% endif %}{% if domain_filter %}&domain={{ domain_filter | urlencode }}{% endif %}{% if sort %}&sort={{ sort | urlencode }}{% endif %}{% endset %}
+
+{% if total > limit %}
+<div class="pagination">
+    {% if offset > 0 %}
+    <a href="/github?offset={{ offset - limit }}&limit={{ limit }}{{ filter_qs }}"
+       hx-get="/github?offset={{ offset - limit }}&limit={{ limit }}{{ filter_qs }}"
+       hx-target="#gh-results" hx-push-url="true" hx-indicator="#page-loader"
+       class="btn btn-sm">&larr; Prev</a>
+    {% endif %}
+    <span class="page-info">{{ offset + 1 }}&ndash;{{ [offset + limit, total] | min }} of {{ total }}</span>
+    {% if offset + limit < total %}
+    <a href="/github?offset={{ offset + limit }}&limit={{ limit }}{{ filter_qs }}"
+       hx-get="/github?offset={{ offset + limit }}&limit={{ limit }}{{ filter_qs }}"
+       hx-target="#gh-results" hx-push-url="true" hx-indicator="#page-loader"
+       class="btn btn-sm">Next &rarr;</a>
+    {% endif %}
+</div>
+{% endif %}
+
+{% else %}
+<div class="empty-state">
+    <h2>No projects found</h2>
+    <p>{% if search or language or domain_filter %}Try adjusting your filters.{% else %}Run the GitHub pipeline to discover trending projects.{% endif %}</p>
+</div>
+{% endif %}

src/web/templates/partials/paper_card.html

@@ -0,0 +1,29 @@
+{% set pct = ((p.composite or 0) | float / 10 * 100) | round(0) | int %}
+{% set level = 'high' if pct >= 65 else ('mid' if pct >= 40 else 'low') %}
+<div class="paper-card">
+    <div class="card-top">
+        <div style="min-width:0">
+            <span class="rank">#{{ rank if rank is defined else "" }}</span>
+            <div class="title"><a href="/papers/{{ p.domain }}/{{ p.id }}">{{ p.title }}</a></div>
+            <div class="meta">
+                {% if p.code_url or p.github_repo %}<span class="badge-code">CODE</span>{% endif %}
+                {% if p.hf_models %}<span class="badge-hf">HF</span>{% endif %}
+                {% if p.source == "both" %}<span class="badge-source">HF+arXiv</span>{% endif %}
+                {% if p.is_discovery is defined and p.is_discovery %}<span class="badge badge--discover">DISCOVER</span>{% endif %}
+                <span>{{ p.published[:10] if p.published else "" }}</span>
+            </div>
+        </div>
+        <div style="display:flex; align-items:center; gap:0.35rem">
+            {% if p.preference_boost is defined and p.preference_boost > 0.1 %}
+            <span class="boost-pip boost-pip--up" title="Preference boost: {{ '%+.1f'|format(p.preference_boost) }}">&#9650;</span>
+            {% elif p.preference_boost is defined and p.preference_boost < -0.1 %}
+            <span class="boost-pip boost-pip--down" title="Preference penalty: {{ '%+.1f'|format(p.preference_boost) }}">&#9660;</span>
+            {% endif %}
+            <div class="score-badge {{ level }}">{{ p.composite }}</div>
+        </div>
+    </div>
+    {% if p.summary %}<div class="summary-text">{{ p.summary[:200] }}{% if p.summary | length > 200 %}&hellip;{% endif %}</div>{% endif %}
+    <div class="score-mini-track">
+        <div class="score-mini-fill {{ level }}" style="width:{{ pct }}%"></div>
+    </div>
+</div>

src/web/templates/partials/paper_row.html

@@ -0,0 +1,41 @@
+{% set pct = ((p.composite or 0) | float / 10 * 100) | round(0) | int %}
+{% set level = 'high' if pct >= 65 else ('mid' if pct >= 40 else 'low') %}
+<tr>
+    <td class="col-rank">{{ rank if rank is defined else "" }}</td>
+    <td>
+        <a href="/papers/{{ p.domain }}/{{ p.id }}" class="paper-title-link">
+            {{ p.title[:80] }}{% if p.title | length > 80 %}&hellip;{% endif %}
+        </a>
+        {% if p.topics is iterable and p.topics is not string and p.topics | length > 0 %}
+        <div style="margin-top:2px">
+            {% for t in p.topics[:2] %}
+            <span class="badge badge--accent" style="font-size:0.55rem; padding:0.08rem 0.35rem">{{ t }}</span>
+            {% endfor %}
+        </div>
+        {% endif %}
+        {% if p.is_discovery is defined and p.is_discovery %}
+        <span class="badge badge--discover">DISCOVER</span>
+        {% endif %}
+    </td>
+    <td class="col-score composite score-{{ level }}">
+        {% if p.preference_boost is defined and p.preference_boost != 0 %}
+        <span title="Adjusted: {{ p.adjusted_score }} (raw {{ p.composite }}{{ ' %+.1f'|format(p.preference_boost) }})">
+            {{ p.adjusted_score }}
+            {% if p.preference_boost > 0 %}<span class="boost-arrow boost-up">&#9650;</span>{% elif p.preference_boost < 0 %}<span class="boost-arrow boost-down">&#9660;</span>{% endif %}
+        </span>
+        <div class="score-raw">({{ p.composite }})</div>
+        {% else %}
+        {{ p.composite }}
+        {% endif %}
+    </td>
+    <td class="col-score">{{ p.score_axis_1 | default("&mdash;") }}</td>
+    <td class="col-score">{{ p.score_axis_2 | default("&mdash;") }}</td>
+    <td class="col-score">{{ p.score_axis_3 | default("&mdash;") }}</td>
+    <td class="col-code">{% if p.code_url or p.github_repo %}<span class="code-check">&#10003;</span>{% else %}<span class="code-dash">&mdash;</span>{% endif %}</td>
+    <td class="col-signals">
+        {% set paper_id = p.id %}
+        {% set user_signal = p.user_signal if p.user_signal is defined else None %}
+        {% include "partials/signal_buttons.html" %}
+    </td>
+    <td class="col-summary">{{ (p.summary or "")[:100] }}{% if (p.summary or "") | length > 100 %}&hellip;{% endif %}</td>
+</tr>

src/web/templates/partials/papers_results.html

@@ -0,0 +1,55 @@
+{% if papers %}
+<table class="paper-table">
+    <thead>
+        <tr>
+            <th class="col-rank">#</th>
+            <th>Title</th>
+            <th class="col-score">Score</th>
+            <th class="col-score" title="{{ axis_labels[0] }}">{{ abbreviate_label(axis_labels[0]) }}</th>
+            <th class="col-score" title="{{ axis_labels[1] }}">{{ abbreviate_label(axis_labels[1]) }}</th>
+            <th class="col-score" title="{{ axis_labels[2] }}">{{ abbreviate_label(axis_labels[2]) }}</th>
+            <th class="col-code">Code</th>
+            <th class="col-signals-header">Rate</th>
+            <th class="col-summary">Summary</th>
+        </tr>
+    </thead>
+    <tbody>
+        {% for p in papers %}
+        {% set rank = offset + loop.index %}
+        {% include "partials/paper_row.html" %}
+        {% endfor %}
+    </tbody>
+</table>
+
+{% if not has_preferences is defined or not has_preferences %}
+<div class="cold-start-hint">
+    Rate papers to personalize your feed &mdash; use the arrows to tell the system what you like.
+</div>
+{% endif %}
+
+{% set filter_qs %}{% if search %}&search={{ search | urlencode }}{% endif %}{% if min_score %}&min_score={{ min_score }}{% endif %}{% if has_code %}&has_code=1{% endif %}{% if topic %}&topic={{ topic | urlencode }}{% endif %}{% if sort %}&sort={{ sort | urlencode }}{% endif %}{% endset %}
+
+{% if total > limit %}
+<div class="pagination">
+    {% if offset > 0 %}
+    <a href="/papers/{{ domain }}?offset={{ offset - limit }}&limit={{ limit }}{{ filter_qs }}"
+       hx-get="/papers/{{ domain }}?offset={{ offset - limit }}&limit={{ limit }}{{ filter_qs }}"
+       hx-target="#paper-results" hx-push-url="true" hx-indicator="#page-loader"
+       class="btn btn-sm">&larr; Prev</a>
+    {% endif %}
+    <span class="page-info">{{ offset + 1 }}&ndash;{{ [offset + limit, total] | min }} of {{ total }}</span>
+    {% if offset + limit < total %}
+    <a href="/papers/{{ domain }}?offset={{ offset + limit }}&limit={{ limit }}{{ filter_qs }}"
+       hx-get="/papers/{{ domain }}?offset={{ offset + limit }}&limit={{ limit }}{{ filter_qs }}"
+       hx-target="#paper-results" hx-push-url="true" hx-indicator="#page-loader"
+       class="btn btn-sm">Next &rarr;</a>
+    {% endif %}
+</div>
+{% endif %}
+
+{% else %}
+<div class="empty-state">
+    <h2>No papers found</h2>
+    <p>{% if search or min_score or has_code or topic %}Try adjusting your filters.{% else %}Run the {{ domain_label }} pipeline to get started.{% endif %}</p>
+</div>
+{% endif %}

src/web/templates/partials/signal_buttons.html

@@ -0,0 +1,17 @@
+<div class="signal-buttons" id="signal-{{ paper_id }}">
+    <button class="signal-btn signal-btn--up{% if user_signal == 'upvote' %} active{% endif %}"
+            hx-post="/api/signal/{{ paper_id }}/upvote"
+            hx-target="#signal-{{ paper_id }}"
+            hx-swap="outerHTML"
+            title="More like this">&#9650;</button>
+    <button class="signal-btn signal-btn--down{% if user_signal == 'downvote' %} active{% endif %}"
+            hx-post="/api/signal/{{ paper_id }}/downvote"
+            hx-target="#signal-{{ paper_id }}"
+            hx-swap="outerHTML"
+            title="Less like this">&#9660;</button>
+    <button class="signal-btn signal-btn--dismiss{% if user_signal == 'dismiss' %} active{% endif %}"
+            hx-post="/api/signal/{{ paper_id }}/dismiss"
+            hx-target="#signal-{{ paper_id }}"
+            hx-swap="outerHTML"
+            title="Not interested">&times;</button>
+</div>

src/web/templates/preferences.html

@@ -0,0 +1,85 @@
+{% extends "base.html" %}
+{% block title %}Preferences — Research Intelligence{% endblock %}
+{% block content %}
+<div class="page-header">
+    <div style="display:flex; justify-content:space-between; align-items:flex-start; flex-wrap:wrap; gap:0.5rem">
+        <div>
+            <h1>Preferences</h1>
+            <div class="subtitle">
+                {{ total_prefs }} learned preference{{ 's' if total_prefs != 1 else '' }}
+                {% if updated_at %} · Last updated {{ updated_at[:16] }}{% endif %}
+            </div>
+        </div>
+        <div style="display:flex; gap:0.5rem">
+            <button class="btn btn-sm" onclick="this.disabled=true;this.textContent='Recomputing...';fetch('/api/preferences/recompute',{method:'POST'}).then(function(){showToast('Preferences recomputed','success');setTimeout(function(){location.reload()},500)}).catch(function(){showToast('Failed','error')})">Recompute</button>
+            <button class="btn btn-sm" style="color:var(--red)" onclick="if(confirm('Reset all preferences and signal history?')){this.disabled=true;fetch('/api/preferences/reset',{method:'POST'}).then(function(){showToast('Preferences reset','success');setTimeout(function(){location.reload()},500)}).catch(function(){showToast('Failed','error')})}">Reset All</button>
+        </div>
+    </div>
+</div>
+
+{# Signal summary #}
+<div class="stats-grid" style="grid-template-columns:repeat(5, 1fr); margin-bottom:2rem">
+    <div class="stat-card stat-card--green">
+        <div class="label">Saves</div>
+        <div class="value">{{ signal_counts.get('save', 0) }}</div>
+    </div>
+    <div class="stat-card stat-card--blue">
+        <div class="label">Upvotes</div>
+        <div class="value">{{ signal_counts.get('upvote', 0) }}</div>
+    </div>
+    <div class="stat-card stat-card--purple">
+        <div class="label">Views</div>
+        <div class="value">{{ signal_counts.get('view', 0) }}</div>
+    </div>
+    <div class="stat-card stat-card--red">
+        <div class="label">Downvotes</div>
+        <div class="value">{{ signal_counts.get('downvote', 0) }}</div>
+    </div>
+    <div class="stat-card" style="background:var(--bg-card)">
+        <div class="label">Dismissed</div>
+        <div class="value">{{ signal_counts.get('dismiss', 0) }}</div>
+    </div>
+</div>
+
+{% if total_prefs == 0 %}
+<div class="empty-state">
+    <h2>No preferences yet</h2>
+    <p>Rate papers using the arrow buttons to build your preference profile. The system learns from saves, upvotes, downvotes, and dismissals.</p>
+</div>
+{% else %}
+
+{# Preference groups #}
+{% set pref_labels = {'topic': 'Topics', 'keyword': 'Keywords', 'category': 'Categories', 'author': 'Authors', 'axis_pref': 'Axis Preferences'} %}
+
+<div class="pref-groups">
+{% for prefix, items in grouped.items() %}
+{% set label = pref_labels.get(prefix, prefix | capitalize) %}
+<div class="pref-group">
+    <div class="section-header">
+        <h2>{{ label }}</h2>
+        <span class="badge badge--accent">{{ items | length }}</span>
+    </div>
+
+    <div class="pref-list">
+        {% for item in items[:20] %}
+        <div class="pref-item">
+            <span class="pref-item__name">{{ item.name }}</span>
+            <span class="pref-item__count" title="{{ item.count }} signal{{ 's' if item.count != 1 else '' }}">{{ item.count }}x</span>
+            <div class="pref-bar-container">
+                {% set abs_val = (item.value | abs * 100) | round(0) | int %}
+                {% if item.value > 0 %}
+                <div class="pref-bar pref-bar--positive" style="width:{{ abs_val }}%"></div>
+                {% else %}
+                <div class="pref-bar pref-bar--negative" style="width:{{ abs_val }}%"></div>
+                {% endif %}
+            </div>
+            <span class="pref-item__value {% if item.value > 0 %}pref-positive{% else %}pref-negative{% endif %}">{{ '%+.2f'|format(item.value) }}</span>
+        </div>
+        {% endfor %}
+    </div>
+</div>
+{% endfor %}
+</div>
+
+{% endif %}
+{% endblock %}

src/web/templates/seed_preferences.html

@@ -0,0 +1,178 @@
+{% extends "base.html" %}
+{% block title %}Pick Papers You Like — Research Intelligence{% endblock %}
+{% block content %}
+<div class="page-header">
+    <h1>Pick Papers You Like</h1>
+    <div class="subtitle">Rate a few papers to personalize your feed. Click thumbs up or down, then hit Done.</div>
+</div>
+
+{% if papers %}
+<div class="seed-grid" id="seed-grid">
+    {% for p in papers %}
+    <div class="seed-card" data-arxiv="{{ p.arxiv_id }}">
+        <div class="seed-card__body">
+            <div class="seed-card__domain">
+                {% if p.domain == 'aiml' %}
+                <span class="badge badge--accent">AI/ML</span>
+                {% elif p.domain == 'security' %}
+                <span class="badge badge--red">Security</span>
+                {% endif %}
+            </div>
+            <div class="seed-card__title">{{ p.title }}</div>
+            {% if p.summary %}
+            <div class="seed-card__summary">{{ p.summary[:150] }}{% if p.summary | length > 150 %}&hellip;{% endif %}</div>
+            {% endif %}
+        </div>
+        <div class="seed-card__actions">
+            <button type="button" class="seed-btn seed-btn--up" onclick="seedRate(this, 'upvote')" title="More like this">
+                <svg width="16" height="16" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2.5" stroke-linecap="round" stroke-linejoin="round"><path d="M14 9V5a3 3 0 0 0-3-3l-4 9v11h11.28a2 2 0 0 0 2-1.7l1.38-9a2 2 0 0 0-2-2.3H14z"/><path d="M7 22H4a2 2 0 0 1-2-2v-7a2 2 0 0 1 2-2h3"/></svg>
+            </button>
+            <button type="button" class="seed-btn seed-btn--down" onclick="seedRate(this, 'downvote')" title="Less like this">
+                <svg width="16" height="16" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2.5" stroke-linecap="round" stroke-linejoin="round"><path d="M10 15v4a3 3 0 0 0 3 3l4-9V2H5.72a2 2 0 0 0-2 1.7l-1.38 9a2 2 0 0 0 2 2.3H10z"/><path d="M17 2h3a2 2 0 0 1 2 2v7a2 2 0 0 1-2 2h-3"/></svg>
+            </button>
+        </div>
+    </div>
+    {% endfor %}
+</div>
+
+<div style="text-align:center; margin-top:2rem">
+    <span id="seed-count" style="color:var(--text-muted); font-size:0.85rem; margin-right:1rem">0 rated</span>
+    <button type="button" class="btn btn-primary" id="seed-done" onclick="seedDone()">Done</button>
+</div>
+
+<style>
+.seed-grid {
+    display: grid;
+    grid-template-columns: repeat(auto-fill, minmax(280px, 1fr));
+    gap: 1rem;
+}
+.seed-card {
+    background: var(--bg-card);
+    border: 1px solid var(--border);
+    border-radius: var(--radius-lg);
+    padding: 1rem 1.25rem;
+    display: flex;
+    flex-direction: column;
+    justify-content: space-between;
+    transition: border-color 0.2s, box-shadow 0.2s;
+    animation: fadeSlideUp 0.35s ease-out both;
+}
+.seed-card.rated-up {
+    border-color: rgba(16, 185, 129, 0.4);
+    box-shadow: 0 0 12px rgba(16, 185, 129, 0.1);
+}
+.seed-card.rated-down {
+    border-color: rgba(239, 68, 68, 0.3);
+    opacity: 0.6;
+}
+.seed-card__title {
+    font-weight: 600;
+    font-size: 0.88rem;
+    line-height: 1.45;
+    margin: 0.35rem 0;
+}
+.seed-card__summary {
+    font-size: 0.8rem;
+    color: var(--text-muted);
+    line-height: 1.5;
+    margin-top: 0.25rem;
+}
+.seed-card__actions {
+    display: flex;
+    gap: 0.5rem;
+    margin-top: 0.75rem;
+    padding-top: 0.75rem;
+    border-top: 1px solid var(--border);
+}
+.seed-btn {
+    flex: 1;
+    display: flex;
+    align-items: center;
+    justify-content: center;
+    gap: 0.35rem;
+    padding: 0.45rem;
+    border-radius: var(--radius);
+    border: 1px solid var(--border);
+    background: transparent;
+    color: var(--text-muted);
+    cursor: pointer;
+    font-size: 0.8rem;
+    transition: all 0.15s;
+}
+.seed-btn:hover {
+    background: var(--bg-surface);
+    color: var(--text-secondary);
+}
+.seed-btn--up.active {
+    background: var(--emerald-glow);
+    color: var(--emerald);
+    border-color: rgba(16, 185, 129, 0.3);
+}
+.seed-btn--down.active {
+    background: var(--red-glow);
+    color: var(--red);
+    border-color: rgba(239, 68, 68, 0.3);
+}
+</style>
+
+<script>
+var seedRatings = {};
+
+function seedRate(btn, action) {
+    var card = btn.closest('.seed-card');
+    var arxivId = card.dataset.arxiv;
+    var upBtn = card.querySelector('.seed-btn--up');
+    var downBtn = card.querySelector('.seed-btn--down');
+
+    // Toggle off if already active
+    if (seedRatings[arxivId] === action) {
+        delete seedRatings[arxivId];
+        upBtn.classList.remove('active');
+        downBtn.classList.remove('active');
+        card.classList.remove('rated-up', 'rated-down');
+    } else {
+        seedRatings[arxivId] = action;
+        upBtn.classList.toggle('active', action === 'upvote');
+        downBtn.classList.toggle('active', action === 'downvote');
+        card.classList.toggle('rated-up', action === 'upvote');
+        card.classList.toggle('rated-down', action === 'downvote');
+    }
+
+    document.getElementById('seed-count').textContent = Object.keys(seedRatings).length + ' rated';
+}
+
+function seedDone() {
+    var count = Object.keys(seedRatings).length;
+    if (count === 0) {
+        window.location.href = '/';
+        return;
+    }
+
+    var btn = document.getElementById('seed-done');
+    btn.disabled = true;
+    btn.textContent = 'Saving...';
+
+    fetch('/api/seed-preferences', {
+        method: 'POST',
+        headers: {'Content-Type': 'application/json'},
+        body: JSON.stringify({ratings: seedRatings})
+    })
+    .then(function(r) { return r.json(); })
+    .then(function(data) {
+        window.location.href = '/?toast=Preferences+initialized+from+' + data.count + '+ratings';
+    })
+    .catch(function() {
+        btn.disabled = false;
+        btn.textContent = 'Done';
+        alert('Error saving preferences');
+    });
+}
+</script>
+{% else %}
+<div class="empty-state">
+    <h2>No seed papers available</h2>
+    <p>Run a pipeline first to populate papers, then come back to seed your preferences.</p>
+    <a href="/" class="btn btn-primary" style="margin-top:1rem">Go to Dashboard</a>
+</div>
+{% endif %}
+{% endblock %}

src/web/templates/setup.html

@@ -0,0 +1,596 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+    <meta charset="utf-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1">
+    <title>Setup — Research Intelligence</title>
+    <meta name="theme-color" content="#0b1121">
+    <link rel="icon" href="/static/favicon.svg" type="image/svg+xml">
+    <link rel="stylesheet" href="/static/style.css">
+    <style>
+    .setup-wrap {
+        min-height: 100vh;
+        display: flex;
+        align-items: center;
+        justify-content: center;
+        padding: 2rem 1rem;
+    }
+    .setup-card {
+        background: var(--bg-card);
+        border: 1px solid var(--border);
+        border-radius: var(--radius-xl);
+        max-width: 640px;
+        width: 100%;
+        padding: 2.5rem;
+        box-shadow: var(--shadow-lg);
+        animation: fadeSlideUp 0.5s ease-out both;
+    }
+    .setup-logo {
+        display: flex;
+        align-items: center;
+        gap: 10px;
+        margin-bottom: 2rem;
+    }
+    .setup-logo .logo-dot {
+        width: 10px;
+        height: 10px;
+    }
+    .setup-logo span {
+        font-family: var(--font-display);
+        font-size: 1.3rem;
+        font-weight: 700;
+        letter-spacing: -0.02em;
+    }
+    .setup-step {
+        display: none;
+    }
+    .setup-step.active {
+        display: block;
+        animation: fadeSlideUp 0.35s ease-out both;
+    }
+    .setup-step h2 {
+        font-family: var(--font-display);
+        font-size: 1.35rem;
+        font-weight: 700;
+        letter-spacing: -0.03em;
+        margin-bottom: 0.5rem;
+    }
+    .setup-step .step-desc {
+        color: var(--text-muted);
+        font-size: 0.9rem;
+        margin-bottom: 1.5rem;
+        line-height: 1.6;
+    }
+    .setup-field {
+        margin-bottom: 1.25rem;
+    }
+    .setup-field label {
+        display: block;
+        font-size: 0.8rem;
+        font-weight: 600;
+        color: var(--text-secondary);
+        margin-bottom: 0.4rem;
+        text-transform: uppercase;
+        letter-spacing: 0.04em;
+    }
+    .setup-field input[type="text"],
+    .setup-field input[type="password"],
+    .setup-field select {
+        width: 100%;
+        background: var(--bg);
+        border: 1px solid var(--border-strong);
+        border-radius: var(--radius);
+        color: var(--text);
+        padding: 0.6rem 0.85rem;
+        font-size: 0.9rem;
+        font-family: var(--font-body);
+        transition: border-color 0.15s, box-shadow 0.15s;
+    }
+    .setup-field input:focus,
+    .setup-field select:focus {
+        outline: none;
+        border-color: var(--accent);
+        box-shadow: 0 0 0 2px var(--accent-muted);
+    }
+    .setup-field .hint {
+        font-size: 0.78rem;
+        color: var(--text-dim);
+        margin-top: 0.3rem;
+    }
+    .setup-toggle {
+        display: flex;
+        align-items: center;
+        justify-content: space-between;
+        padding: 0.85rem 1rem;
+        background: var(--bg);
+        border: 1px solid var(--border);
+        border-radius: var(--radius);
+        margin-bottom: 0.6rem;
+        cursor: pointer;
+        transition: border-color 0.15s;
+    }
+    .setup-toggle:hover {
+        border-color: var(--border-strong);
+    }
+    .setup-toggle .toggle-label {
+        font-weight: 600;
+        font-size: 0.9rem;
+    }
+    .setup-toggle .toggle-desc {
+        font-size: 0.78rem;
+        color: var(--text-muted);
+        margin-top: 0.15rem;
+    }
+    .toggle-switch {
+        position: relative;
+        width: 42px;
+        height: 24px;
+        flex-shrink: 0;
+    }
+    .toggle-switch input {
+        opacity: 0;
+        width: 0;
+        height: 0;
+    }
+    .toggle-switch .slider {
+        position: absolute;
+        inset: 0;
+        background: var(--bg-surface);
+        border-radius: 12px;
+        transition: background 0.2s;
+        cursor: pointer;
+    }
+    .toggle-switch .slider::after {
+        content: '';
+        position: absolute;
+        width: 18px;
+        height: 18px;
+        left: 3px;
+        top: 3px;
+        background: var(--text-muted);
+        border-radius: 50%;
+        transition: transform 0.2s, background 0.2s;
+    }
+    .toggle-switch input:checked + .slider {
+        background: var(--accent);
+    }
+    .toggle-switch input:checked + .slider::after {
+        transform: translateX(18px);
+        background: var(--bg-deep);
+    }
+    .setup-axes {
+        margin-top: 0.75rem;
+        padding: 0.85rem;
+        background: var(--bg);
+        border-radius: var(--radius);
+        border: 1px solid var(--border);
+    }
+    .setup-axes.hidden {
+        display: none;
+    }
+    .axis-row {
+        display: flex;
+        gap: 0.75rem;
+        align-items: center;
+        margin-bottom: 0.5rem;
+    }
+    .axis-row:last-child {
+        margin-bottom: 0;
+    }
+    .axis-row .axis-name {
+        flex: 1;
+        font-size: 0.84rem;
+        font-weight: 500;
+    }
+    .axis-row input[type="number"] {
+        width: 5rem;
+        background: var(--bg-card);
+        border: 1px solid var(--border);
+        border-radius: var(--radius);
+        color: var(--text);
+        padding: 0.35rem 0.5rem;
+        font-size: 0.84rem;
+        font-family: var(--font-mono);
+        text-align: center;
+    }
+    .axis-row input[type="number"]:focus {
+        outline: none;
+        border-color: var(--accent);
+    }
+    .setup-nav {
+        display: flex;
+        justify-content: space-between;
+        align-items: center;
+        margin-top: 2rem;
+        padding-top: 1.25rem;
+        border-top: 1px solid var(--border);
+    }
+    .step-dots {
+        display: flex;
+        gap: 6px;
+    }
+    .step-dot {
+        width: 8px;
+        height: 8px;
+        border-radius: 50%;
+        background: var(--bg-surface);
+        transition: background 0.2s, box-shadow 0.2s;
+    }
+    .step-dot.active {
+        background: var(--accent);
+        box-shadow: 0 0 6px var(--accent);
+    }
+    .step-dot.done {
+        background: var(--emerald);
+    }
+    .api-key-status {
+        display: inline-flex;
+        align-items: center;
+        gap: 0.35rem;
+        font-size: 0.82rem;
+        font-weight: 500;
+        padding: 0.3rem 0.7rem;
+        border-radius: var(--radius-full);
+        margin-top: 0.5rem;
+    }
+    .api-key-status.valid {
+        background: var(--emerald-glow);
+        color: var(--emerald);
+    }
+    .api-key-status.invalid {
+        background: var(--red-glow);
+        color: var(--red);
+    }
+    .api-key-status.checking {
+        background: var(--amber-glow);
+        color: var(--amber);
+    }
+    .setup-summary {
+        background: var(--bg);
+        border: 1px solid var(--border);
+        border-radius: var(--radius);
+        padding: 1rem;
+    }
+    .summary-item {
+        display: flex;
+        justify-content: space-between;
+        padding: 0.4rem 0;
+        font-size: 0.85rem;
+    }
+    .summary-item:not(:last-child) {
+        border-bottom: 1px solid var(--border);
+    }
+    .summary-item .label {
+        color: var(--text-muted);
+    }
+    .summary-item .value {
+        font-weight: 600;
+        color: var(--text);
+    }
+    </style>
+</head>
+<body>
+    <div class="setup-wrap">
+        <div class="setup-card">
+            <div class="setup-logo">
+                <span class="logo-dot"></span>
+                <span>Research Intelligence</span>
+            </div>
+
+            <!-- Step 1: Welcome -->
+            <div class="setup-step active" data-step="1">
+                <h2>Welcome</h2>
+                <p class="step-desc">
+                    Research Intelligence monitors academic papers and GitHub projects,
+                    scores them with AI, and learns your preferences over time.
+                    This setup will configure your instance.
+                </p>
+                <div style="padding:1rem; background:var(--bg); border-radius:var(--radius); border-left:3px solid var(--accent); font-size:0.85rem; color:var(--text-secondary); line-height:1.65">
+                    You'll configure:<br>
+                    &bull; API key for AI scoring<br>
+                    &bull; Which research domains to monitor<br>
+                    &bull; GitHub project tracking<br>
+                    &bull; Pipeline schedule
+                </div>
+            </div>
+
+            <!-- Step 2: API Key -->
+            <div class="setup-step" data-step="2">
+                <h2>API Key</h2>
+                <p class="step-desc">
+                    An Anthropic API key is required for paper scoring.
+                    It will be stored in a local <code>.env</code> file, not in the config.
+                </p>
+                <div class="setup-field">
+                    <label for="api-key">Anthropic API Key</label>
+                    <input type="password" id="api-key" name="api_key" placeholder="sk-ant-..." autocomplete="off">
+                    <div class="hint">Get one at <a href="https://console.anthropic.com/" target="_blank">console.anthropic.com</a></div>
+                </div>
+                <div id="api-key-result"></div>
+                <button type="button" class="btn btn-sm" onclick="validateApiKey()" id="validate-btn">Validate Key</button>
+            </div>
+
+            <!-- Step 3: Domains -->
+            <div class="setup-step" data-step="3">
+                <h2>Research Domains</h2>
+                <p class="step-desc">Choose which research areas to monitor.</p>
+
+                <div class="setup-toggle" onclick="this.querySelector('input').click()">
+                    <div>
+                        <div class="toggle-label">AI / ML</div>
+                        <div class="toggle-desc">Papers from arXiv + HuggingFace trending</div>
+                    </div>
+                    <label class="toggle-switch" onclick="event.stopPropagation()">
+                        <input type="checkbox" id="domain-aiml" checked onchange="toggleAxes('aiml', this.checked)">
+                        <span class="slider"></span>
+                    </label>
+                </div>
+                <div class="setup-axes" id="axes-aiml">
+                    <div style="font-size:0.72rem; font-weight:600; text-transform:uppercase; letter-spacing:0.04em; color:var(--text-muted); margin-bottom:0.5rem">Scoring Weights</div>
+                    <div class="axis-row">
+                        <span class="axis-name">Code & Weights</span>
+                        <input type="number" id="aiml-w1" value="30" min="0" max="100" step="5">
+                        <span style="font-size:0.78rem; color:var(--text-dim)">%</span>
+                    </div>
+                    <div class="axis-row">
+                        <span class="axis-name">Novelty</span>
+                        <input type="number" id="aiml-w2" value="35" min="0" max="100" step="5">
+                        <span style="font-size:0.78rem; color:var(--text-dim)">%</span>
+                    </div>
+                    <div class="axis-row">
+                        <span class="axis-name">Practical Applicability</span>
+                        <input type="number" id="aiml-w3" value="35" min="0" max="100" step="5">
+                        <span style="font-size:0.78rem; color:var(--text-dim)">%</span>
+                    </div>
+                </div>
+
+                <div class="setup-toggle" onclick="this.querySelector('input').click()" style="margin-top:0.75rem">
+                    <div>
+                        <div class="toggle-label">Security</div>
+                        <div class="toggle-desc">Security research from arXiv cs.CR</div>
+                    </div>
+                    <label class="toggle-switch" onclick="event.stopPropagation()">
+                        <input type="checkbox" id="domain-security" checked onchange="toggleAxes('security', this.checked)">
+                        <span class="slider"></span>
+                    </label>
+                </div>
+                <div class="setup-axes" id="axes-security">
+                    <div style="font-size:0.72rem; font-weight:600; text-transform:uppercase; letter-spacing:0.04em; color:var(--text-muted); margin-bottom:0.5rem">Scoring Weights</div>
+                    <div class="axis-row">
+                        <span class="axis-name">Has Code / PoC</span>
+                        <input type="number" id="sec-w1" value="25" min="0" max="100" step="5">
+                        <span style="font-size:0.78rem; color:var(--text-dim)">%</span>
+                    </div>
+                    <div class="axis-row">
+                        <span class="axis-name">Novel Attack Surface</span>
+                        <input type="number" id="sec-w2" value="40" min="0" max="100" step="5">
+                        <span style="font-size:0.78rem; color:var(--text-dim)">%</span>
+                    </div>
+                    <div class="axis-row">
+                        <span class="axis-name">Real-World Impact</span>
+                        <input type="number" id="sec-w3" value="35" min="0" max="100" step="5">
+                        <span style="font-size:0.78rem; color:var(--text-dim)">%</span>
+                    </div>
+                </div>
+            </div>
+
+            <!-- Step 4: GitHub -->
+            <div class="setup-step" data-step="4">
+                <h2>GitHub Monitoring</h2>
+                <p class="step-desc">Track trending open-source projects via OSSInsight collections.</p>
+                <div class="setup-toggle" onclick="this.querySelector('input').click()">
+                    <div>
+                        <div class="toggle-label">Enable GitHub tracking</div>
+                        <div class="toggle-desc">Monitor trending repos in AI/ML and Security</div>
+                    </div>
+                    <label class="toggle-switch" onclick="event.stopPropagation()">
+                        <input type="checkbox" id="github-enabled" checked>
+                        <span class="slider"></span>
+                    </label>
+                </div>
+            </div>
+
+            <!-- Step 5: Schedule -->
+            <div class="setup-step" data-step="5">
+                <h2>Schedule</h2>
+                <p class="step-desc">How often should pipelines run automatically?</p>
+                <div class="setup-field">
+                    <label for="schedule">Frequency</label>
+                    <select id="schedule">
+                        <option value="weekly" selected>Weekly (Sunday night)</option>
+                        <option value="daily">Daily (midnight UTC)</option>
+                        <option value="manual">Manual only</option>
+                    </select>
+                    <div class="hint">You can always trigger runs manually from the dashboard.</div>
+                </div>
+            </div>
+
+            <!-- Step 6: Review -->
+            <div class="setup-step" data-step="6">
+                <h2>Review & Save</h2>
+                <p class="step-desc">Here's your configuration. Click Save to get started.</p>
+                <div class="setup-summary" id="setup-summary"></div>
+            </div>
+
+            <div class="setup-nav">
+                <div class="step-dots" id="step-dots"></div>
+                <div style="display:flex; gap:0.5rem">
+                    <button type="button" class="btn btn-sm" id="btn-prev" onclick="prevStep()" style="display:none">Back</button>
+                    <button type="button" class="btn btn-primary btn-sm" id="btn-next" onclick="nextStep()">Get Started</button>
+                </div>
+            </div>
+        </div>
+    </div>
+
+    <script>
+    var currentStep = 1;
+    var totalSteps = 6;
+    var apiKeyValid = false;
+
+    function initDots() {
+        var dots = document.getElementById('step-dots');
+        dots.innerHTML = '';
+        for (var i = 1; i <= totalSteps; i++) {
+            var dot = document.createElement('div');
+            dot.className = 'step-dot' + (i === 1 ? ' active' : '');
+            dot.dataset.step = i;
+            dots.appendChild(dot);
+        }
+    }
+    initDots();
+
+    function showStep(n) {
+        var steps = document.querySelectorAll('.setup-step');
+        steps.forEach(function(s) { s.classList.remove('active'); });
+        var target = document.querySelector('.setup-step[data-step="' + n + '"]');
+        if (target) target.classList.add('active');
+
+        var dots = document.querySelectorAll('.step-dot');
+        dots.forEach(function(d) {
+            var s = parseInt(d.dataset.step);
+            d.className = 'step-dot' + (s === n ? ' active' : (s < n ? ' done' : ''));
+        });
+
+        document.getElementById('btn-prev').style.display = n > 1 ? '' : 'none';
+        var nextBtn = document.getElementById('btn-next');
+        if (n === totalSteps) {
+            nextBtn.textContent = 'Save & Start';
+            buildSummary();
+        } else if (n === 1) {
+            nextBtn.textContent = 'Get Started';
+        } else {
+            nextBtn.textContent = 'Next';
+        }
+    }
+
+    function nextStep() {
+        if (currentStep === totalSteps) {
+            saveConfig();
+            return;
+        }
+        currentStep++;
+        showStep(currentStep);
+    }
+
+    function prevStep() {
+        if (currentStep > 1) {
+            currentStep--;
+            showStep(currentStep);
+        }
+    }
+
+    function toggleAxes(domain, enabled) {
+        var el = document.getElementById('axes-' + domain);
+        if (enabled) {
+            el.classList.remove('hidden');
+        } else {
+            el.classList.add('hidden');
+        }
+    }
+
+    function validateApiKey() {
+        var key = document.getElementById('api-key').value.trim();
+        if (!key) return;
+        var result = document.getElementById('api-key-result');
+        var btn = document.getElementById('validate-btn');
+        result.innerHTML = '<span class="api-key-status checking">Checking...</span>';
+        btn.disabled = true;
+
+        fetch('/api/setup/validate-key', {
+            method: 'POST',
+            headers: {'Content-Type': 'application/json'},
+            body: JSON.stringify({api_key: key})
+        })
+        .then(function(r) { return r.json(); })
+        .then(function(data) {
+            if (data.valid) {
+                result.innerHTML = '<span class="api-key-status valid">Valid</span>';
+                apiKeyValid = true;
+            } else {
+                result.innerHTML = '<span class="api-key-status invalid">Invalid — ' + (data.error || 'check your key') + '</span>';
+                apiKeyValid = false;
+            }
+            btn.disabled = false;
+        })
+        .catch(function() {
+            result.innerHTML = '<span class="api-key-status invalid">Connection error</span>';
+            btn.disabled = false;
+        });
+    }
+
+    function getScheduleCron() {
+        var v = document.getElementById('schedule').value;
+        if (v === 'daily') return '0 0 * * *';
+        if (v === 'manual') return '';
+        return '0 22 * * 0';
+    }
+
+    function buildSummary() {
+        var aiml = document.getElementById('domain-aiml').checked;
+        var sec = document.getElementById('domain-security').checked;
+        var gh = document.getElementById('github-enabled').checked;
+        var sched = document.getElementById('schedule').value;
+        var hasKey = document.getElementById('api-key').value.trim().length > 0;
+
+        var html = '';
+        html += '<div class="summary-item"><span class="label">API Key</span><span class="value">' + (hasKey ? (apiKeyValid ? 'Validated' : 'Set (unvalidated)') : 'Not set') + '</span></div>';
+        html += '<div class="summary-item"><span class="label">AI/ML</span><span class="value">' + (aiml ? 'Enabled' : 'Disabled') + '</span></div>';
+        html += '<div class="summary-item"><span class="label">Security</span><span class="value">' + (sec ? 'Enabled' : 'Disabled') + '</span></div>';
+        html += '<div class="summary-item"><span class="label">GitHub</span><span class="value">' + (gh ? 'Enabled' : 'Disabled') + '</span></div>';
+        html += '<div class="summary-item"><span class="label">Schedule</span><span class="value">' + sched.charAt(0).toUpperCase() + sched.slice(1) + '</span></div>';
+        document.getElementById('setup-summary').innerHTML = html;
+    }
+
+    function saveConfig() {
+        var btn = document.getElementById('btn-next');
+        btn.disabled = true;
+        btn.textContent = 'Saving...';
+
+        var payload = {
+            api_key: document.getElementById('api-key').value.trim(),
+            domains: {
+                aiml: {
+                    enabled: document.getElementById('domain-aiml').checked,
+                    scoring_weights: [
+                        parseInt(document.getElementById('aiml-w1').value) / 100,
+                        parseInt(document.getElementById('aiml-w2').value) / 100,
+                        parseInt(document.getElementById('aiml-w3').value) / 100
+                    ]
+                },
+                security: {
+                    enabled: document.getElementById('domain-security').checked,
+                    scoring_weights: [
+                        parseInt(document.getElementById('sec-w1').value) / 100,
+                        parseInt(document.getElementById('sec-w2').value) / 100,
+                        parseInt(document.getElementById('sec-w3').value) / 100
+                    ]
+                }
+            },
+            github: {enabled: document.getElementById('github-enabled').checked},
+            schedule: getScheduleCron()
+        };
+
+        fetch('/api/setup/save', {
+            method: 'POST',
+            headers: {'Content-Type': 'application/json'},
+            body: JSON.stringify(payload)
+        })
+        .then(function(r) { return r.json(); })
+        .then(function(data) {
+            if (data.status === 'ok') {
+                window.location.href = '/';
+            } else {
+                btn.disabled = false;
+                btn.textContent = 'Save & Start';
+                alert('Error: ' + (data.error || 'Unknown error'));
+            }
+        })
+        .catch(function() {
+            btn.disabled = false;
+            btn.textContent = 'Save & Start';
+            alert('Connection error');
+        });
+    }
+    </script>
+</body>
+</html>

src/web/templates/weeks.html

@@ -0,0 +1,83 @@
+{% extends "base.html" %}
+{% block title %}Archive — Research Intelligence{% endblock %}
+{% block content %}
+<div class="page-header">
+    <h1>Weekly Archives</h1>
+    <div class="subtitle">Past weekly reports and pipeline runs</div>
+</div>
+
+{% if archives %}
+<div class="section-header">
+    <h2>Reports</h2>
+</div>
+<table class="paper-table" style="margin-bottom:2.5rem">
+    <thead>
+        <tr>
+            <th>Week</th>
+            <th>Domain</th>
+            <th>File</th>
+        </tr>
+    </thead>
+    <tbody>
+        {% for a in archives %}
+        <tr>
+            <td style="font-family:var(--font-mono); font-size:0.82rem">{{ a.date }}</td>
+            <td>
+                {% if a.domain == 'aiml' %}
+                <span class="badge badge--accent">AI/ML</span>
+                {% elif a.domain == 'security' %}
+                <span class="badge badge--red">SECURITY</span>
+                {% else %}
+                <span class="badge" style="background:var(--bg-surface); color:var(--text-muted)">{{ a.domain | upper }}</span>
+                {% endif %}
+            </td>
+            <td><a href="/weeks/{{ a.filename }}">{{ a.filename }}</a></td>
+        </tr>
+        {% endfor %}
+    </tbody>
+</table>
+{% else %}
+<div class="empty-state" style="padding:2rem">
+    <h2>No archives yet</h2>
+    <p>Weekly reports will appear here after pipeline runs.</p>
+</div>
+{% endif %}
+
+{% if runs %}
+<div class="section-header">
+    <h2>Recent Runs</h2>
+</div>
+<table class="paper-table">
+    <thead>
+        <tr>
+            <th>ID</th>
+            <th>Domain</th>
+            <th>Date Range</th>
+            <th>Papers</th>
+            <th>Status</th>
+            <th>Started</th>
+        </tr>
+    </thead>
+    <tbody>
+        {% for r in runs %}
+        <tr>
+            <td style="font-family:var(--font-mono); font-size:0.82rem">{{ r.id }}</td>
+            <td>
+                {% if r.domain == 'aiml' %}
+                <span class="badge badge--accent">AI/ML</span>
+                {% elif r.domain == 'security' %}
+                <span class="badge badge--red">SECURITY</span>
+                {% else %}
+                <span class="badge" style="background:var(--bg-surface); color:var(--text-muted)">{{ r.domain | upper }}</span>
+                {% endif %}
+            </td>
+            <td style="font-size:0.82rem">{{ r.date_start }} to {{ r.date_end }}</td>
+            <td style="font-family:var(--font-mono)">{{ r.paper_count }}</td>
+            <td class="status-{{ r.status }}">{{ r.status }}</td>
+            <td style="font-size:0.82rem; color:var(--text-muted)">{{ r.started_at[:16] }}</td>
+        </tr>
+        {% endfor %}
+    </tbody>
+</table>
+{% endif %}
+{% endblock %}