app.py

"""
PhD Research OS — Taxonomy Manager GUI
========================================
A Gradio companion application for managing domain-specific study type
taxonomies that integrate with the Research OS confidence scoring engine.

Features:
  - View and edit the 8-tier Quantum-Bio base taxonomy
  - Create, edit, delete domain-specific taxonomies
  - Add custom study types per domain with calibrated weights
  - Live confidence scoring calculator
  - Migration status and rollback controls
  - Full audit trail
"""

import os
import sys
import json
import gradio as gr

# Add parent to path for imports
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))

from phd_research_os.taxonomy import (
    TaxonomyManager, STUDY_TYPE_WEIGHTS, STUDY_TYPE_DESCRIPTIONS,
    ALLOWED_STUDY_TYPES, TAXONOMY_VERSION, PIPELINE_VERSION,
    DEFAULT_DOMAINS, LEGACY_TO_V2_MAP
)
from phd_research_os.db import init_db, from_fixed

# Initialize
DB_PATH = os.environ.get("RESEARCH_OS_DB", "data/research_os.db")
os.makedirs("data", exist_ok=True)
tm = TaxonomyManager(db_path=DB_PATH)


# ============================================================
# Helper Functions
# ============================================================

def get_base_taxonomy_table():
    """Get the base 8-tier taxonomy as a table."""
    rows = []
    for st in ALLOWED_STUDY_TYPES:
        weight = STUDY_TYPE_WEIGHTS[st]
        desc = STUDY_TYPE_DESCRIPTIONS.get(st, "")
        rows.append([st, from_fixed(weight), desc])
    return rows


def get_domain_list():
    """Get list of domain names for dropdown."""
    domains = tm.list_domains()
    return [f"{d['domain_id']} — {d['name']}" for d in domains]


def get_domain_table(domain_selection):
    """Get study types table for a selected domain."""
    if not domain_selection:
        return []
    domain_id = domain_selection.split(" — ")[0]
    all_types = tm.get_all_study_types(domain_id)

    rows = []
    for st, info in sorted(all_types.items(), key=lambda x: -x[1]["weight"]):
        rows.append([
            st,
            info["weight_float"],
            info["description"],
            info["source"],
        ])
    return rows


def score_confidence_ui(evidence_strength, study_type, journal_tier, 
                        is_complete, domain_selection):
    """Calculate confidence score from UI inputs."""
    if not domain_selection:
        domain_id = "quantum_bio"
    else:
        domain_id = domain_selection.split(" — ")[0]

    tier_map = {"Tier 1": 1, "Tier 2": 2, "Tier 3": 3, "Preprint": 0}
    tier = tier_map.get(journal_tier, 2)

    result = tm.score_confidence(
        evidence_strength=evidence_strength,
        study_type=study_type,
        journal_tier=tier,
        is_complete=is_complete,
        domain_id=domain_id,
    )

    breakdown = f"""## Confidence Score: **{result['confidence']:.3f}**

### Breakdown
| Factor | Value |
|--------|-------|
| Evidence Strength | {result['evidence_strength']:.3f} |
| Study Quality Weight | {result['study_quality_weight']:.3f} |
| Journal Tier Weight | {result['journal_tier_weight']:.3f} |
| Completeness Penalty | {result['completeness_penalty']:.3f} |

### Formula
`{result['evidence_strength']:.3f} × {result['study_quality_weight']:.3f} × {result['journal_tier_weight']:.3f} × {result['completeness_penalty']:.3f} = {result['confidence']:.3f}`

### Metadata
- **Study Type (normalized):** `{result['study_type_normalized']}`
- **Taxonomy Version:** `{result['taxonomy_version']}`
- **Domain:** `{result['domain_id']}`
"""
    return breakdown


def create_domain_ui(domain_id, name, description):
    """Create a new domain taxonomy."""
    if not domain_id or not name:
        return "❌ Domain ID and Name are required", get_domain_list()
    
    domain_id = domain_id.strip().lower().replace(" ", "_").replace("-", "_")
    
    try:
        tm.create_domain(domain_id, name.strip(), description.strip())
        return f"✅ Domain '{name}' created successfully!", get_domain_list()
    except Exception as e:
        return f"❌ Error: {str(e)}", get_domain_list()


def add_study_type_ui(domain_selection, type_name, weight, description):
    """Add a custom study type to a domain."""
    if not domain_selection or not type_name:
        return "❌ Select a domain and provide a type name"
    
    domain_id = domain_selection.split(" — ")[0]
    type_name = type_name.strip().lower().replace(" ", "_").replace("-", "_")
    
    try:
        tm.add_study_type(domain_id, type_name, weight, description.strip())
        return f"✅ Study type '{type_name}' added to {domain_id} (weight={weight})"
    except Exception as e:
        return f"❌ Error: {str(e)}"


def remove_study_type_ui(domain_selection, type_name):
    """Remove a custom study type from a domain."""
    if not domain_selection or not type_name:
        return "❌ Select a domain and provide a type name"
    
    domain_id = domain_selection.split(" — ")[0]
    type_name = type_name.strip().lower().replace(" ", "_")
    
    if tm.remove_study_type(domain_id, type_name):
        return f"✅ Removed '{type_name}' from {domain_id}"
    return f"❌ Type '{type_name}' not found in domain {domain_id}"


def delete_domain_ui(domain_selection):
    """Delete (deactivate) a domain."""
    if not domain_selection:
        return "❌ Select a domain", get_domain_list()
    
    domain_id = domain_selection.split(" — ")[0]
    if tm.delete_domain(domain_id):
        return f"✅ Domain '{domain_id}' deactivated", get_domain_list()
    return f"❌ Cannot delete base taxonomy 'quantum_bio'", get_domain_list()


def run_migration_ui():
    """Run V2 migration."""
    result = tm.migrate_to_v2()
    if result.get("already_migrated"):
        return "ℹ️ Database already migrated to V2"
    if result.get("errors"):
        return f"❌ Migration errors: {result['errors']}"
    return f"✅ Migration complete! {result['rows_backfilled']} claims backfilled to V2"


def run_rollback_ui():
    """Run V1 rollback."""
    result = tm.rollback_to_v1()
    if result.get("errors"):
        return f"❌ Rollback errors: {result['errors']}"
    return f"✅ Rollback complete! {result['rows_reverted']} claims reverted to V1"


def get_audit_log_ui():
    """Get taxonomy audit log."""
    entries = tm.get_audit_log(limit=20)
    if not entries:
        return "No audit entries yet"
    
    lines = ["| Timestamp | Action | Domain | Details |", "|-----------|--------|--------|---------|"]
    for e in entries:
        ts = e.get("timestamp", "")[:19]
        lines.append(f"| {ts} | {e.get('action', '')} | {e.get('domain_id', '-')} | {e.get('details', '')[:60]} |")
    return "\n".join(lines)


def normalize_type_ui(raw_type):
    """Show how a raw study type would be normalized."""
    normalized = tm.normalize_study_type(raw_type)
    weight = tm.get_weight_float(normalized)
    return f"**`{raw_type}`** → **`{normalized}`** (weight: {weight:.3f})"


# ============================================================
# Gradio UI
# ============================================================

THEME = gr.themes.Base(
    primary_hue="blue",
    secondary_hue="slate",
    neutral_hue="slate",
    font=gr.themes.GoogleFont("Inter"),
).set(
    body_background_fill="*neutral_950",
    body_background_fill_dark="*neutral_950",
    block_background_fill="*neutral_900",
    block_background_fill_dark="*neutral_900",
    block_border_color="*neutral_700",
    input_background_fill="*neutral_800",
    input_background_fill_dark="*neutral_800",
)

CSS = """
.gradio-container { max-width: 1200px !important; }
.dark { background: #0a0a0a; }
"""

with gr.Blocks(theme=THEME, css=CSS, title="PhD Research OS — Taxonomy Manager") as app:

    gr.Markdown("""
    # 🧬 PhD Research OS — Taxonomy Manager
    
    **Quantum-Bio Taxonomy V2** | Manage domain-specific study type taxonomies for confidence scoring
    
    `taxonomy_version: quantum_bio_v1` | `pipeline_version: 2.1.0`
    """)

    with gr.Tabs():
        # ── Tab 1: Base Taxonomy ──────────────────────────
        with gr.Tab("📊 Base Taxonomy (8-Tier)"):
            gr.Markdown("### Quantum-Bio 8-Tier Study Type Weights\nThese are the core weights used across all domains. Fixed-point math (×1000).")
            
            base_table = gr.Dataframe(
                headers=["Study Type", "Weight", "Description"],
                value=get_base_taxonomy_table(),
                interactive=False,
                wrap=True,
            )
            
            gr.Markdown("### Study Type Normalizer\nTest how legacy or alias study types map to V2:")
            with gr.Row():
                raw_input = gr.Textbox(label="Raw Study Type", placeholder="e.g., PrimaryExperimental, clinical_trial, simulation")
                norm_output = gr.Markdown()
            raw_input.change(normalize_type_ui, inputs=raw_input, outputs=norm_output)

        # ── Tab 2: Domain Taxonomies ─────────────────────
        with gr.Tab("🌐 Domain Taxonomies"):
            gr.Markdown("### Manage Domain-Specific Taxonomies\nEach domain extends the base 8-tier system with custom study types.")
            
            domain_dropdown = gr.Dropdown(
                label="Select Domain",
                choices=get_domain_list(),
                interactive=True,
            )
            
            domain_table = gr.Dataframe(
                headers=["Study Type", "Weight", "Description", "Source"],
                value=[],
                interactive=False,
                wrap=True,
            )
            
            domain_dropdown.change(get_domain_table, inputs=domain_dropdown, outputs=domain_table)
            
            gr.Markdown("---\n### Create New Domain")
            with gr.Row():
                new_domain_id = gr.Textbox(label="Domain ID", placeholder="e.g., organic_chemistry")
                new_domain_name = gr.Textbox(label="Domain Name", placeholder="e.g., Organic Chemistry")
            new_domain_desc = gr.Textbox(label="Description", placeholder="Taxonomy for organic synthesis and reaction mechanisms")
            create_btn = gr.Button("Create Domain", variant="primary")
            create_status = gr.Markdown()
            
            create_btn.click(
                create_domain_ui,
                inputs=[new_domain_id, new_domain_name, new_domain_desc],
                outputs=[create_status, domain_dropdown],
            )
            
            gr.Markdown("---\n### Add Custom Study Type")
            with gr.Row():
                add_type_name = gr.Textbox(label="Type Name", placeholder="e.g., single_crystal_xrd")
                add_type_weight = gr.Slider(0.0, 1.0, value=0.8, step=0.05, label="Weight")
            add_type_desc = gr.Textbox(label="Description", placeholder="e.g., Single-crystal X-ray diffraction structure determination")
            add_type_btn = gr.Button("Add Study Type", variant="primary")
            add_type_status = gr.Markdown()
            
            add_type_btn.click(
                add_study_type_ui,
                inputs=[domain_dropdown, add_type_name, add_type_weight, add_type_desc],
                outputs=add_type_status,
            )
            
            gr.Markdown("---\n### Remove Study Type / Delete Domain")
            with gr.Row():
                rm_type_name = gr.Textbox(label="Type to Remove", placeholder="e.g., single_crystal_xrd")
                rm_btn = gr.Button("Remove Type", variant="secondary")
            rm_status = gr.Markdown()
            rm_btn.click(remove_study_type_ui, inputs=[domain_dropdown, rm_type_name], outputs=rm_status)
            
            del_btn = gr.Button("🗑️ Deactivate Selected Domain", variant="stop")
            del_status = gr.Markdown()
            del_btn.click(delete_domain_ui, inputs=domain_dropdown, outputs=[del_status, domain_dropdown])

        # ── Tab 3: Confidence Calculator ─────────────────
        with gr.Tab("🔬 Confidence Calculator"):
            gr.Markdown("### Live Confidence Scoring\nCalculate confidence using the V2 formula with domain-aware weights.")
            
            with gr.Row():
                calc_evidence = gr.Slider(0.0, 1.0, value=0.85, step=0.01, label="Evidence Strength")
                calc_type = gr.Dropdown(
                    label="Study Type",
                    choices=ALLOWED_STUDY_TYPES + ["simulation", "primary_experimental", "clinical_trial"],
                    value="direct_physical_measurement",
                )
            with gr.Row():
                calc_tier = gr.Dropdown(
                    label="Journal Tier",
                    choices=["Tier 1", "Tier 2", "Tier 3", "Preprint"],
                    value="Tier 1",
                )
                calc_complete = gr.Checkbox(label="Claim is Complete (all fields present)", value=True)
            
            calc_domain = gr.Dropdown(
                label="Domain",
                choices=get_domain_list(),
                value=get_domain_list()[0] if get_domain_list() else None,
            )
            
            calc_btn = gr.Button("Calculate Confidence", variant="primary")
            calc_result = gr.Markdown()
            
            calc_btn.click(
                score_confidence_ui,
                inputs=[calc_evidence, calc_type, calc_tier, calc_complete, calc_domain],
                outputs=calc_result,
            )

        # ── Tab 4: Migration & Rollback ──────────────────
        with gr.Tab("⚙️ Migration & Rollback"):
            gr.Markdown(f"""### Database Migration Controls
            
**Current Taxonomy Version:** `{TAXONOMY_VERSION}`  
**Current Pipeline Version:** `{PIPELINE_VERSION}`

⚠️ **Migration** converts legacy 4-tier study types to V2 8-tier. **Rollback** reverts to V1.
Both operations are idempotent and logged.
""")
            
            with gr.Row():
                migrate_btn = gr.Button("🔄 Migrate to V2", variant="primary")
                rollback_btn = gr.Button("⏪ Rollback to V1", variant="stop")
            
            migration_status = gr.Markdown()
            
            migrate_btn.click(run_migration_ui, outputs=migration_status)
            rollback_btn.click(run_rollback_ui, outputs=migration_status)
            
            gr.Markdown("---\n### Legacy → V2 Mapping Reference")
            mapping_rows = [[k, v] for k, v in sorted(LEGACY_TO_V2_MAP.items())]
            gr.Dataframe(
                headers=["Legacy/Alias", "Maps To (V2)"],
                value=mapping_rows,
                interactive=False,
            )

        # ── Tab 5: AI Model Council ──────────────────────
        with gr.Tab("🏛️ AI Model Council"):
            gr.Markdown("""### AI Model Council — Multi-Agent Claim Extraction
            
The Council is a 4-member deliberation pipeline that produces higher-quality claim extraction:

| Member | Role | What It Does |
|--------|------|-------------|
| **Query Planner** | Decomposer | Breaks complex questions into 2-4 search queries |
| **Extractor** | Claim Miner | Extracts atomic claims with epistemic tags |
| **Critic** | Challenger | Reviews claims against source, flags errors |
| **Chairman** | Synthesizer | Produces final claims with 0.7 completeness penalty |
""")
            
            with gr.Row():
                council_text = gr.Textbox(
                    label="Scientific Text",
                    placeholder="Paste a paragraph from a scientific paper...",
                    lines=8,
                )
            council_query = gr.Textbox(
                label="Research Question (optional)",
                placeholder="e.g., What is the detection limit for GFET cardiac sensors?",
            )
            council_btn = gr.Button("🏛️ Convene Council", variant="primary")
            
            with gr.Row():
                council_claims = gr.JSON(label="Final Claims (Chairman Output)")
            
            with gr.Accordion("Council Deliberation Details", open=False):
                council_queries = gr.JSON(label="Query Plan (Planner)")
                council_raw = gr.JSON(label="Raw Extraction (Extractor)")
                council_critique = gr.JSON(label="Critique (Critic)")
                council_meta = gr.JSON(label="Round Metadata")
            
            def run_council_ui(text, query):
                if not text or len(text.strip()) < 50:
                    return (
                        [{"error": "Please provide at least 50 characters of scientific text"}],
                        [], [], {}, {}
                    )
                from phd_research_os.council import ModelCouncil
                council = ModelCouncil(brain=None, db_path=DB_PATH)
                result = council.deliberate(text, query=query if query else None)
                return (
                    result.final_claims,
                    result.query_plan,
                    result.raw_extraction,
                    result.critique,
                    result.metadata,
                )
            
            council_btn.click(
                run_council_ui,
                inputs=[council_text, council_query],
                outputs=[council_claims, council_queries, council_raw, council_critique, council_meta],
            )
            
            gr.Markdown("---\n### Council History")
            history_btn = gr.Button("🔄 Load History")
            history_display = gr.Dataframe(
                headers=["Round ID", "Claims", "Tokens", "Started"],
                value=[],
            )
            
            def load_council_history():
                from phd_research_os.council import ModelCouncil
                council = ModelCouncil(brain=None, db_path=DB_PATH)
                history = council.get_council_history(limit=10)
                return [[h["round_id"], h.get("final_claim_count", 0),
                         h.get("total_tokens", 0), h.get("started_at", "")[:19]]
                        for h in history]
            
            history_btn.click(load_council_history, outputs=history_display)

        # ── Tab 6: Audit Log ─────────────────────────────
        with gr.Tab("📋 Audit Log"):
            gr.Markdown("### Taxonomy Audit Trail\nEvery domain change, migration, and rollback is logged.")
            
            refresh_btn = gr.Button("🔄 Refresh Log")
            audit_display = gr.Markdown(value=get_audit_log_ui())
            refresh_btn.click(get_audit_log_ui, outputs=audit_display)


# ============================================================
# Launch
# ============================================================

if __name__ == "__main__":
    app.launch(server_name="0.0.0.0", server_port=7860, share=False)