Delete alpha_factory/infra/rag.py, alpha_factory/infra/rag.py

Upload alpha_factory/config.py

Upload alpha_factory/run.py

Upload alpha_factory/orchestration/pipeline.py

Delete factor_store/alphas.duckdb, prompts/templates/*.j2, .gitattributes

Upload alpha_factory/orchestration/__init__.py

Upload .gitignore

Upload alpha_factory/deterministic/proven_templates.py

Upload alpha_factory/deterministic/expression_mutator.py

Upload tests/test_pipeline.py

Upload alpha_factory/orchestration/pipeline.py

Upload alpha_factory/run.py

Upload alpha_factory/deterministic/lint.py

Upload alpha_factory/deterministic/theme_sampler.py

Upload alpha_factory/cleanup.py

Upload alpha_factory/data/brain_fields.py

Upload alpha_factory/__init__.py

Upload alpha_factory/config.py

fix: honest README with proper metadata"

fix: infra/__init__.py — only export what exists, remove BrainClient from active imports"

fix: orchestration/__init__.py — clean import"

fix: personas/__init__.py — only import what pipeline actually uses, stub the rest"

fix: run.py v3 — add --proven flag as primary mode, fix sign logic in compiler, add retry to llm_client"

fix: pipeline v3 — remove dead imports, add --proven mode as primary, check token budget, honest behavior"

fix: honest README — removes false claims, documents what actually works, adds proper usage instructions"

feat: generate_proven.py — standalone script that generates BRAIN-ready expressions using proven Alpha 15/6 structures, no LLM needed"

feat: winner_memory.py — stores BRAIN results, identifies winning patterns, feeds back to generation"

feat: winner_memory.py — tracks what works, feeds back winning patterns + proven_templates.py — hardcoded Alpha 6/15 archetypes that ALWAYS produce upward curves"

feat: expression_mutator.py — generates 5 variants from any alpha (decay, horizon, neutralization, vol-scaling, sign flip)"

fix: pipeline v2 — prevent theme repetition in batch, track used themes per batch for diversity"

feat: add cleanup utility to purge old bad alphas with fake fields + prevent theme repetition in pipeline"

fix: hypothesis_hunter v3 — inject novel group keys, prevent theme repetition, force no-quotes rule"

fix: expression_compiler v4 — strip quotes from fields, post-compile validation, wire novel groups, fix all known issues"

fix: ui.py — force PYTHONIOENCODING=utf-8 and NO_COLOR=1 in subprocess to prevent Windows cp1252 crash"

fix: ui.py — fix dry run button by loading .env before subprocess and handling async properly"

fix: ui.py — compatible with Gradio 6.x (remove show_copy_button, move theme to launch)"

feat: add Gradio UI for viewing generated alphas with timestamps, copy button, and live generation

fix: hypothesis_hunter v2 — full field metadata to LLM, strict field constraint

fix: hypothesis_hunter v2 — pass field descriptions, signs, arities + strict constraint to use ONLY listed fields"

fix: add explicit operator arities to LLM compiler prompt — prevents 'Invalid number of inputs' BRAIN errors"

fix: add operator arity validation to lint + enrich LLM prompt with exact arities to prevent 'Invalid number of inputs' errors"

fix: auto-load .env file at startup using python-dotenv

fix: add PROVEN_ARCHETYPES constant back to theme_sampler.py for direct imports

fix: restore PROVEN_ARCHETYPES export + add new exports (get_theme_archetype, get_theme_fields, THEME_TO_ARCHETYPE)"

feat: config v2 — add dataset tier priorities, field selection strategy, and novel group preference

feat: expression_compiler v2 — new archetypes for model77 SUE/PEAD, supply chain, analyst guidance, PCR contrarian + auto field-aware backfill/sign"

feat: lint v2 — loads 71 operators from data/operators.csv, validates field IDs against registry, improved checks

feat: theme_sampler v2 — real BRAIN fields from model77/analyst4/news12/option9/pv13, data-driven gap scoring

feat: add alternative neutralization groups (85+ under-explored keys with AC≤100)

feat: add complete BRAIN operators catalog (71 operators with categories)