2aba6418e6
Add overlay engine, ranking models, snapshot pipelines, and research tools
...
New libs: overlay curve builder, ranking models, continuation/merged
snapshot export, intraday features. New tools: overlay evaluator,
ranking model builder, deep evaluation, fullsplit batch runner.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
5 months ago
57d38ecfe0
Add earnings surprise feature pipeline and snapshot export improvements
...
Adds earnings surprise extraction to parser/features/labeler pipeline,
improves filing fetcher robustness, and extends snapshot export with
new field support.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
5 months ago
16613f1758
Fix _rows_to_table to collect keys from ALL rows, not just first
...
Previously only used rows[0].keys() — columns present in later rows
(like earnings_surprise_pct from sparse features) were silently dropped.
Now collects all unique keys across all rows.
YoY earnings surprise tested: WR spread only 2.5pp (55.2% vs 52.7%).
Not actionable — YoY growth != analyst consensus surprise.
v6new.30 remains the framework optimum.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
5 months ago
8dacdaab4e
Add v10 macro regime scoring (VIX+HY) and FRED macro features — SQS 59.1 (rejected)
...
New features:
- _enrich_macro_features() in snapshot_export: adds macro_vix, macro_hy_spread from FRED
- compute_return_max_long_score_v10: +12% bonus in favorable regime (VIX>18+HY>3.25)
- _macro_regime_score(): regime-aware scoring component
Findings:
- VIX signal is statistically strong: 62.3% WR (VIX>18+HY>3.25) vs 50.8% (other)
- But scoring bonus promotes marginal trades, diluting OOS quality
- Same pattern as eps_growth, drift bonus: raw signal ≠ scoring improvement
- v6new.17 SQS 59.1 < v6new.9 SQS 63.3
v6new.9 remains best at SQS 63.3 after 17 experiments.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
5 months ago
85c4d98987
Add v6new experiment suite: scoring v8/v9/v9g, cross-event drift, coverage engines
...
Scoring additions (libs/backtest/scoring.py):
- v8: conditional financial bonus (eps_growth_qoq/revenue_growth_qoq)
- v9: cross-event drift momentum (+/-10% from prior same-ticker 5d return)
- v9g: gated variant — reject events with negative prior drift
Snapshot export (libs/export/snapshot_export.py):
- _enrich_prior_event_drift: computes prior_event_fwd5d for all snapshots
- smallcap-liquid-long-v1 universe profile ($500M-$2B)
- market_cap_max support in screener and filtering
8 experiment configs (v6new.1-v6new.8):
- v6new.1: unknown event reclassification (neutral)
- v6new.2: financial features (neutral, EPS growth is noise)
- v6new.3: small-cap (blocked, survivorship bias)
- v6new.4-6: cross-event drift variants (rejected)
- v6new.7: engine pruning (quality up, count down)
- v6new.8: coverage expansion with 2 new post-market engines (best result)
Best result: v6new.8 SQS 41.5 vs v6.29 control 32.3 on same conditions.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
5 months ago
2395a0c0c3
feat: PEAD mid-cap strategy + pipeline hardening + README cleanup
...
- Implement PEAD 7% Long+Short strategy with mid-cap universe expansion
- Add Stock Oracle screener/company clients, text sentiment features
- Enhance backtest engine: short-side execution, walk-forward CV, MFE/MAE analysis
- Harden pipeline: sequential Oracle API calls, scoring recalibration (event_quality 65%)
- Add experiment configs for 60+ strategy variants and journal tracking
- Add review/analysis CLI tools
- Remove obsolete dev/phase0-4 design documents and analysis scripts
- Clean README to reflect only implemented features (remove unbuilt adapters/engines)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
5 months ago
d4ae900286
feat: expand data pipeline and integrate event features into scoring
...
Data expansion:
- Poll SEC 8-K filings from 2025-10-01 to 2026-03-12 (was ~2 months)
- Pipeline: 36 new filings → 24 new events → total 44 events, 15 symbols
- Re-export with merged features (market_v1+event_v1+financial_v1)
- Parquet columns: 7 → 41 (adds signal_strength, guidance_direction,
document_quality, oneoff_penalty, eps_growth_qoq, etc.)
Score model v2:
- Add event quality component (15% weight): signal_strength_score,
guidance_direction_score, document_quality_score
- Add risk penalty component (10% weight): inverted oneoff_penalty
- Rebalance market weights: reaction 25%, close 25%, volume 15%, gap 10%
- Graceful degradation when event features are absent (returns 0.5)
Export pipeline:
- Add --feature-versions CLI flag to merge multiple feature types
- export_dataset_snapshot() accepts feature_versions list parameter
- Groups features by event_id and merges feature_json dicts
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
5 months ago
bbe0e31150
feat: implement Phase 3 -- LLM enrichment, labeler, review queue, dataset export
...
- libs/llm/: OllamaClient (httpx async, retry), LLMCacheStore (SHA-256 DB cache),
prompt registry (event_classifier_v1), LLMParser (cache→prompt→validate→repair)
- libs/parser/merger.py: rule+LLM canonical merge with provenance tracking,
conflict detection (both confident + disagree), should_queue_for_review()
- libs/db/models.py: LLMCallCache, ReviewItem, EventLabel 3개 ORM 모델 추가
- libs/db/migrations/versions/0002_phase3_tables.py: Phase 3 Alembic migration
- libs/labeler/: filing_time_bucket→reaction_date, 1D/3D/5D fwd return, MFE/MAE
- libs/review/queue.py: create(dedup)/resolve/list ReviewItem
- libs/export/snapshot_export.py: temporal split + Parquet + manifest.json
- apps/: label_generator, dataset_export, review CLI, gold set evaluator
- 42개 신규 테스트 추가 (unit 32 + integration 4 + replay 1) — 152/152 통과
- libs/common/config.py: OLLAMA_URL/MODEL/TIMEOUT 설정 추가
- libs/common/logging.py: bugfix — add_logger_name incompatible with PrintLoggerFactory
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
5 months ago
I Luk Kim