126445ae0e
Promote v9.5.36 PEAD champion + prune experiments to 3-tier set; ORB soft-day parity; TGTC infra
...
- v9.5.36 promoted: megacap reaction_min 0.02→0.03 on v9.5.20 cap=0.70 base; Full 13,244%/MDD 11.67%/Sharpe 3.579/OOT 304%; clean Pareto win
- Pruned 130 experiment configs → keep 3 (v9.4.5 max-return, v9.5.36 balanced/champion, v9.5.18 cap=0.55 conservative); rebuilt index
- PEAD engine probes: low_vol anomaly + breakout_52w (retained as falsification probes; not promoted)
- Risk analysis: 11.67% MDD confirmed structurally HYMC (6-day giveback, not controllable); per-name cap sweep on next_open_long/reaction_close falsified (0 MDD impact)
- ORB: port soft-day primary-trigger filter to live engine (6-gate parity with simulator)
- TGTC trader infrastructure + advisor service (new)
- ORB v46/v49 sweep configs, intraday scripts, docs
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
3 months ago
bba1ce6789
Add V47 near-miss composite diagnostic on V46 400d base
...
Tests momentum_20d, grav_pull_20_50, range_pos_52w, obv_slope_20 on V46's
284-trade set (400d). All signals collapse vs V24 base: best G2=0.118R
(momentum_20d) vs 0.291R on V24 and pre-committed gate of 0.30R.
Composite gate (2+ features ≥ 0.20R) also fails. Axis permanently closed.
V46's PEAD event catalyst selection fundamentally changes the ticker/entry
universe vs V24's pure momentum base — momentum signals non-transferable.
V47 requires new data sources (options flow, 13F institutional ownership).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
4 months ago
9095b376d9
Clean up superseded configs and commit accumulated R&D infrastructure
...
Key changes:
- Delete superseded strategy configs: orb_gainers safe_v2-v9, orb_pullback, vwap_reclaim, hypergap, leader_safe
- Add V46 prior_event_types param to domain.py + run.py event type wiring
- Major simulator.py enhancements: sector thrust sleeve, sector proxy mapping, helper functions
- Improve screener.py with better scoring/filtering
- Add new test coverage: test_simulator.py (776 lines) + test_screener.py (313 lines)
- Add V24.1 research candidate configs (w002/w003/w004/entrycap/losscap010 variants)
- Add leader momentum research configs and sweep files
- Update configs/snapshots/registry.json with new strategy registrations
- Add docs/leader_intraday_momentum_workflow.md
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
4 months ago
92840b857a
Close V25 FINRA short-volume axis: backfill CDN data + Phase 1 diagnostic
...
- scripts/backfill_finra_short_volume_cdn.py: bulk backfill FINRA short
sale CDN files (400d) into short_sale_daily table (382K rows inserted)
- apps/intraday_bt/scripts/diag_orb_short_volume_v46.py: Phase 1 diagnostic
on V46 400d trade set (180 trades, 99% coverage)
- scripts/audit_short_volume_coverage.py: fix asyncpg date param types
Result: all three short-ratio features fail G2 (max +0.168R vs gate 0.30R).
Direction is INVERTED from Boehmer (short squeeze dominates over informed-bear
signal in gap-up ORB). V25 axis permanently closed. V46 terminal.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
4 months ago
08e41831bc
Promote ORB Gainers V46: PEAD prior-event signal (D-7 lookback, w=0.12)
...
V24 → V46 via PEAD (Post-Earnings Announcement Drift) signal. Stocks with
earnings_release or guidance_update in prior 7 calendar days show +14.5pp
win rate improvement and +0.348R advantage on ORB breakouts.
Phase 1 diagnostic (291 V24 200d trades):
Pearson=+0.135, Δ=+0.348R, WR gap=+14.5pp — all gates pass.
Phase 2 validation (w=0.12, Pareto-optimal from sweep):
200d: V46 +114.60% / -11.83% / 3.21 vs V24 +94.78% / -11.29% / 2.83
400d: V46 +173.78% / -14.11% / 2.60 vs V24 +162.1% / -13.70% / 2.471
Code changes:
- libs/intraday/domain.py: add prior_event_lookback_days: int = 0 param
- libs/intraday/orb_simulator.py: fix bug — weight_event_catalyst now wired
for gainers_leader engine (was restricted to stocks_in_play_dual_regime only)
- apps/intraday_bt/run.py: _prefetch_prior_event_features_db() helper +
DB routing in both catalyst trigger blocks when prior_event_lookback_days>0
V24 → status: superseded. V46 → status: live_champion.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
4 months ago
4e2d2c0d13
Add V45 sector ETF diagnostic script; document exhausted signal axes
...
diag_orb_sector_etf.py tests sector ETF gap/momentum as ORB quality signals.
Finding: sector ETF signals null (best G2=0.122R vs threshold 0.30R); QQQ regime
filter already captures all market-level information.
Also documents exhausted axes in lineage memory:
- Momentum curve 60-252d (best G2=0.259R, 252d)
- Calendar effects (period-specific artifact)
- Event catalyst (n=39, below G1 n≥120 threshold)
- CMF 20d (G2=0.046R)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
4 months ago
16f49411cb
Add V34-V44 ORB diagnostic scripts; wire obv_slope_5 + min_obv_slope_20d infra
...
Signal axes tested (V34-V44, all failed G2 ≥ 0.30R gate):
- V34 obv_slope_5 (5d): G2=0.083R (null)
- V35 obv_slope composite (5d+20d): regime artifact (200d +18pp, 400d -12pp)
- V36 RSI-14: G2=0.148R, G5a=0.742 (redundant with OBV)
- V37 BB %B / BB width: G2=0.186R
- V38 dollar_vol_trend / sleep_streak / prior_day / vol_trend: all fail G2
- V39 premarket acceleration + hold ratio: G2=0.013R (null)
- V40 prior-day market breadth: G2=0.008R (null)
- V41 min_obv_slope_20d=0.0 hard filter: -15pp (OBV as gate too aggressive)
- V42 52w-high proximity + range position: G2=0.200R (best near-miss, fails)
- V43 30-min ORB window: -14.79% (catastrophic)
- V44 trailing multiplier sweep 0.6-1.0: 0.80 confirmed global optimum
Infrastructure added (backward-compatible, V24 parity preserved):
- features.py: obv_slope_5 enrichment key
- domain.py: weight_obv_slope_5=0.0, min_obv_slope_20d=None
- orb_simulator.py: 5 wiring sites for obv_slope_5; min_obv_slope_20d gate
V24 remains live champion. 20 signal axes exhausted.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
4 months ago
8bd4dba89c
V32 exit sweep FAILED; V33 QQQ ORB quality FAILED — V24 absolute terminal
...
V32 parametric sweep on V24 base (4 variants — all failed):
- risk6 (0.05→0.06): -1.93pp 200d FAIL
- trail_loose (0.8→1.0): +7.69pp 200d but -6.72pp 400d FAIL (regime-specific)
- breakeven_early (1.0→0.75R): -7.39pp 200d FAIL
- trail_09 (0.8→0.9): -14.83pp 400d FAIL (non-monotonic; 0.8 is global optimum)
V33 QQQ ORB candle quality diagnostic (new script):
- qqq_orb_body_pct/return/close_loc: Pearson ~0.001–0.022, day-level ρ ≈ 0
- QQQ daily gap filter already captures all market-level regime signal
- All 3 features fail G1 and G2 completely
All improvement axes now exhausted: 7 signal axes + exit sweep + market intraday.
V24 (OBV-slope w=0.05, 200d +95% / 400d +162%) is the absolute final champion.
Next R&D cycle requires new data source: sector ETF ORB / options flow / architecture.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
4 months ago
5c6950af9d
Fix gap_vs_market computation in diag_orb_trending_features.py
...
Was using non-existent gap_pct key in enrichment dict; now computes from
today_open and prev_close fields. V30 results: hurst INVERTED+G2 FAIL,
ou_theta G1 FAIL (near-zero Pearson). gap_vs_market re-running as V30b.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
4 months ago
733d8dabbd
Add V29/V30 ORB diagnostic scripts; document V29 structural axis failure
...
V29 (diag_orb_structural_features.py): grav_pull G2 FAIL (+0.203R), market_temp G1 FAIL,
momentum_20d near-miss G2 FAIL (+0.291R). All 3 pass G1+G3 but miss G2 threshold.
V30 (diag_orb_trending_features.py): hurst_60d, ou_theta_60d, gap_vs_market (idiosyncratic
gap above QQQ). Running.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
4 months ago
5b7e17d50a
Add V28/V29 ORB diagnostic scripts; V27/V28 axis verdicts documented
...
V28 (diag_orb_volatility_features.py): tests gap_zscore_20d, range_compression_10_60,
atr_ratio_10_60. At 400d: gap_zscore passes G1(n=176)+G3(+10.6pp WR)+G4+G5 but
fails G2 (avg_R gap 0.181R < 0.30R threshold). ABORT — V24 champion.
V29 (diag_orb_structural_features.py): tests grav_pull_20_50, market_temp_5_20,
momentum_20d — structural/regime axis orthogonal to OBV-slope.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
4 months ago
ab34c3ef1b
Diagnose RSI-14/BB%B axis for V27: ABORT — redundant with OBV-slope
...
RSI-14: Pearson=+0.157, G2 passes (+0.316R gap, direction confirmed).
BUT ρ(RSI-14, obv_slope_20)=+0.726 — G5a FAIL. RSI captures same momentum
information as V24's OBV-slope. BB %B is ρ=0.889 with RSI — identical axis.
Both fail G1 (n=98 < 120 threshold). V24 remains champion.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
4 months ago
137b413084
Diagnose ORB tape ignition (V26): NULL result, V24 remains champion
...
5 1-min microstructure features tested on V24 200d trade set (n=101).
range_coil_orb shows directional signal (Pearson=-0.128, WR gap +9.9pp)
but fails G1 (n=101 < 120) and G2 (avg_R gap 0.195R < 0.30R threshold).
p-value ~0.10 — insufficient for promotion. V24 remains champion.
Script fetches 1-min bars from Oracle and caches to data/cache/intraday_1min/.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
4 months ago
1e7fcde14d
Diagnose FINRA short-volume axis for V25: NULL result, V24 remains champion
...
Phase 1 diagnostic (diag_orb_short_volume.py) tests 3 short-ratio features
(prior-day, 20d avg, 20d z-score) on V24 200d trade set via FINRA CDN data.
All features fail all gates (|Pearson| ≤ 0.062, avg_R gap < 0.30R, WR gap < 5pp).
Direction inverts Boehmer 2020 hypothesis: high short ratio correlates positively
with ORB r_multiple (short-squeeze dynamics dominate at intraday ORB timescale).
V25 short-volume overlay not wired. V24 remains live champion.
Also adds scripts/audit_short_volume_coverage.py (Phase 0 DB coverage audit).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
4 months ago
4b8a157a67
Promote V24 ORB Gainers: add OBV-slope(20d) accumulation quality weight
...
Phase 1 diagnostic (diag_orb_quality_features.py) on V23 200d trade set found
obv_slope_20 passes all edge gates: Pearson=+0.2349 with r_multiple, top-tercile
WR 75% vs bottom 59.4% (+15.6pp), avg_R gap +0.394R. Hurst_60 and OU-θ_60 failed.
Weight sweep: 0.05 is Pareto-dominant (0.10/0.15 blow DD).
200d (same window): V24 +94.8% DD-11.3% Sharpe 2.83 vs V23 +85.0% DD-11.6% Sharpe 2.66
400d (same window): V24 +162.1% DD-13.7% Sharpe 2.47 vs V23 +149.4% DD-13.7% Sharpe 2.36
V24 Pareto-dominates V23 on both windows. V23 marked superseded.
Code changes:
- libs/intraday/features.py: add compute_obv_slope_approx() + enrich_daily_bars field
- libs/intraday/domain.py: add weight_obv_slope field to ORBStrategyParams
- libs/intraday/orb_simulator.py: wire obv_slope_20 read/store/score in gainers_leader branch
- configs: orb_gainers_v24_quality_overlay.yaml (new champion, live_readiness: experimental)
- configs: orb_gainers_v23.yaml status → superseded
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
4 months ago
9e622c6614
Investigate compound mode: V23 is absolute champion in all modes
...
- V23 pure compound (live-equivalent): 200d +148.13% DD-14.23%, 400d +209.41% DD-17.35%
- Hybrid V2 compound tested: 200d +175.41% looks promising but 400d +181.71% DD-23.72%
loses to V23 by -27.7pp return AND -6.4pp worse DD → rejected
- Safe v9 compound 400d: +128.49% DD-14.43% — better DD but -81pp return vs V23 → rejected
- V23 tight governor compound 400d: +192.54% DD-17.32% — marginal gain, not worth config
- Live paper trader uses compound mode (engine.py:1816 session_equity = initial_equity + P&L)
- All improvement axes exhausted; V23 daily_reset declared TERMINAL
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
4 months ago
b98442b28a
Add paper trader improvements, web GUI updates, and experiment registry cleanup
...
- Paper trader: Alpaca broker fixes, catchup-thread state improvements
- Web GUI: intraday backtest duplicate run button, paper trading fixes
- Experiment registry: cleanup old v15/v16 experiments, update index
- Tests: Oracle client test additions
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
4 months ago
I Luk Kim