PM→MOO System — Scoring Consolidation + 100 Endpoints Research Plan

Generated 2026-04-17 | ZAPAS / CSE Bridge / MOO→9:55

1. Scoring Engines — Canonical Mapping

File	Strategy	Entry point	Scale	Last touch	Status	Action
scoring_pm_moo_v1.py	ZAPAS + CSE Bridge	`PmMooScorer().score_zapas() / score_cse()`	0-25 (min_B+=13)	Apr 15 audit	CANONICAL — production	Keep. Build CSE backtest on top.
scoring_v7.py	MOO→9:55	`score_dataframe()`	combo sum (unbounded)	Apr 14	CANONICAL — production	Keep. Used by backtest_moo_955_v8.
scoring_v8_overlay.py	MOO→9:55 overlay	`apply_v8_overlay(stocks_list)`	±5 adj on v7	Apr 17	CANONICAL — add-on	Keep. Apply after v7.
scoring_pm_moo_v1_beta_overlay.py	ZAPAS β-overlay	—	±2 adj on v1	Apr 16	EXPERIMENTAL	Evaluate: integrate into v1 or deprecate.
scoring_zapas_v2.py	ZAPAS (old)	—	0-25	Apr 15	DEPRECATED	Delete after confirming pm_moo_v1 covers all factors.
scoring_v5.py / v6.py / v6_auto.py	MOO→9:45 (old)	—	0-25	Apr 12	DEPRECATED	v7 replaces. Move to `archive/`.
scoring_pm_moo_v1_backup.py	—	—	—	—	BACKUP	Delete.

Known scoring bugs (from Apr 17 audit)

v7 squeeze-fade blindspot: на SPY_gap > 2% выдаёт массовый LONG signal (1369 сделок Apr 8 2026), но мелкие стоки фейдят → WR 32.5%, -$19K. Нужен trap-combo: SPY_gap>2% AND LONG AND mcap<2B → skip.
v7 MASTER_FILTERED требует sector_etf_gaps.csv + foreign_etf_gaps.csv + SPY/QQQ/ARKK/vxx_gap. Если хоть одна колонка NaN → score=0, grade=C.
v8_overlay hard-skip pre_vol<50K — но pre_vol иногда 0 из-за missing enrichment. Fix: treat 0 as unknown, не skip.
pm_moo_v1 min_score=13 → 82% win days. Но при hot strategy может дропать слишком много. Нужна adaptive threshold.

2. Data Pipeline Canonical Files

Role	Canonical file	Rebuilt by	Used by
Primary trades (1y)	trades_enriched_v6.csv	update_all.py + refill_v6_*.py	scoring_v7, v8_overlay
Backfill (2y)	all_trades_backfill_2023-04-15_2025-04-14.csv	manual (old)	research_v8_3y_expanded (missing vix_level)
PM snapshots	pm_timing_all_cache.pkl	_fetch_pm_timing_all.py + pull_fresh_daily.py	build_zapas_dataset
ZAPAS dataset	zapas_directional_full.parquet	build_zapas_dataset.py	backtest_apr17_full.py
CSE signals (static)	zapas_87k_scored.parquet	— (frozen)	backtest_zapas_cse_apr17.py
Intraday 1m	intraday_1m_gap_days_v2.parquet	pull_fresh_daily.py / pyarrow concat	backtest_moo_955_v8.py
Daily OHLCV + RSI	cache_daily/daily_bars_cache.parquet	(daily_refresh)	scoring_v7, refill_v6_technicals
Daily regime (SPY/QQQ etc)	daily_regime.parquet	(daily_refresh)	refill_v6_technicals (breadth proxy)
Sector ETF gaps	sector_etf_gaps.csv	refresh_etf_gaps.py	scoring_v7
Foreign ETF gaps	foreign_etf_gaps.csv	refresh_etf_gaps.py	scoring_v7
Beta profiles	beta_profiles_v3.json	nightly_beta_refresh.py	scoring_v8_overlay
intraday_1m_gap_days.parquet	LEGACY (Apr 5)	—	— (archive)

3. Datum API — 100 Endpoints Research Plan

Currently used: /daily, /intraday, /intraday/v3 partial, /calculations/volatility_20d/90d/260d, /calculations/avg_true_range_14d, /calculations/high_low_52w, /calculations/corr_beta/v2, /calculations/profiler_top_correlation, /auctions/gaps, /imbalance/imb_09_25, /tickers/etf.

UNUSED или недоиспользуемые (приоритеты по потенциальному edge):

Priority 1 — высокий edge потенциал

Endpoint	Hypothesis	Estimated edge	Effort
/calculations/profiler_volume	Volume concentration per ticker — подсветит squeeze candidates (низкий обычно vol + высокий обычный = squeeze trap)	+3-5% WR на squeeze-fade day detection (решает Apr 8 баг)	1 day
/calculations/median_opg_volume_20d	Typical opening auction volume → detect abnormal open (если opg_vol > 3× median = fade сигнал)	+2-4% WR на MOO→9:55	1 day
/auctions, /auctions/gaps	Full auction data (not just gap) — imbalance+size+print_price → predict open drift	+2-3% WR на MOO direction	2 days
/imbalance/imb_09_25	MOC/MOO imbalance @ 09:25 — signed directional pressure	+3-5% WR (Factor: imbalance aligned with direction)	1 day
/intraday/extended_high_low + extended_high_low_postmarket	52w high/low hit в PM = PM breakout detection	+2-4% WR на ZAPAS (HIGH + 52wHi confluence)	1 day
/corporate_actions/listings + ticker_changes	IPO/spinoff detection — flag untradeable; ticker rename = avoid	Avoid -10% disasters (-2 traps/year)	0.5 day

Priority 2 — средний edge

Endpoint	Hypothesis	Notes
/calculations/avg_premarket_value_traded_ex_finr_20d/3m	$-value PM volume vs shares — captures large-cap PM activity better than raw vol	Replace pre_vol threshold
/calculations/avg_postmarket_volume_ex_finr_90d	PM→AH continuation (AH vol prev day predicts PM next day)	ZAPAS entry timing
/intraday/v3/benchmark	Relative performance vs sector/SPY in real-time	Replaces manual SPY_gap calc
/intraday/money_traded_segments, volume_segments	Intraday money flow by time segment → confirm momentum legitimacy	Exit timing refinement
/intraday/ticker_count_at_open	Breadth at open — how many tickers gapping → detect squeeze days	Apr 8 bug fix candidate
/ratings, /ratings/grades	Analyst ratings — upgrade/downgrade edge	Already partial in v6, unify
/dvd/cash_dividends, stock_dividends, splits	Avoid ex-div trades (price drop ≠ trade signal)	Trap filter
/snapshots/yesterday_values	Prev-day snapshot (close, high, low, vol) in one call	Simplifies enrichment pipeline
/opg_and_clo, /opg_and_clo/adjusted_closes	Opening+closing auction prints, split-adjusted	Improve MOO price accuracy

Priority 3 — low/niche

/bics_sectors/{level} — BICS sector hierarchy (already have via Finviz; Datum version cleaner)
/countries — for ADR research expansion
/holidays, /short_days, /working_days — calendar util, integrate into event calendar step
/futures/active_contracts — ES/NQ/RTY future PM tracking
/calculations/median_auction_money_traded_20d — refined auction size normalization
/reports, /reports/extended — delivery reports (likely not useful for trading)
/calendar, /calendar/timerange_chunks — session calendar (already have locally)
/latest_close_prices — fast last-close lookup (replace manual)

4. Research Hypothesis Queue (ordered by ROI)

Apr 8 squeeze-fade fix → /intraday/ticker_count_at_open + /calculations/profiler_volume
Proxy for "mass gap day" → add trap-combo to scoring_v7/v8. Immediate priority.
Imbalance 9:25 → /imbalance/imb_09_25 already partial.
Full research: direction-aligned imbalance predicts MOO→9:55 WR boost.
Auction edge → /auctions full data (print price vs imbalance).
Hypothesis: when auction print differs from 9:25 bid-ask mid, direction predicts 9:55.
52-week high/low confluence → /intraday/extended_high_low.
ZAPAS entry at PM HIGH confluent with 52wHi → higher conviction.
Money traded segments → /intraday/money_traded_segments.
Detect: if PM drive money-weighted vs volume-weighted differ → trap.
PM value-traded vs shares → /calculations/avg_premarket_value_traded_ex_finr_20d.
Mega-cap with $50M+ PM volume but 30K shares ≠ retail squeeze; discriminates.
Adjusted closes → /opg_and_clo/adjusted_closes.
Fix split/div distortions in historical backtests.
Ticker changes / listings → /corporate_actions/*.
Filter new listings (first 5 days) and renames.
Futures active contracts → /futures/active_contracts.
Ensure futures PM ret pulled correctly across contract rolls.
3-year MOO→9:55 backfill → pull 9:55 bars for 347K backfill trades via /intraday.
Blocker for v8 3y research. ~8 hours pull.

5. Nightly Cron (current: nightly_full.bat, 6 steps)

[1/6] daily_refresh.bat           — enrichment/v6/finviz/screen/briefing
[2/6] pull_fresh_daily.py         — PM snapshots + intraday 1m (Datum)
[3/6] refresh_etf_gaps.py         — sector+foreign ETF gaps from Datum /daily
[4/6] refill_v6_etf_gaps.py       — fill SPY/QQQ/ARKK/VXX in v6 for missing rows
[5/6] refill_v6_technicals.py     — rsi14/atr14/pre_vol/breadth/prev_day_ret
[6/6] refresh_pm_moo_dashboard.py — rebuild 3 backtests + regen HTML

User action: schtasks /change /tn "DailyRefreshNY1AM" /tr "C:\datum-api-examples-main\nightly_full.bat"

6. Next Sprint — Concrete Tasks

#	Task	Effort	Owner
1	Delete deprecated scoring files (v5, v6, v6_auto, zapas_v2, pm_moo_v1_backup)	15 min	easy
2	Build Apr 8 trap-combo into scoring_v7 (SPY_gap>2% & LONG & mcap<2B → skip)	2 hrs	scoring team
3	Integrate 6 Priority-1 endpoints into refresh pipeline	3 days	backend
4	Research hypothesis #1-6 (one per day)	6 days	research
5	3y MOO→9:55 ret_955 pull (347K trades, ~8 hrs Datum)	1 day	data
6	Consolidate backtest runners (3 scripts → 1)	4 hrs	refactor