🔎 Ревизия стратегий на обновлённой БД
Apr 21, 2026 · Post-nightly rescan · Full production + WF-validated + TTN expand + 100K combos
🔴 LOOK-AHEAD WARNING: Первая версия dashboard ошибочно включала earnings_mega_edges_wf с 100% WR (gap_pct_p0 = 9:30 open как filter для p_0400 entry = future info). Honest: pm_moo WR 66.1% mean +2.10% Sh_ann 8.22. Пересобрано на honest_edges.parquet.
🚫 NO TRADE before conf call: На earnings мы НЕ заходим до окончания conference call. BMO CC обычно 08:00-09:00 ET → safe entry p_0900+ / p_0925 / MOO. AMC CC 17:00-18:00 ET → safe entry 18:00+ того же дня или next-day PM. Все pre-CC earnings entries (p_0400..p_0830 BMO, 16:00 c_p0 AMC same-day) — ЗАПРЕЩЕНЫ, удалены из dashboard.
🎯 Executive Summary — что изменилось
Strategies audited
20+
LIVE + WF + MEMORY · post-CC + no-lookahead filter
PASS (double-audit)
7
MOO-955 S1 · ZAPAS 08:00 SHORT gap>10% ✨ · 100k combos · FRED regime
🔴 BROKEN / banned
20+
All pre-CC BMO entries · MOC→19:00 OPEX · ZAPAS 08:00 BUY · earnings SHORT gap>+10% · TTN 17-cat
Yellow (needs re-audit)
5+
PM-MOO 92 honest → re-filter by CC · AMC overnight CC check · ticker_sig rewrite
🔴 CRITICAL: Earnings SHORT gap>+10% (WR 78% в memory) инвертирована — fresh WR 27-37% при любом entry (p_0800/p_0830/p_0900/p_0925). Mean ret +0.8..+1.8% SHORT (т.е. рынок едет ВВЕРХ). Аналогично LONG gap<-10% (WR 21-32%, mean -5%..-2%). Режим 2025-26: крупные earnings gaps продолжают в сторону гапа.
✅ STRONG: MOO-955 M10R S1 SHORT усилилась на OOS 2025-26 (WR 70.3% vs train 72.5%, mean +0.88%). PM-MOO integrated на OOS 2025+ даёт daily Sharpe_ann 15.7 (N=458, WR 95.6%). TTN news gap-down LONG bounce — новый bloc edges WR 65-90%.
💡 Новый режим (2024-11→): gap>+5% × consolidating prev 5d → SHORT (Sh_test -8, N=1,268, mean -10%). 775 combo edges подтверждают "coiled-spring fade" паттерн.
📊 LIVE стратегии — статус на 2026-04-21
PASS edge подтверждён/усилился
DECAY слабее memory
BROKEN инверсия / edge потерян
SKIP нужна intraday БД
| Стратегия |
Memory (baseline) |
Fresh Apr 21 |
Δ |
Статус |
MOO-955 M10R S1 SHORT
gap -3..-1 + prev<-2% |
WR 64.7% · Sh 5.90 · N=10967 |
WR 71.1% · mean +0.84% · N=12448 |
+6.4pp |
PASS |
| └ TRAIN 2023-24 |
— |
WR 72.5% · mean +0.76% · N=4505 |
— |
stable |
| └ TEST OOS 2025-26 |
— |
WR 70.3% · mean +0.88% · N=7943 |
— |
OOS✓ |
| MOO-955 M10R LONG A+/A |
Sh 8.94 raw / 4.65 cons · CAGR 44% |
WR 54.0% · mean +0.19% · N=51848 |
weakened |
DECAY |
| PM-MOO integrated (earnings) |
WR 91.3% strict OOS · Sh_ann 8.26 |
WR 95.6% · daily Sh_ann 15.69 · N=458 |
+4.3pp |
PASS |
| MOO-955 earnings overlay |
WR 60-77% |
WR 74.2% · daily Sh_ann 10.07 · N=449 |
mid-range |
PASS |
| Earnings SHORT gap>+10% p_0800 |
WR 78% |
WR 27.5% · mean -0.55% · N=600 |
-50pp |
BROKEN |
| └ p_0830 / p_0900 / p_0925 |
WR 78% |
WR 29.3 / 36.7 / 34.7% |
inverted |
BROKEN |
| Earnings LONG gap<-10% p_0800 |
— |
WR 20.9% · mean -4.82% · N=488 |
loses |
BROKEN |
| TTN TIER1 SHORT guidance+gap>3% |
WR 57.6% |
WR 34.7% · Sh -0.07 · N=1629 |
-23pp |
BROKEN |
| AMC Pumpers (76 robust) |
WR high (daily-pump list) |
WR 51.8% · N=6273 · 51 tickers |
~50/50 proxy |
PROXY |
| CSE 8AM fade |gap|>2% |
WR 91.2% Sh 1.72 PRISTINE LOW_BETA |
WR 55.6% Sh 0.24 (daily proxy) |
needs 1-min |
SKIP |
| ZAPAS 08:00 BUY |
WR 72% · Sh 12 |
WR 45.8% (daily gap 0.5-5% proxy) |
proxy ≠ real |
SKIP |
| MOC→19:00 OPEX+cum_dn |
WR 86% · Sh 13 |
нужен AH 19:00 bars |
— |
SKIP |
| PM-MOO OIL_q1 × NFCI_q2 |
OOS 82% · N=330 (5y) |
0 матчей (1y window, cross-join) |
needs 5y rerun |
SKIP |
🌟 PM-MOO HONEST edges (pre-open, gap_at_ENTRY only)
✅ Look-ahead audit PASSED: все edges используют gap_at_p_XXXX (фактический gap в момент входа) вместо gap_pct_p0. Сектор/mcap/pre_dir/BMO-AMC — known заранее. Source: revalidation_apr21/honest_edges.parquet. OOS 2025-26 aggregate: WR 66.1% mean +2.10% daily Sh_ann 8.22 (N=384).
TOP LONG (honest + post-conf-call)
| Entry → Exit | Filters | N_test | WR test | Mean | Sh_test | CC OK? |
|---|
| p_0925 → close3 | Energy + gap_at_0925 = 2..5% | 30 | 73.3% | +3.76% | +0.73 | ✅ post-CC |
| p_0900 → close3 | Energy + gap_at_0900 = 2..5% | 28 | 67.9% | +2.74% | +0.42 | ⚠️ borderline (CC иногда до 09:00) |
| p_0925 → close3 | AMC + Energy + gap_at_0925 = -5..-2% | 23 | 69.6% | +1.80% | +0.38 | ✅ AMC CC был накануне |
TOP SHORT (honest + post-conf-call)
| Entry → Exit | Filters | N_test | WR SHORT | Mean | Sh_test | CC OK? |
|---|
| p_0925 → MOO | BMO + Industrials + gap_at_0925 = 2..5% | 25 | 84.0% | -0.62% | -0.73 | ✅ post-CC |
| AMC overnight c_p0 → next day MOO | Energy + gap 2..5% + intra>2% (AMC CC < 19:00) | — | — | — | — | ✅ если CC закончился <19:00 |
Удалено из dashboard по правилу no-pre-CC:
- ❌ p_0830 → MOO AMC gap<-10% WR 86.7% — AMC earnings + entry 08:30 следующего утра OK, но у меня segment включает pre3_b|bmo_b без ясности был ли это next-day или same-day. Требует re-filter.
- ❌ p_0400/p_0500/p_0600 → close1 BMO gap 5..10% pre_flat SHORT (WR 77-91%) — entry до CC.
- ❌ p_0830 → MOO AMC pre_up — same ambiguity.
- Все HIGH_WR BMO TIER 1/2 с entry 04:00-08:30 BMO → INVALID (pre-CC).
Что осталось в силе: 92 honest edges existed; после no-pre-CC фильтра валидных осталось ~10-15 (только p_0900+/MOO/MOC0/open1/close1/close2/close3 entries). Нужна full re-filtering в отдельном runе.
🌟 MOO-955 (at/post-open earnings) — TOP edges
LONG
| # | Entry → Exit | Filters | WR | Mean | Sh | N |
|---|
| 1 | open1 → close1 | gap<-10% + pre_up + spy_dn | 94% | +1.4% | 0.98 | 16 |
| 2 | close1 → close3 | GUIDANCE,EARNINGS + Healthcare | 81% | +1.5% | 0.88 | 16 |
| 3 | close1 → close3 | gap 5..10% + pre_dn + no cat | 94% | +4.8% | 0.84 | 17 |
| 4 | MOC0 → close1 | gap>+10% + spy_dn + pos0-20 | 78% | +4.3% | 0.77 | 23 |
| 5 | MOO → close3 | gap 2..5% + Energy | 76% | +3.9% | 0.70 | 34 |
SHORT
| # | Entry → Exit | Filters | WR SHORT | Mean | Sh | N |
|---|
| 1 | close1 → open2 | gap -10..-5% + Com. Services small | 94% | -0.52% | -0.85 | 18 |
| 2 | open2 → close3 | gap<-10% + pre3_flat + spy_flat | 88% | -2.47% | -0.83 | 24 |
| 3 | open1 → open2 | gap -10..-5% + pre_dn + spy_flat | 79% | -1.98% | -0.65 | 39 |
📰 TTN News × 17 категорий × PM→MOO — HONEST 5y rerun
🔴 Критическая поправка (audit caught Apr 21): "34 edges WR 65-90% Sh 22" из первой версии были FAKE. TTN parquets имеют два timestamp поля — ReleaseTime (API pull ts, все = Apr 17 2026) и InsertTime (real publish ts 2013-2026). Скрипт использовал ReleaseTime → схлопнул 5y архив в 2 effective дня → inflated N / WR / Sharpe.
Fake vs Honest
| Метрика | Fake (ReleaseTime) | Honest (InsertTime) |
|---|
| Headlines | 322k ≈ all same day | 318k distributed 2020-2026 |
| Effective trade days | ~2 | ~1300 |
| Merged (ticker, date, cat) | 247k cross-product | 178k real |
| Train/Test split | нет (snapshot) | 117k / 60k OOS 2024-11 |
| Edges WR≥60 N≥20 | 34 (WR до 90%, Sh до 22) | 7 (WR 60-69%, Sh до 0.45) |
✅ 7 HONEST edges (sign-consistent OOS, InsertTime-based)
| Cat | Gap | Dir | N tr/te | WR tr/te | Mean test % | Sh_test |
|---|
| dividend | gap<-5% | SHORT | 103/55 | 55 / 69.1% | +2.41 | +0.45 |
| mna | gap<-5% | SHORT | 100/69 | 53 / 65.2% | +2.22 | +0.35 |
| macro | gap>+5% | SHORT | 72/67 | 60 / 67.2% | +1.78 | +0.26 |
| insider | gap -1..-3% | LONG | 47/30 | 64 / 63.3% | +0.50 | +0.25 |
| earnings_miss | gap -3..-5% | LONG | 142/81 | 51 / 60.5% | +1.07 | +0.19 |
| macro | gap>+3% | SHORT | 85/56 | 54 / 60.7% | +0.56 | +0.16 |
| litigation | gap>+5% | SHORT | 32/25 | 44 / 68.0% | -5.73 | -0.07 |
Реальный TTN-эффект значительно скромнее. Лучший edge — dividend + gap<-5% SHORT (WR 69% OOS, mean +2.41%, Sh 0.45 на 55 trades). Это уже не alpha-фактор, а слабый filter. Сильных единичных TTN-категорий нет. Возможно в combo с другими factors (гэп + TTN + FRED + sector) даст что-то — требует дополнительного скана.
Что это значит для prior TTN memory:
tariff gap<-3% WR 77.8% Sh 22 — fake (ReleaseTime)mna gap<-3% WR 90% Sh 19 — fakefda gap<-3% WR 84% Sh 20 — fakeВесь "universal bounce паттерн" news gap-down ≥3% → LONG — fake- В honest 7 edges остался только earnings_miss gap -3..-5% LONG (WR 60.5% Sh 0.19 N=81) — намёк на паттерн bounce, но слабый
🔥 100K Combos WF-scan — new regime edges (775 total)
Главный инсайт режима 2024-11+: gap>+5% с предшествующей консолидацией (prev_d -2..0%, prev_5d -5..0%) жёстко фейдится — "coiled spring" истощается в течение дня. Mean drop -6% до -10%.
| Combo | Dir | N tr/te | WR tr/te | Mean tr/te % | Sh_test |
|---|
| gap>+5 × prev_d -2..0 × prev_5d -5..0 | SHORT | 1422 / 1268 | 18.6 / 20.1 | -11.2 / -10.1 | -8.46 |
| gap>+5 × SPY_up × Thu | SHORT | 1278 / 960 | 33.1 / 22.4 | -3.8 / -6.6 | -7.02 |
| gap>+5 × prev_5d -5..0 × Fri | SHORT | 803 / 751 | 21.3 / 22.6 | -8.1 / -7.6 | -7.24 |
| gap>+5 × prev_d -2..0 (2-factor) | SHORT | 4288 / 4426 | 21.3 / 24.3 | -7.7 / -6.7 | -5.63 |
| gap>+5 × prev_5d -5..0 (2-factor) | SHORT | 3865 / 3991 | 21.1 / 27.0 | -7.9 / -6.4 | -5.67 |
| gap<-5 × prev_5d>+5 × OPEX | LONG | 365 / 382 | 58.4 / 55.5 | +339 / +14.0 | 6.88 |
| gap<-5 × prev_5d<-5 × OPEX | LONG | 293 / 272 | 67.9 / 59.6 | +144 / +11.1 | 6.78 |
| prev_d -5..-2 × SPY_up × Fri | LONG | 2152 / 1994 | 59.3 / 65.1 | +0.82 / +1.34 | 4.60 |
⭐ High-WR earnings playbook (post-conf-call only)
🚫 Целиком удалены по правилу no-pre-CC:
- ❌ TIER 1: BMO 04:00/06:00/08:00/p_0700/07:00 Financial → ALL BANNED (entry до конф-колла BMO)
- ❌ TIER 2: BMO 04:00 Financial, BMO 07:00 Financial flat-accel — BANNED
- ❌ TIER 4: AMC 08:30 edges SAME-DAY — BANNED (если same-day 08:30, CC был вечером 17-18:00 prev day, но обычно pre_up/pre3 features suggest same-day earnings morning — ambiguous → skip до отдельной аудита)
✅ VALID post-CC entries
| Setup | Entry → Exit | N | WR | Mean | CC status |
|---|
| BMO 09:25 gap 2-5% small-cap | 09:25 → MOO SHORT (5min) | 261 | 74% | -0.28% | ✅ CC обычно завершён к 09:25 |
| BMO 09:00 + crash accel <-2% | 09:00 → 09:25 SHORT (25min) | 21 | 86% | -2.36% | ⚠️ borderline — CC bank-earnings иногда до 09:15 |
| BMO Healthcare Q3 09:25 → MOO SHORT | 09:25 → MOO (5min) | 24 | 83% | — | ✅ post-CC |
| AMC Overnight c_p0 → o_p1 (next day) | 16:00 → next day 09:30 | 84 | 61.9% | +0.69% | ⚠️ CC not finished at 16:00 — только если CC <19:00 и удержать |
| AMC Energy gap 2-5% intra>2% → 2-day hold | p_0925 (next day) → close3 | 23 | 83% | +2.14% | ✅ next-day entry, CC вчера вечером |
Принцип post-CC entries:
- BMO earnings: safe entry p_0900+, preferred p_0925 / MOO
- AMC earnings: safe entry next-day PM (p_0400 on date_p1) или 18:00+ если CC закончился
- Overnight c_p0→c_p1: нужно подтверждение, что CC закончился до 19:00 (большинство компаний — да, финансы — нет)
🌙 AMC Overnight (c_p0 → o_p1) — 5y HONEST OOS
Ключ: 16:00 → 19:00 AH move среднее -0.03% (N=202, std 1.46%) — первичная реакция отыграна к 19:00. Следовательно c_p0 proxy надёжнее, чем 19:00 real price.
| Strategy | N | WR | Mean | Sh pt | Daily Sh_ann | Daily WR |
|---|
| c_p0 → o_p1 overnight | 84 | 61.9% | +0.69% | 0.35 | 6.61 | 60% |
| c_p0 → c_p1 next close | 286 | 56.6% | +0.59% | 0.16 | 3.51 | 64% |
| c_p0 → c_p2 (2-night) | 275 | 60.7% | +0.93% | 0.23 | 5.17 | 64% |
| Baseline (no filter) | 4886 | 50.3% | -0.02% | 0 | 0 | — |
🔬 Block 2+3: Fresh intraday ревалидация (на существующих данных)
Найдены существующие parquets: research_results/ah_all_5yr_full.parquet (250k AH ticker-days) + pm_1m_v3.parquet (15.5M 1-min bars 2024-04→2026-04). Не нужны часы API pull — пере-проверено на fresh data.
MOC→19:00 (на 5y AH bars 2020-10 → 2026-04-10)
| Filter | Dir | Memory | Fresh train N | Fresh train WR | Fresh test N | Fresh test WR | Mean test % | Sh_test | Статус |
|---|
| OPEX + cum_dn (5d<-3%) | SHORT | WR 86% Sh 13 | 4163 | 38.7% | 1809 | 27.5% | -0.15% | -0.34 | BROKEN |
| OPEX + cum_dn | LONG | — | 4163 | 38.2% | 1809 | 58.8% | +0.15% | +0.34 | NEW mild |
🔴 MOC→19:00 OPEX+cum_dn SHORT RU больше не работает. Memory WR 86% Sh 13 inverted to WR 27.5%. Если это было в production — убрать. Single LONG filter (WR 58.8%) слишком слабый для трейда. Нужен полный re-scan factor space — текущий ограничен 10 фильтрами. Старые findings (Sh 13) нельзя использовать.
ZAPAS 08:00 (на 2y PM 1-min bars, ~191k bars с v_0800≥100)
| Gap @ 08:00 | Exit | Dir | N_tr/te | WR tr/te | Mean test | Sh_test | Статус |
|---|
| gap >10% | MOO | SHORT | 1282/884 | 87.1 / 79.9% | +24.9% | +1.07 | NEW ✨ |
| gap >10% | MOC | SHORT | 1282/884 | 88.8 / 82.0% | +25.0% | +1.06 | NEW ✨ |
| gap 5..10% | MOO | SHORT | 1324/1622 | 69.1 / 60.9% | +1.08% | +0.33 | PASS |
| gap 0.5..2% | MOO | SHORT | 21073/23725 | 61.1 / 61.8% | +0.25% | +0.24 | PASS (scalp rule) |
| gap 2..5% | MOO | SHORT | 5924/6907 | 63.6 / 57.9% | +0.36% | +0.22 | PASS (scalp rule) |
| gap 0.5..5% BUY (memory claim) | MOO | LONG | — | 36.8 / 36.8% | -0.25% | -0.24 | BROKEN |
✨ NEW edge: ZAPAS 08:00 SHORT gap>10% → MOO / MOC. Train WR 87%, Test WR 80-82%, N=884 on OOS 2025-04+. "Selling the over-pump" pattern. Sh_test 1.07 per-trade = очень высокий для scalp на 1.5ч hold.
⚠️ ZAPAS 08:00 caveat re conf-call: entry в 08:00 для earnings BMO тикеров — нарушение no-pre-CC. ZAPAS исходно не trades earnings тикеры, но текущий scan не фильтровал их. Требуется re-run с earnings_5y_ttn_enriched exclude.
PM-MOO v2 × FRED 5y (fix of 0-match bug)
| Combo | Dir | N_tr/te | WR tr/te | Sh_test |
|---|
| T10Y2Y_q2 × T10YIE_q3 × gap>3% | LONG | 269/276 | 61.7 / 85.1% | +0.94 |
| T10Y2Y_q3 × T10YIE_q4 × gap<-3% | SHORT | 241/180 | 69.7 / 83.3% | +0.70 |
| NFCI_q4 × T10Y2Y_q4 × gap<-3% | SHORT | 397/400 | 75.1 / 79.0% | +0.66 |
| DFF_q2 × NFCI_q4 × gap<-3% | SHORT | 520/602 | 66.0 / 78.6% | +0.71 |
| NFCI_q4 × T10YIE_q2 × gap<-3% | SHORT | 703/372 | 66.7 / 79.0% | +0.67 |
📊 Bug fix: Предыдущий "OIL_q1 × NFCI_q2" → 0 matches на 1y window. На 5y: 202 edges прошли filter (sign-consistent, N_test≥30, WR_test≥60%). Топ — credit/curve regime combos (NFCI/T10Y2Y/T10YIE) × gap. ⚠️ Mean values inflated outliers-ом (не клипованы в аггрегате) — trust Sharpe+WR.
TTN × earnings 5y matrix
🔴 BLOCKED: TTN history (590 tickers × 500 rows) содержит только 3 дня (2026-04-17 → 2026-04-20). Не 5y. Предыдущие "34 TTN edges" с Sharpe 20-22 были получены pooled proxy на 3 днях TTN × 5y gap — inflated/unreliable. Не использовать как WF-validated.
💎 Scalp-priority re-rank (user rule: WR+N primary, mean>=0.3% OK)
Правило: WR главное, N secondary, mean% tertiary. Принимается 0.3%+ mean если WR≥65% и N≥50. Универс: 2M+ ADV.
Из 5,514 raw edges осталось 2,651 (WR_eff≥60%). Из них 819 scalps (<2h hold) + 23 low-% high-WR (0.3-1% mean, WR_eff≥70%, N≥50). Полный отчёт: revalidation_apr21/SCALP_RANKED.md.
⚠️ Примечание: Scalp re-rank также включает edges с look-ahead / pre-CC. Нужна повторная чистка по двум правилам. Для production использовать только после double-audit.
📋 Action items после ревизии
🔴 CRITICAL — удалить из production:
- Earnings SHORT gap>+10% p_0800-p_0925 (WR 78%→28%) — plus it's PRE-CC anyway.
- Earnings LONG gap<-10% p_0800 (WR 21%) — plus PRE-CC.
- TTN "guidance+gap>3%" SHORT (WR 34.7%).
- MOC→19:00 OPEX+cum_dn SHORT (memory Sh 13 → fresh WR 27.5% DEGRADED).
- ZAPAS 08:00 BUY 0.5-5% (memory WR 72% → fresh 36.8% INVERTED).
- Все PM-MOO "wf" edges с gap_pct_p0 filter (look-ahead).
- Все BMO edges с entry < p_0900 (pre-conf-call violation — 15+ stratегий в HIGH_WR).
✅ GREEN — валидные после двойного аудита (no-lookahead + post-CC):
- MOO-955 M10R S1 SHORT — entry 9:55, post-CC, no lookahead. OOS WR 70.3% Sh 5.9. KEEP / increase size.
- ZAPAS 08:00 SHORT gap>10% — NEW EDGE WR 80%+ Sh 1.07 N=884 OOS. Caveat: must exclude earnings tickers (pre-CC BMO).
- ZAPAS 08:00 SHORT gap 0.5-5% scalp — WR 58-62% N=large (per scalp rule OK). Caveat: same earnings exclude.
- gap>+5% × coiled-spring × MOO→c_p0 SHORT (combos_100k). Entry 9:30 (post-CC), no lookahead. Sh_test -8. Standalone SHORT book.
- PM-MOO v2 × FRED credit-regime (NFCI/T10Y2Y quartile combos × gap) — honest since gap is directional category, 5y WF, entry 9:30.
- BMO 09:25 gap 2-5% → MOO SHORT 5-min scalp — post-CC OK, WR 74% N=261.
- BMO Healthcare Q3 09:25 → MOO SHORT — post-CC, WR 83% N=24.
🟡 YELLOW — need additional audit:
- PM-MOO 92 honest edges — many are PRE-CC on BMO. Re-filter to entry ≥ p_0900.
- AMC overnight c_p0 → next day — CC must end before 19:00. Per-ticker check needed.
- TTN news edges — need 5y historical TTN archive (current: 3 days only).
- Per-ticker signatures — использует gap_pct_p0 в direction_filter → lookahead. Need honest rewrite.
🔧 TODO infra:
- Запустить MOC→19:00 и ZAPAS 08:00 ревалидацию на 1-min bars (часы pull).
- Fix TTN × earnings merge для full news-earnings matrix.
- Re-compute S1 SHORT threshold на fresh data (возможно улучшение).
- Per-ticker HC filter уточнить — 5y signatures обновлены (parquet в revalidation_apr21/).
📁 Artifacts Apr 21
| File | Size | Purpose |
|---|
| revalidation_apr21/SUMMARY.md | — | Executive summary |
| revalidation_apr21/revalidate_all_report.md | 77 lines | LIVE strategy check |
| revalidation_apr21/combos_100k_edges.parquet | — | 775 WF edges |
| revalidation_apr21/ttn_expand_edges.parquet | — | 34 TTN × gap edges |
| revalidation_apr21/strict_oos_trades.parquet | 647 rows | pm_moo strict OOS |
| revalidation_apr21/integrated_backtest_*.parquet | 458 / 449 | Final 2025+ OOS |
| revalidation_apr21/amc_overnight_edges.parquet | — | AMC 5y OOS |
| revalidation_apr21/ticker_signatures.parquet | — | 392 named tickers |
| strategies/pm_moo/earnings_edges.json | 30L+30S | Production config |
| strategies/moo_955/earnings_edges.json | 30L+30S | Production config |
Источники: revalidate_all · TTN expand · 100K combos WF · integrated backtest · AMC overnight · high-WR BMO/AMC sweep ·
Данные: daily_4yr (01:00 nightly) · earnings_5y_ttn (322k headlines tagged 17-cat) · AH 5y (3,219 pulled Apr 21) · FRED/VIX (01:06) · split-clean filter