2026-11-06 17:00 ET: Software/SaaS earnings bifurcation under Friday STRESS — TEAM day-1 crush (SM) vs APP LM day-2 & NET day-1 gap-fade; ABNB LM vs DKNG crush; propose S-151

Slot: Fri 2026-11-06 17:00 ET hourly research (post-close / last research hour of the day). Status: written ~17:10–17:35 ET (America/New_York); whiggity skipped (executor has no computerUse / Monica browser — needs parent computerUse dispatch). Screening/research, not investment advice. No rule is implemented by this report.

Actionable for today's (Fri Nov 6) trading: NO. Live-rule entry window was 10:00 AM EST (60-minute time stop → ~11:00). Session is closed. Post-close paths under STRESS (VIX 25.64 >22; VIX9D/VIX 1.060) with short_momentum suspended are descriptive only — not a validated fade, chase, or SM reactivation. Weekend / Mon prep remains the 11:00 gap stack (KKR BMO Mon 11-09, BRK Sat 11-07 ~7:00 a.m. CT filing → Mon reaction, plus TTD/HIMS/OXY/MARA/CEG).

1. Incremental focus (rotate from 10:00–16:00)

Prior Fri hours covered VST/CEG (10:00 — briefly noted NET/TEAM/MCHP as do-not-MR-fade day-1 leftovers, not a software sector report), the Mon gap stack (11:00), industrials (S-146), staples (S-147), memory/storage (S-148), healthcare (S-149), and financials (S-150).

This hour rotates to the software / SaaS earnings-week bifurcation still visible on the post-close run, with adjacent travel vs gaming day-1 paths:

1. TEAM — sole short_momentum name (suspended): day-1 −16.15% after Thu AMC outlook scare.
2. APP — long_momentum day-2 continuation +9.86% after Wed AMC blowout / Thu +31% reaction.
3. NET — day-1 after Thu AMC beat/raise: gap +17.4%, close +9.91%, rangePos 0.18 (gave back most of the open).
4. Soft enterprise SaaS — CRM −3.94%, INTU −3.76%, WDAY −3.63%, ADBE −2.90% (no same-session prints).
5. ABNB — long_momentum day-1 +11.67% after Thu AMC beat/raise + $12B buyback (Internet/Media; travel cluster-3).
6. DKNG — day-1 −21.30% (gap −26.28%) after Thu AMC miss + Predictions-driven guide cut (Consumer Disc. singleton; ADV liquidity fail).

2. Live TRAI evidence and clock audit

Read-only API checks returned HTTP 200 around drafting (~17:07 ET). Linked endpoints are mutable. Sources: latest (postclose), events, correlations, regime, scorecard, macro. Yahoo daily bars via public chart API (UA; unofficial).

Scorecard constraint [TRAI]: live-rule 10:00 AM EST entry, 1 ATR stop / 2 ATR target, 60-minute time stop. Do not claim new live-rule edge. Post-close paths are descriptive only.

Post-close tape (Yahoo daily + TRAI latest)

Takeaway: Under Friday post-close SPY/QQQ/SMH green + VIX still >22, Software did not trade as one book. TEAM crushed inside cluster-11 while CRM/NOW soft-to-flat; APP singleton ripped day-2; NET gapped +17% then closed +10% at rangePos 0.18; cluster-9 productivity soft without prints; travel ABNB LM vs gaming DKNG crush vs BKNG −4.8% inside the same cluster-3 home as ABNB. That is the same class of sector-flag / correlation-home split as S-146…S-150. Proposed S-151 below.

3. Correlation structure vs Software sector flag [TRAI + S-001]

60-day clusters (/api/correlations, asOf 2026-11-05):

Label error: TRAI Software sector label still mixes cluster 11 collab, cluster 7 security/edge, cluster 9 productivity/adtech, APP singleton, and (historically) ORCL-in-semis. Fading or chasing "software" as one residual book after this earnings week is the S-146…S-150 class of error. Proposed S-151.

4. Business facts and primary catalysts (with citations)

TEAM — Atlassian (collab SaaS; cluster 11) — SM day-1 [V-P prior + S]

• Business: Jira/Confluence/service-management seat-based cloud; TRAI Software.
• Print: FQ1 FY27 AMC 2026-11-05 (Atlassian IR event [V-P schedule]). Issuer release (Business Wire, prior KB [V-P] on 11-05 17:00): revenue $1.772B (+24%), Cloud $1.302B (+31%), subscription ARR $6.812B (+23%), non-GAAP EPS $1.65, FCF $463M; Q2 rev $1.825–1.835B; FY guide raises vs Aug: ARR/rev/Cloud growth 18.5%/13.75%/26.0%, DC −16%, non-GAAP op. margin 25.5% (Business Wire 2026-11-05 — primary; this-hour WebFetch timed out; numbers match prior verified KB + secondary Investing.com [S]).
• Reaction: Yahoo Fri −16.15% to 153.68 (O 154.92 / H 165.31 / L 151.50); TRAI SM; residZ −5.04. Secondary narrative: investors priced moderation of ARR/Cloud vs Q1 run-rate despite beat/raise (Investing.com / Yahoo syndicated, 2026-11-06 [S]).
• TRAI hist: avg \|gap\| 13.71% (2.05 ATR), n=4 — Fri day-1 c/c ≈ in-line with hist mega-gap class. S-006: do not same-day MR-fade. S-141: do not reactivate SM.
• Feeds S-151 (cluster-11 vs rest of Software).

APP — AppLovin (adtech AI; cluster 27 singleton) — LM day-2 [V-P]

• Business: AXON AI mobile advertising / app monetization; TRAI Software.
• Print: Q3 AMC 2026-11-04 (AppLovin IR / Business Wire [V-P]): revenue $2,185M (+55%), net income $1,439M (+72%), diluted EPS $4.30, adj. EBITDA $1,842M (84% margin); Q4 rev guide $2,365–2,395M; board +$3.2B buyback auth → $3.6B remaining Eo Oct.
• Reaction path: Thu c/c +31.04% to 356.10; Fri Yahoo +9.86% to 391.22 (H 398) → TRAI LM. Day-2 continuation after mega up-gap — descriptive; S-005 / S-066 / S-137: no proven drift premium into flat-by-3:55 books; do not treat as Monday LM seed.
• Feeds S-151 (singleton ≠ Software book).

NET — Cloudflare (edge/security/dev platform; cluster 7) — day-1 gap-fade [V-P]

• Business: connectivity cloud (CDN, security, Workers/AI platform); TRAI Software.
• Print: Q3 AMC 2026-11-05 (Cloudflare press release [V-P]): revenue $782.5M (+39%); non-GAAP EPS $0.38; non-GAAP op. income $137.1M (17.5% of rev); cRPO +37%; Q4 rev $845–846M; FY26 rev $2,963–2,964M; FY non-GAAP EPS $1.35.
• Reaction: TRAI gap +17.37%; Yahoo Fri close 268.75 (+9.91%) after Thu close 244.51; rangePos 0.18 — opened near highs, closed near lows of day. S-005 / S-066: big up-gap reaction days often finish the move early; do not chase into close / do not invent Monday LM from rangePos 0.18. ATR% >8% → liquidity fail for live sizing.
• Opposite sign vs TEAM same AMC night (S-014 class: same calendar, different catalyst quality). Feeds S-151 (cluster-7 ≠ cluster-11).

ABNB — Airbnb (travel marketplace; cluster 3) — LM day-1 [V-P]

• Business: short-term stays + expanding hotels/services; TRAI Internet/Media.
• Print: Q3 AMC 2026-11-05 (Airbnb Newsroom [V-P]): revenue ~$4.9B (+19%); GBV $27.7B (+21%); nights/seats +16%; net income ~$1.9B; adj. EBITDA ~$2.5B (margin 52%); Q4 rev $3.22–3.28B (+16–18%); FY26 rev growth high teens, adj. EBITDA margin ~36%; new $12B Class A repurchase authorization.
• Reaction: TRAI LM +11.67% to 189.99 (gap +12.35%); residZ +5.42. BKNG −4.77% same day inside cluster-3 — print-driven twin break (S-118 related; do not size BKNG as ABNB co-move on ABNB print day).
• Feeds S-151 companion (travel cluster-3 print ≠ Software; and ≠ BKNG residual).

DKNG — DraftKings (sportsbook / iGaming / Predictions; cluster 28) — day-1 crush [V-P]

• Business: digital sports betting, iGaming, prediction markets; TRAI Consumer Disc.
• Print: Q3 AMC 2026-11-05; call Fri 8:30 AM ET (DraftKings IR [V-P]): revenue $1,329M (+16%); adj. EPS −$0.53; adj. EBITDA −$316.9M; MUPs 4.1M (+13%); ARPMUP $109 (+3%). FY26 guide cut: rev $6.35–6.55B, adj. EBITDA $175–275M (Predictions adj. EBITDA investment $650–700M). FY27 intro: rev $7.35–7.75B, adj. EBITDA $800M–1.0B incl. Predictions investment $250–350M; Predictions rev $650–750M.
• Reaction: Yahoo Fri −21.30% to 18.51 (gap −26.28%, L 16.55, rangePos 0.875 bounce-off-lows); residZ −6.97; relVol 5.66. Not in SM (ADV < $500M). S-006: do not same-day MR-fade a mega earnings gap. Singleton cluster — do not fold into Software or ABNB travel risk.
• Feeds S-151 companion (Consumer Disc. gaming singleton vs travel cluster-3).

Soft non-print Software (cluster 9 / 11 peers) [TRAI]

• CRM −3.94%, INTU −3.76%, WDAY −3.63%, ADBE −2.90% with no verified Fri issuer 8-K this hour — consistent with TEAM sympathy + STRESS duration under XLK 20d lag, not a shared print. Next prints: INTU 11-19, WDAY 11-24, CRM estimate 12-02 (/api/events).

5. Strategy implications (proposals only — 6 PM decides)

Existing rules touched

• S-001 / S-014: Software sector flag over-aggregates; Fri post-close path is the cleanest live case since 10-30.
• S-005 / S-066 / S-137: APP day-2 LM and NET day-1 gap-fade (rangePos 0.18) — no chase / no invent Monday LM.
• S-006: TEAM and DKNG mega down-gaps — do not same-day MR-fade.
• S-141: TEAM is sole SM name — research-only; do not reactivate SM on outlook-scare crush.
• S-144: STRESS + inversion still on at close (VIX 25.64 / ratio 1.06); MR bucket empty on postclose — no fade-long sizing claim.
• S-145: Mon gap stack unchanged — not re-litigated.
• S-118: ABNB print day vs BKNG divergence — log as live case for MAR→travel read-through family (sign disagreement on peer day).

S-151: Split TRAI "Software" SM/MR/LM risk into cluster-11 collab vs cluster-7 security/edge vs cluster-9 productivity vs APP singleton · PROPOSAL (extends S-001 / S-014 / S-146…S-150)

• Rule (proposal): when clusterFlag or sector label is Software, do not count cluster 11 (TEAM/CRM/NOW), cluster 7 (NET/DDOG/ZS/CRWD/PANW/PLTR/SNOW), cluster 9 (INTU/WDAY/ADBE/TTD), and APP (cluster 27 singleton) as one correlated residual book. Cap exposure using 60d correlation clusters. Optional hand map: collab/seat SaaS (11), security/observability/edge (7), productivity/adtech (9), adtech-AI singleton (APP). ORCL remains semis cluster-1 (prior notes) — exclude from Software risk unit.
• Companion tests: (1) after a Software SM crush under VIX>22, measure next-session cluster-11 mean vs cluster-7 vs cluster-9 vs APP return spreads; (2) on same-night dual AMC (TEAM+NET class), test sign agreement of day-1 c/c (hypothesis: disagree when guide quality differs); (3) travel/gaming adjacency — do not inherit ABNB LM or DKNG crush into Software caps; test ABNB–BKNG same-day ρ break on ABNB print days. Costs, day clustering, OOS, multiple-testing caution.
• Why (live case): Fri 2026-11-06 post-close — TEAM −16% SM vs APP LM +9.9% vs NET +9.9% rangePos 0.18 vs CRM/INTU/WDAY soft −3–4%; ABNB LM +11.7% vs DKNG −21% vs BKNG −4.8%.
• Evidence: this report §2–4, §8.

6. Whiggity status

Skipped — needs computerUse. This executor cannot drive Monica/whiggity in-browser. Prefer skip over blocking the hour. Recent Fri hours exhausted budget after ≤4 rounds; no consult attempted via API.

7. Actionable for TODAY

NO. Market closed; 10:00 entry window long gone; SM suspended; STRESS risk still halved. Report is knowledge-base / Mon-prep context only.

8. Open questions for 6 PM refinement

1. Accept S-151 Software cluster split (mirrors S-146…S-150 family)?
2. Score TEAM day-1 and DKNG day-1 into the S-006 mega-gap do-not-fade case file; score NET rangePos 0.18 into S-005/S-066 big-up-gap fade-from-highs file; score APP day-2 into S-137.
3. Whether BKNG's −4.8% on ABNB print day should tighten S-118 (peer read-through only when ρ gate holds intraday).
4. Mon handoff: still S-145 gap stack — software leftovers (TEAM/APP/NET) are not Mon calendar risk unless new filings appear over the weekend [UNV].

Sources (selected)

• TRAI postclose latest generatedAtET Nov 6, 2026, 4:49 PM ET; correlations asOf 2026-11-05; events asOf 2026-11-06.
• Yahoo Finance public chart API daily bars 2026-11-02→11-06 (TEAM/APP/NET/ABNB/DKNG) — unofficial.
• AppLovin Q3 IR (2026-11-04).
• Cloudflare Q3 PR (2026-11-05).
• Airbnb Q3 Newsroom (2026-11-05).
• DraftKings Q3 IR (2026-11-05).
• Atlassian FQ1 FY27 Business Wire (2026-11-05) — primary; this-hour fetch timed out; numbers from prior verified KB + matching secondary.

TRAI strategy insights — living document

All items are PROPOSALS. Nothing here is built into the engine until the 6 PM refinement session accepts it. Each proposal links to the report with its evidence. Screening/research, not investment advice.

Status legend: PROPOSAL (new) · ACCEPTED (built) · REJECTED · NEEDS DATA.

S-001 — Use correlation clusters, not sector labels, for the "1 per cluster" cap · PROPOSAL

• Rule: when enforcing "max 1 position per correlated sector cluster", group names by TRAI's 60-day correlation clusters (/api/correlations) in addition to sector labels.
• Why: GEV, VRT, ANET, ETN and CAT sit in the same cluster as the semis (cluster 1, 24 names, avg ρ 0.588), but GEV is labelled Industrials so the engine's sector flag ("Semis ×6") did not include it. On Oct 22 GEV fell −10.64% and VST −9.81% alongside LRCX −11.76%.
• Evidence: 2026-10-23 09:00 report §4, §9.2. Also 2026-10-23 10:00 report §4, §8.3: here the sector label over-counts. "Consumer Disc. ×3" groups HD, LOW and AMZN, but HD–AMZN ρ is 0.00. The "Software" label also mixes app software (cluster 6), security/data (cluster 5) and ORCL (semis cluster 1).

S-002 — Suppress or downsize short-momentum entries in transitional + choppy regimes · PROPOSAL

• Rule: when VIX label = transitional and trend = choppy, skip short-momentum entries (or halve risk) unless a stronger filter is met.
• Why: TRAI replay (2026-04-23 → 2026-10-21, next open→close, net): short momentum n = 412, mean −0.46%/trade, t −3.72; transitional n = 329, mean −0.51%, t −3.66; choppy n = 173, hit 38.7%, mean −0.60%, t −2.75. Caveats: daily-bar VWAP proxy, survivorship bias.
• Evidence: 2026-10-23 09:00 report §8, §9.1.

S-003 — After a sector-wide shock, don't hold opposite-side trades in the same cluster · PROPOSAL

• Rule: if the same correlation cluster has names in both short-momentum and fade-long buckets (e.g., KLAC short, AMD fade-long on 2026-10-23), take at most one side.
• Why: with intra-cluster ρ ≈ 0.6–0.9 the two legs largely offset on the factor while each pays costs.
• Evidence: 2026-10-23 09:00 report §3, §9.2. Also 2026-10-23 10:00 report §8.1: ADBE (long_momentum) vs INTU (fade_short), ρ 0.79, same cluster 6. clusterFlag is computed per bucket, so this cross-bucket conflict is not flagged.

S-004 — Extend gap-risk flags to correlated peers of a reporter · PROPOSAL

• Rule: if a name reports BMO today / AMC yesterday, flag same-cluster bucket candidates with ρ ≥ 0.5 as "peer-event risk" (e.g., BX long while AXP reports; KLAC/AMAT/ASML after LRCX).
• Why: LRCX's print moved AMAT −8.16%, KLAC −9.55%, ASML −6.93% (ρ 0.84–0.91). AXP (indicated −2.43%) is in BX's cluster 8 (BX–AXP ρ 0.53).
• Evidence: 2026-10-23 09:00 report §2, §6, §9.4. Also 2026-10-23 10:00 report §3, §8.4: NOW reports 10-28 AMC (avg |gap| 7.58%). Cluster-6 peers ADBE/INTU/CRM/WDAY (ρ 0.57–0.78 to NOW) carry read-through risk that night.
• Evidence add (2026-11-06 11:00): Mon crowded peer night (TTD/HIMS/OXY/MARA AMC + KKR/CEG BMO); OXY ρ≥0.89 to EOG/COP/DVN; KKR–BX 0.85; MARA cluster-8. 2026-11-06-1100-mon-gap-stack-ttd-kkr-hims-oxy-mara-brk §4–5.

S-005 — Don't count on post-earnings drift for day-2 momentum entries · PROPOSAL

• Rule: for long/short-momentum candidates whose signal day was an earnings reaction (BX, SPOT on 2026-10-23), require the normal confirmation and do not add a drift premium.
• Why: TRAI aggregate (n = 622 earnings): avg drift days 2–5 in gap direction +0.02%; avg day-1 in gap direction −0.16%; continuation 49%.
• Evidence: 2026-10-23 09:00 report §5, §9.4. Also 2026-10-23 10:00 report §3, §8.5: IBM is a long_momentum candidate on day 2 after its 10-21 AMC print. Its last-4 continuation rate is 25% and its avg day-1 move in gap direction is −2.17%. Replay for IBM long_momentum: n 6, mean −0.68%.
• Added evidence (15:00): TRAI replay, long_momentum where the signal day was the name's own earnings reaction day: n 48, mean −0.56%, hit 43.8%, t −1.19 (net, my calc). BX's own day-2 history (Yahoo, n 17 prints): mean day-2 o→c −0.43%. SPOT after up reactions (n 10): −0.27%. Evidence: 2026-10-23 15:00 report §2.6, §3.3, §4.

S-006 — Big earnings gaps rarely fill: avoid fading them same-day · PROPOSAL

• Rule: exclude mean-reversion fades on names with an earnings gap > 2 ATR on the reaction day.
• Why: TRAI big-gap subset (n = 172, median |gap| 8.75%): fill rate 2.3%.
• Evidence: 2026-10-23 09:00 report §5.
• Evidence add (2026-11-06 11:00): TTD last-4 avg |gap| 13.56% (2.58 ATR), fill 25%, max 27.16% — mega-gap class for Mon reaction. 2026-11-06-1100-mon-gap-stack-ttd-kkr-hims-oxy-mara-brk §4.

S-007 — Hold first entry until scheduled 10:00 ET macro prints clear · PROPOSAL

• Rule: on days with a 10:00 ET release (e.g., UMich final, Oct 23), push first entry to after the print (e.g., 10:05).
• Why: the 10:00 release lands exactly when TRAI's no-entry window ends; flash PMIs also at 9:45.
• Evidence: 2026-10-23 09:00 report §7, §9.7.

S-008 — Earnings-calendar completeness check · PROPOSAL (data quality)

• Rule: after the 7:30 AM events job, list universe names with next = null and backfill from a second public calendar.
• Why: WDC (reported 10-22 AMC, ~−2% premarket) and NEM (reported, lower premarket) had no next date in /api/events, so neither was flagged.
• Evidence: 2026-10-23 09:00 report §6, §9.6. Also 2026-10-23 10:00 report §3: IBM has next = null after reporting 10-21. INTU's 11-19 date is an estimate; Intuit says the date will be announced at quarter end.

S-009 — Add a macro/FOMC calendar to the event-risk layer · PROPOSAL (data quality)

• Rule: load scheduled macro events (FOMC decisions, GDP, PCE, CPI, ECI, payrolls) next to earnings in /api/events. Flag any day where an FOMC decision falls on the same date as earnings from cluster members.
• Why: /api/macro has no calendar feed. Next week stacks the FOMC decision (Wed Oct 28) on MSFT/GOOGL/KLAC AMC and VRT/GEV BMO. GDP + PCE follow on Thu Oct 29 8:30 and ECI on Fri Oct 30 8:30 (Fed, BEA, BLS schedules). The research partner (whiggity) missed the FOMC meeting entirely.
• Evidence: 2026-10-23 09:00 report, whiggity consult, items 5–6.

S-010 — Customer→supplier catalyst map for peer-event risk (complements S-004) · PROPOSAL

• Rule: keep a small hand-built map (hyperscalers MSFT/GOOGL/AMZN/META/ORCL → cluster-1 suppliers). When a mapped customer reports, flag cluster-1 bucket candidates as "peer-event risk" even when the return correlation is low.
• Why: hyperscalers barely co-move with semis on daily returns (avg 60-day ρ to core semis: MSFT 0.33, AMZN 0.17, GOOGL 0.07, META 0.05; TRAI /api/correlations). A ρ ≥ 0.5 rule (S-004) would never connect them, yet META's capex/financing print helped drive the Oct 22 semis selloff.
• Evidence: 2026-10-23 09:00 report §2, whiggity consult ("Checked against TRAI data").

S-011 — Test the sector-shock next-day rule before using it (whiggity H1) · NEEDS DATA

• Rule to test: after SMH < −5% with VIX ≥ 19 and the 10y down > 5 bp, measure next-day open→close for cluster-1 names (not just SMH), split into gap and intraday parts.
• Why: a quick SMH-only check (Yahoo daily bars, 2010 → 2026-10-21; not TRAI's engine) found n = 16, next-day open→close mean +1.00%, median +0.49%, hit 56%, t ≈ 1.2. That is not significant, it is driven by March-2020 outliers, and most of the edge is in the overnight gap, which TRAI can't hold.
• Evidence: 2026-10-23 09:00 report, whiggity consult, hypothesis 1.
• Added evidence (2026-11-02 12:00): 5-year Yahoo replay, SMH ≤ −5% days (n 23, to 10-29): next-day open→close SMH +0.13% (48% up), ASML −0.07%, AMAT −0.39% (39% up), ANET +0.09%; close→close negative for all four. No day-after bounce in the window TRAI trades; the 2010+ result above depends on March 2020. Proposal: skip or down-weight cluster-1 fade-longs (mean_reversion longs) on the session after SMH ≤ −5% until a cluster-wide test says otherwise. Live case 11-02: SMH +0.04% at 12:10 ET, below VWAP; ASML/AMAT flat, ANET −1.4%. 2026-11-02 12:00 report §2, §5.

S-012 — Residual-reversion replay on TRAI's own residualZ (whiggity H4) · NEEDS DATA

• Rule to test: each day, take the universe names with the most negative residualZ (e.g., ≤ −2) that have no earnings in the last 2 sessions, then score next-day open→close in /api/scorecard, split by VIX label and trend.
• Why: TRAI already computes beta-adjusted residuals (e.g., GEV −3.33, ASML −2.46, KLAC −2.41 on Oct 22), so this can be tested on existing daily bars. The current fade-long bucket (mean +0.165%, t 1.19) may get sharper if no-news residual names are separated out.
• Evidence: 2026-10-23 09:00 report, whiggity consult, hypothesis 4; §8.

S-013 — Check that the driver still holds before fading a macro-driven move · PROPOSAL (NEEDS DATA)

• Rule: when a mean-reversion candidate's prior-day move came from a macro factor rather than company news (e.g., HD/LOW/TGT on a 10y yield drop), check that factor's move since the prior close and any 9:45/10:00 ET prints before entry. Skip the fade if the factor has already reversed and the name has moved ≥ 1 ATR toward the fade target before 10:00, because the reward left is below the 2-ATR target and the entry is late.
• Why: on Oct 22, HD +4.33%, LOW +3.97% and TGT +3.76% came with no company news as the 10y fell 17 bp. By 10:14 ET Oct 23, after hot flash-PMI prices and UMich inflation expectations, the 10y was ~4.99% and all three were −2.7% to −2.8% (about 1.1–1.2 ATR) before TRAI's first allowed entry window. Replay shows no fade-short edge in transitional regimes (n 263, −0.05%, t −0.34). The skip threshold needs a backtest.
• Evidence: 2026-10-23 10:00 report §5, §6, §8.3.

S-014 — Net cluster exposure across buckets, including negatively correlated hedges · PROPOSAL

• Rule: (a) compute clusterFlag across all buckets, not per bucket. Flag any long/short pair in the same correlation cluster (ties into S-003). (b) Before entry, compute the book's net exposure to the core-semis factor. Treat cluster-6 app-software longs (avg ρ to core semis −0.22 to −0.41) as partly offsetting semis fade-longs, not as independent bets. (c) Split TRAI's "Software" sector into app (cluster 6), security/data (cluster 5) and AI-infra (ORCL, in cluster 1).
• Why: on Oct 23, ADBE (long) and INTU (fade-short) sit in the same cluster at ρ 0.79, and the per-bucket flag misses it. On Oct 22 the "Software" label split cleanly: app software up (ADBE +6.07, INTU +5.75), AI-infra software down (ORCL −2.78, DDOG −5.62). At 10:14 ET Oct 23, SMH +0.86% vs ADBE −1.40%: the inverse factor relationship held intraday.
• Evidence: 2026-10-23 10:00 report §1, §4, §8.1–8.2.

S-015 — Rotation-dispersion day: t+1 gap vs open→close for the software winners (whiggity H1/H3) · NEEDS DATA (open research question)

• Question: after a day when semis crash and app software rallies, does the software winners' t+1 move split into a give-back gap and an intraday recovery, or (whiggity's claim) does the move extend into the t+1 open and reverse by ~11 AM ET? Define the event with an app-software basket (cluster 6) or IGV − SOXX. Do not use "IGV > +2%": IGV was −0.09% on Oct 22, so that rule misses the very day that prompted the question. Extension (whiggity H3): when a large share of the software gain came in the last hour with MOC imbalances, test fading the t+1 morning VWAP breach. Both whiggity's "> 45%" threshold and the 11 AM timing are unsourced.
• Why: a quick daily-bar check (mine, Yahoo, 2010 → 2026-10-22; not TRAI's engine) with SOXX < −4% and IGV − SOXX > 5 pp gives n = 16, 13 of them in 2026 (one regime, overlapping). Next-day gaps averaged −0.3% to −0.7% (IGV/ADBE/INTU/CRM/NOW). Open→close averaged +0.8% to +1.5% (IGV t 3.4, NOW t 3.2). That is the opposite of whiggity's intraday claim, but it rests on daily bars and a small, clustered sample. Oct 23 so far breaks the pattern (gap down, then further weakness by 10:25 ET). Needs intraday bars and an out-of-sample split. Not for use until tested.
• Evidence: 2026-10-23 10:00 report, whiggity consult, hypotheses 1 and 3 and conflict 5.

S-016 — Split rotation days by whether defensives/housing also rallied with yields falling (whiggity H2) · NEEDS DATA (open research question)

• Question: on semis-crash days, does the next-day momentum-vs-fade outcome differ between Regime A (app software beats SPY, defensives/housing flat) and Regime B (app software and HD/LOW/TGT each beat SPY by > 2.5 pp while SOXX < −4%)? Control for the same-day and next-morning 10y change. whiggity's hypothesis: Regime B is macro/duration caution, so fade; Regime A is real software preference, so momentum.
• Why: Oct 22 fits Regime B only with an app-software basket. HD/LOW/TGT beat SPY by ~5 pp, but IGV beat it by only ~1.1 pp. On Oct 23 both groups reversed as the 10y rebounded to ~4.99% after hot PMI prices. This builds on S-013 (check that the macro driver still holds before fading). Untested.
• Evidence: 2026-10-23 10:00 report §5, whiggity consult, hypothesis 2 and counter-evidence (b).

S-017 — Pre-earnings sympathy drift after a rotation-day rally (NOW, T-4 → T-0) (whiggity H4) · NEEDS DATA (open research question)

• Question: when a name with earnings within 5 sessions rallies > 2% on a sector-rotation day with no company news, does it drift further into T-0, and does that run-up predict a negative T+1 reaction?
• Why: NOW rose +2.53% on Oct 22, which is T-4 for its Oct 28 AMC print. Its last 4 prints: continuation 75%, but day-1 −0.71% in the gap direction, and two large down-gaps came on EPS beats (TRAI /api/events). TRAI holds nothing overnight, so any rule would be about intraday sizing in the run-up and about peer read-through (S-004), not holding through the print. Use TRAI's stored Yahoo EPS estimates for the surprise variable. Public.com's estimate series doesn't match them (consult, "Partly verified").
• Evidence: 2026-10-23 10:00 report §3 (NOW), whiggity consult, hypothesis 4.

S-018 — News-driven shocks to defensive clusters (telecom, Oct 23) · PROPOSAL

• Proposal: when a defensive cluster (e.g., cluster 10: T, VZ, TMUS, CMCSA) falls > 1.5 ATR on non-earnings competitive news (here, SpaceX's tower talks), (1) drop that cluster from defensive-hedge netting in S-014 for the session, and (2) don't open same-day mean-reversion fades in it until the news has been checked against primary sources. Also flag when the mean-reversion longs (semis) and the defensive shorts are winning together: that is one anti-defensive factor bet, not two positions.
• Why: on Oct 23 the telecom names fell 2–6% while SMH rose 1.1%. Telecom's normal role as a semis hedge (sector ρ −0.27) broke on a story about competitive structure, not on rates or earnings. If a rescan flags T or TMUS as short_momentum, TRAI's replay for that bucket in VIX-transitional regimes is weak (n = 329, mean −0.51%/trade, t −3.66; report §6.2).
• Evidence: 2026-10-23 11:00 report §2, §6.8.

S-019 — Context symbols outside the universe (towers / SpaceX) · PROPOSAL

• Proposal: add CCI, AMT and SBAC (and the SpaceX ticker once it lists) to the pipeline as non-tradable context symbols: fetch daily and intraday bars and show them next to cluster 10 on the dashboard. They are the mirror image of carrier moves on spectrum and tower news. Context only, no signals.
• Why: the strongest move today (CCI +10%, residual z +5.0) was invisible to TRAI because CCI isn't in the universe.
• Evidence: 2026-10-23 11:00 report §2, §6.5; towers.

S-020 — Residual z for context names vs SPY and Δ10y (whiggity 8a) · NEEDS DATA (open research question)

• Question: does a context symbol's residual z (252-day OLS on SPY and the Δ10y) above ±3 on a news day predict the next-day move in the tradable peers (e.g., CCI z +5 → T/VZ/TMUS t+1)?
• Why: quick check (mine, Yahoo daily) for Oct 23: CCI +5.0σ, AMT +3.1σ, SBAC +2.2σ. That's n = 1, with no history of tower–carrier inverse shocks tested yet. Needs S-019 data first.
• Evidence: 2026-10-23 11:00 report §8 (quick checks).

S-021 — Pair-spread shocks before a peer's print (T−VZ, whiggity 8b) · NEEDS DATA (open research question)

• Question: when an intra-cluster pair spread hits |z| > 3 (252-day) within 2 sessions of one member's earnings, does it mean-revert by the close of the print day (t+1)?
• Why: T−VZ was −3.63 pp (z −3.17) on Oct 23, with VZ reporting Mon 10-26 BMO. In the last year there were 6 prior |z| > 3 days, all near earnings. On the 4 days T lagged, the next-day spread was +1.31/+0.14/+1.80/+0.02 pp (n = 4, anecdote). TRAI holds nothing overnight, so any rule would be an intraday-only spread or a sizing note.
• Evidence: 2026-10-23 11:00 report §6.3, §8.

S-022 — "Quiet open, strong by 11 AM" earnings continuation, plus Friday cyber hold with 10y ≥ 5% (whiggity 8c/8d) · NEEDS DATA (open research question)

• Question (c): when an earnings name opens with less than half its indicated gap (SLB: +0.83% vs +2.00%) and is above +2 ATR by 11:00 ET, does it hold into the close more often than its gap-fill history suggests? (d): does a high-beta cyber basket (cluster 5) that's up > 4% at midday on a Friday with the 10y ≥ 5% give back more than half of the gain by the close?
• Why: SLB broke its 75% fill history today. PANW/NET/CRWD are +4.7% to +5.8% with the 10y at ~4.99%. Both questions need intraday bar history that TRAI doesn't store yet. Not tested.
• Evidence: 2026-10-23 11:00 report §3, §5, §8.

S-023 — clusterFlag from correlation clusters, not sector labels · PROPOSAL

• Proposal: compute clusterFlag (and the "1 per correlated cluster" cap) from the 60-day correlation clusters in /api/correlations (or a pairwise ρ ≥ 0.6 rule), not from the TRAI sector label. Extends S-001.
• Why: in today's short_momentum bucket, ANET/VRT are flagged "Hardware ×2" and six semis "Semis ×6", but GEV and VST have no flag — even though GEV is in cluster 1 with ANET/VRT/semis (GEV–VRT 0.73) and VST is ρ 0.60 to GEV. The reverse also happens: ISRG and JNJ are flagged "Healthcare ×2" in mean_reversion while their ρ is only 0.33 and they sit in different clusters. Sector-label intra-ρ for Industrials is 0.14 — the label carries little information.
• Evidence: 2026-10-23 12:00 report §2.5, §4; AI power.

S-024 — Out-of-universe peer prints as context for the AI-power cluster (BE, PWR, FIX) · PROPOSAL

• Proposal: add BE, PWR and FIX as non-tradable context symbols (same mechanism as S-019) and show their earnings dates and reactions next to cluster 1 / cluster 18 on the dashboard during earnings season. Context only, no signals.
• Why: Bloom Energy fell −11.5% on Oct 23 after a beat-and-raise (FY26 revenue guide raised to $4.4–4.6B) because of −$181M FCF and ~4× capex; it is the closest public read-through before GEV and VRT report Wed Oct 28 BMO. TRAI could not see it.
• Candidate additions (whiggity, 12:00 consult): TLN, NRG (VST peers, 60d ρ 0.81/0.79); NVT, MOD, TT (GEV/VRT peers, ρ 0.56–0.88); CIEN, COHR, FN (ANET peers, ρ 0.62–0.71); SCCO (FCX, ρ 0.94). Co-movement is my Yahoo calc; whiggity's mechanisms (PJM auctions, FERC co-location, spark spreads) are unverified.
• Evidence: 2026-10-23 12:00 report §2.3, §2.4, §8.

S-025 — Classify the catalyst before a mean-reversion fade · NEEDS DATA (open research question)

• Question: do fades on names whose prior-day move had a company-specific catalyst (earnings, clinical data, guidance) perform worse than fades on news-less moves? Requires a catalyst tag per fade candidate (earnings date from /api/events plus a news/8-K check).
• Why: today's fade_short JNJ is fading a clinical-data move (VARIPULSE Pro / Omny-IRE); ISRG and TMO sit in buckets two to three days after their prints; HON's long_momentum is an earnings-day signal (S-005). JNJ's fade is working today (−1.74%) and ISRG's is not (+2.59%) — n = 2, no conclusion. Replay per name: JNJ mean_reversion n4 +0.01%, ISRG n3 +0.80%.
• Evidence: 2026-10-23 12:00 report §3.3, §4, §5.

S-026 — Corporate-action flag for earnings-reaction history (HON spin) · PROPOSAL

• Proposal: record spin-offs, mergers and reorganisations per symbol, and compute "typical reaction" stats in /api/events only from events after the latest corporate action (or show a warning badge when fewer than 4 post-action events exist).
• Why: Honeywell Aerospace was spun off on June 29, 2026; two of HON's last-4 events (Jan, Apr 2026) belong to the pre-spin conglomerate, so HON's avg |gap| 2.59% / 25% fill mixes two companies. HON also shows next = null (S-008).
• Evidence: 2026-10-23 12:00 report §3.3; HON.

S-027 — "Already round-tripped" filter for short_momentum · NEEDS DATA (open research question)

• Question: when a short_momentum candidate is back above its prior close (or has retraced more than half of its opening drop) by 10:00 ET or later, does a subsequent short have negative expectancy? Also: does the day after an AI-power basket drop of more than 5% systematically favour longs?
• Why: GEV/VST/VRT made their lows in the 9:35 bar (inside the no-entry window), were −2.0% to −2.5% at 10:00 and flat by noon; ANET +2.1%. My daily check (Yahoo, 2024-03 → 2026-10): after 24 basket drops of more than 5%, next-day open→close averaged +0.82% (median +0.47%, t 1.16) — not significant, overlapping events. Needs intraday bar history (same dependency as S-022).
• Evidence: 2026-10-23 12:00 report §2.2, §5, §6.3.

S-028 — Pre-earnings short-momentum exhaustion (whiggity H1) · NEEDS DATA (open research question)

• Question: for a short_momentum name 3–5 sessions before its own print (T−5..T−3) whose setup day closed < −7% with close location (close−low)/(high−low) < 0.20, does price continue lower from the open to 11:00 ET and then revert from 11:30 to the close?
• Why: whiggity proposed this for GEV/VRT (GEV and VRT print Wed 10-28 BMO). On Oct 23 (n = 1, Yahoo 5m, my calc) GEV qualified (−10.64%, close location 0.16) and VRT just missed (0.23). The continuation leg did not show up: GEV was −0.12% open→11:00, and its low came in the 9:35 bar, inside the no-entry window. It was +0.28% from 11:30 to 12:14. Same round-trip as S-027, so test the two together. Needs intraday bar history (S-022 dependency).
• Evidence: 2026-10-23 12:00 report §8 (quick checks), §6.3.

S-029 — Day-2 sympathy losers vs the earnings loser, beta-adjusted (whiggity H2) · NEEDS DATA (open research question)

• Question: after a reporter's earnings drop drags a correlated non-reporting basket (LRCX → GEV/VST/VRT), does the basket minus β·reporter spread mean-revert positively over the next 24–48 h when day-1 dispersion was low?
• Why: on Oct 22, raw dispersion was low (basket −9.42% vs LRCX −11.76%), but the basket fell about 2× what its 60-day β to LRCX (0.37) implies. Define "dispersion" before testing. As of 12:14 ET Oct 23 the spread was −0.41 pp (LRCX +1.42% led, with SMH +1.36%), which is slightly against H2. The window is incomplete, n = 1. Daily bars are enough for the 24/48 h version.
• Evidence: 2026-10-23 12:00 report §8; 09:00 report §1.

S-030 — Intraday multi-factor residual fade (CAT on FCX, GEV, XLI) (whiggity H3) · NEEDS DATA (open research question)

• Question: regress a name's 5-minute returns on 2–3 factor proxies (for CAT: FCX for mining, GEV for AI power, XLI for the sector). Does fading the cumulative intraday residual beyond a threshold pay by the close? This is the intraday version of S-012/S-020.
• Why: on Oct 23 CAT was −2.08% at 12:14 with FCX +1.55%, GEV +0.01% and XLI −1.09%, so its residual was clearly negative. CAT kept falling (open→12:14 −0.71%), and a fade would have lost (n = 1). CAT reports Thu 10-29 BMO, so pre-print drift may dominate. Needs intraday history.
• Evidence: 2026-10-23 12:00 report §3.1, §8.

S-031 — Friday afternoon after a large post-print pair divergence (HON−CAT) (whiggity H4) · NEEDS DATA (open research question, low priority)

• Question: after a Thursday where R_HON − R_CAT > +6 pp (HON's own print), does the spread compress or continue from 12:00 to 4:00 PM Friday? whiggity filters this by HON vs XLI over the first hour.
• Why: the condition was met on Oct 22 (+6.56 pp). The first-hour filter was neutral (HON −0.97% vs XLI −0.93%). The 12:00 baseline was HON +0.44% and CAT −1.69% (spread +2.13 pp); score it at the close. HON and CAT are uncorrelated (ρ 0.19) and HON's history mixes pre- and post-spin events (S-026), so the sample per quarter is tiny. Generalize to "post-print divergence vs an uncorrelated sector peer" before testing.
• Evidence: 2026-10-23 12:00 report §3.3, §8.

S-032 — "Low-quality beat" continuation on BMO prints · NEEDS DATA (open research question)

• Question: when a BMO print beats EPS but misses revenue (or a key volume line) and the beat is helped by tax, provisions or one-off gains, does the opening gap continue rather than fill on day 1?
• Why: AXP 2026-10-23: adj EPS $4.68 vs ~$4.51, revenue $20.06B vs $20.12B, tax rate 18.1% vs 24.1%, $982M one-off gain in GAAP; opened −3.12%, −4.15% at 13:07, no fill. AXP has filled 0/4 earnings gaps, and 3 of its last 4 prints were "beat, sell". Requires revenue surprise and a "beat quality" field, which /api/events does not store yet.
• Evidence: 2026-10-23 13:00 report §2, §6.

S-033 — Card networks the session after an issuer's negative earnings reaction · NEEDS DATA (open research question)

• Question: after AXP or COF closes down on its earnings day, do V and MA outperform SPY the next session?
• Why: on the 4 AXP reaction days since Oct 2025, V and MA rose every next session (avg ≈ +1.6%; one day confounded by V's own print). n = 4, anecdotal. Today's AXP print gives a live check on Mon 10-26 (V reports 10-27 AMC, so 10-26 is clean).
• Evidence: 2026-10-23 13:00 report §3.

S-034 — Sanitize EPS-surprise % when estimates are near zero · PROPOSAL (data fix)

• Rule: in /api/events, blank or cap surprisePct when |epsEstimate| < $0.05 or actual and estimate have opposite signs, and exclude those events from beat/miss aggregates.
• Why: COIN shows −364.89%, −3458.88% and −487.47% surprises, which would distort the "beats" vs "misses" aggregates.
• Evidence: 2026-10-23 13:00 report §4 (COIN); COIN.

S-035 — Per-name gap-fill tendency as an earnings-day filter (PYPL vs V) · PROPOSAL

• Rule: on a name's earnings reaction day, don't take a fade of the opening gap when its last-4 fill rate is 0% and avg |gap| ≥ 2 ATR (e.g., PYPL: 9.43%, 3.42 ATR, 0/4 filled); allow fades only when fill rate ≥ 75% (e.g., V: 75% fill, 25% continuation). Applies first to PYPL (10-27 BMO) and V (reaction 10-28).
• Why: per-name history is tiny (n = 4), so treat as a tie-breaker until pooled across the universe (cf. /api/events aggregate fill rate 38.7%, big gaps 2.3%).
• Evidence: 2026-10-23 13:00 report §4, §6.

S-036 — V → MA earnings lead-lag (whiggity) · NEEDS DATA (open research question)

• Question: does MA's earnings-day gap (Thu BMO) load on V's post-print day return minus MA's same-day sympathy return (Wed)? If MA under-reacts on Wednesday, the residual should predict Thursday's gap. Also check whether 8:30 ET macro (Q3 GDP + September PCE on 10-29, per BEA) swamps MA's own print.
• Why: V–MA ρ 0.88; V reports 10-27 AMC (fiscal Q4 + initial FY27 guide), MA 10-29 BMO. Daily bars suffice; ~16 quarters of history available.
• Evidence: 2026-10-23 13:00 report §4, §7.

S-037 — BTC overnight return as a filter on HOOD/COIN post-print trends (whiggity) · NEEDS DATA (open research question)

• Question: for HOOD and COIN AMC prints, does BTC's return from 4 PM ET to 9:30 AM ET predict the reaction day's first-hour trend and open→close, beyond the EPS/revenue surprise sign?
• Why: HOOD–COIN ρ 0.73 (cluster 9 with MSTR/MARA/RIOT). HOOD's reactions were negative c-c in 3 of 4 events despite EPS beats; the crypto tape may matter more than the print. HOOD reports 10-27 AMC, COIN 10-29 AMC.
• Evidence: 2026-10-23 13:00 report §4, §7.

S-038 — Reaction-day risk sizing off the opening range, not ATR(14) alone · PROPOSAL

• Rule: on a name's own earnings reaction session, set the stop distance to max(1×ATR14, 0.5×|opening gap|, first-30-min range). Keep 0.25% account risk, so the share count shrinks accordingly. Apply first to GOOGL and MSFT (Thu 10-29), AMZN (Fri 10-30) and AAPL (Tue 11-03).
• Why:
• TRAI replay (2026-04-23 → 10-21, net) shows reaction-day trades at n 54, mean −0.15%, hit 44.4%, sd 4.34% vs 2.46% for non-event trades (n 1,009).
• Mega-cap gaps average 2.5–3.5 ATR. With a pre-event ATR stop, that variance gets stopped out at random.
• Caveats: daily-bar VWAP proxy, and only the last 4 events per name.
• Evidence: 2026-10-23 14:00 report §3, §4.3.

S-039 — Direction-agnostic handling of supplier read-through on hyperscaler reaction days · PROPOSAL (refines S-010 / S-004)

• Rule: S-010's customer→supplier map already flags suppliers. This adds that on a hyperscaler or META reaction session (MSFT, GOOGL, AMZN, META), flagged cluster-1 AI-hardware names (NVDA, AVGO, VRT, ANET, ORCL, AMD, TSM, ETN, GEV, VST…) get no direction assumption from the reporter's gap and no mean-reversion fades until the first 30 minutes set the SMH direction. Next live cases: Thu 10-29 (GOOGL/MSFT) and Fri 10-30 (AMZN).
• Why:
• 60d ρ between hyperscalers and AI hardware is mostly < 0.35 (GOOGL–NVDA 0.12, MSFT–NVDA 0.22, AMZN–VRT 0.32), so S-004's ρ ≥ 0.5 test misses the link.
• Yet reaction days moved suppliers ±5–10%: 2026-02-06 AMZN −5.55% with VRT +10.03% and NVDA +7.87%; 2026-07-30 MSFT +15.51% with ORCL +8.34% and SMH +6.88%; 2026-10-22 META −6.98% with VRT −7.81% and SMH −5.99%.
• The read-through sign is not the reporter's sign.
• Evidence: 2026-10-23 14:00 report §4.1–4.2; megacap-cloud.
• Added test (whiggity consult, NEEDS DATA): after an SMH 5-day selloff of ≥ 5%, do NVDA/AVGO/VRT morning sympathy gaps on GOOGL/MSFT reaction days revert or trend into the close? Quick look (Yahoo daily; n = 2): on 2026-02-05 (SMH 5d −8.31%) all three reverted; on 2026-07-30 (SMH 5d −14.09%) NVDA and AVGO trended while VRT reverted (+6.35% gap → −4.09% o→c). Mixed. whiggity's "capex divergence vs capex exhaustion" labels fit some past days, but several days fit neither and no pre-print predictor of the regime has been shown. That leaves this rule's no-direction default unchanged. SMH 5d to 10-22 was −6.35%, so the condition may be live on 10-29. Evidence: 2026-10-23 14:00 report §9.

S-040 — Flag EPS surprises dominated by non-operating marks or one-off charges · PROPOSAL (data fix; extends S-034)

• Rule: in /api/events, mark surprisePct as "non-comparable" when GAAP EPS includes a disclosed investment gain or legal charge larger than ~25% of net income. Exclude those events from beat/miss aggregates, or use the company's ex-item EPS when it is disclosed.
• Why:
• GOOGL shows +91.59% and +214.23% surprises (the Q2 2026 EPS of $9.11 includes a $98.0B net gain on equity securities).
• AMZN shows +68.18% and +214.89% (Q2 included a $53.4B pre-tax gain, mainly Anthropic; consensus excludes those marks).
• META shows −29.91% for Q3 ($3.83 GAAP includes a $9.89B legal charge; $7.40 ex-charge).
• These are among the largest names in the universe, so they distort the surprise→reaction stats.
• Evidence: 2026-10-23 14:00 report §3.1, §3.3, §3.5.

S-041 — Crowded-night rule: treat same-night mega-cap reporters plus macro as one event · PROPOSAL

• Rule: when ≥ 2 universe mega-caps report on the same night, or a reaction session coincides with 8:30 ET tier-1 macro or an FOMC day, treat the session as one event. Allow at most one position among the reporters and their ρ ≥ 0.5 peers, and no long/short pairs between reporters. First live case is Thu 10-29: GOOGL and MSFT reactions, Q3 GDP and September PCE at 8:30, MA/CAT BMO, AMZN/INTC AMC.
• Why: same-night reporters diverge heavily (2026-04-30: GOOGL +9.96%, MSFT −3.93%, META −8.55%; 2026-01-29: MSFT −9.99%, META +10.40%). A "pair" is therefore two independent ~2.5-ATR bets, not a hedge. Anecdotal (n ≈ 4 crowded nights).
• Evidence: 2026-10-23 14:00 report §2, §4.2.

S-042 — Do mega-caps belong in the mean_reversion bucket before their own print? · NEEDS DATA (open research question)

• Question: should mega-caps (GOOGL, AMZN, MSFT, META, AAPL, NVDA…) be excluded from mean_reversion fades, or require a stronger z-score, especially in the 5-session pre-earnings window?
• Why:
• Mega-cap mean_reversion replay: n 18, mean −0.46%, hit 38.9%, t −1.38. Universe pre-earnings mean_reversion: n 27, −0.29%.
• Today GOOGL and AMZN (ρ 0.72, one cluster) are fade-long candidates 3–4 trading days before their prints; neither had bounced by 14:05 ET (−0.66% / −0.58%).
• Small n; needs more months of replay.
• Evidence: 2026-10-23 14:00 report §3.3, §4.3, §6.
• Added test (whiggity consult, NEEDS DATA): compare the mean_reversion hit rate and Sharpe, and the 15-minute and 1-hour return autocorrelation, for GOOGL/AMZN during T−3..T−1 after an SMH ≤ −5% / QQQ ≤ −2.5% shock day against normal weeks. Needs intraday history. Oct 22 met the shock condition (SMH −5.99%, QQQ −2.86%), so the 10-23 → 10-28 sessions are the first live sample (n = 1 episode). Evidence: 2026-10-23 14:00 report §9.

S-043 — AAPL as an anti-AI-capex factor leg · NEEDS DATA (open research question, low priority)

• Question: on AI-capex shock days (SMH |move| > 3%), does AAPL reliably move opposite to cluster 1? Should AAPL serve as a factor control in residual calculations (S-030) rather than as a trade candidate?
• Why: AAPL's 60d ρ is AMZN −0.47, VRT −0.36, ORCL −0.32 and NVDA −0.22. On 2026-10-22 AAPL rose +1.23% while QQQ fell −2.86% and SMH −5.99%. AAPL currently fails TRAI liquidity (ATR% 1.75% < 2%), so this is a research and factor question only.
• Evidence: 2026-10-23 14:00 report §3.4, §4.1; AAPL.

S-044 — Do GOOGL/MSFT overnight gaps predict AMZN's pre-print session? (whiggity) · NEEDS DATA (open research question)

• Question: on AMZN's own report day (the session before its AMC print, which is also the GOOGL/MSFT reaction session), does the average GOOGL+MSFT overnight gap predict AMZN's open→close? Regression: r_AMZN_o2c = a + b1·avg(GOOGL, MSFT gap) + b2·AMZN gap. If b1 ≠ 0 on that day, TRAI could skip AMZN mean-reversion fades or use the read-through as a filter. Next live case: Thu 10-29.
• Why: GOOGL–AMZN ρ 0.72 (cluster 20). Quick look (Yahoo daily; n = 3): AMZN's gap and its o→c matched the sign of the reporters' average gap 2/3 times (2025-10-30 −1.85%, 2026-02-05 −0.99%, 2026-07-30 +0.92%). Too few cases; needs every quarter where AMZN reported the day after GOOGL and/or MSFT (~16+ quarters from daily bars). Not covered by S-041, which is a risk-cap rule, not a directional test.
• Evidence: 2026-10-23 14:00 report §9, §4.1–4.2.

S-045 — /api/options must take earnings timing from /api/events · PROPOSAL (data fix)

• Rule: when /api/options builds its earnings block, take timing and reactionSession from /api/events. Do not default an unknown timing to "next session". Compute impliedVsHist only when the expiry covers ≤ 3 non-event sessions; otherwise leave it blank or label it "includes non-event days".
• Why:
• On 2026-10-22, /api/options listed BX and SPOT with timing "unknown" and reactionSession 2026-10-23. Both reported BMO on 10-22, so today has no event, yet it priced 10-23 "implied moves" of ±2.60% and ±2.68%.
• The BX histAvgAbsGap (1.16%) also disagrees with /api/events (1.66%).
• KKR's impliedVsHist of 3.7× uses a Nov 13 expiry that covers ~15 non-event sessions.
• Evidence: 2026-10-23 15:00 report §5.
• Evidence add (2026-11-06 11:00): options impliedVsHist for Mon stack not re-pulled — do not invent; keep event-timed gate. report §8.

S-046 — Peer-factor residual alongside the SPY residual · PROPOSAL (NEEDS DATA)

• Rule: for each bucket candidate, also compute a residual against SPY plus its correlation-cluster basket (or a sector ETF). For names whose strongest 60d link is negative to cluster 1, use SPY plus SMH. Show "% of move explained by group" on the dashboard. Rank long/short-momentum candidates on the peer residual, or require it to be at least 2.
• Why: BX's Oct 22 SPY-only residual z was 6.16 (TRAI 6.19). Against SPY plus an alt-manager basket it was 3.67, because the whole group rose 3–6%. SPOT went from 4.02 to 3.44 against SPY plus SMH: about a third of its +10.5% was the semis-to-non-AI rotation (SMH −5.99%). The peer basket partly reacts to the reporter's own print, so the true idiosyncratic z lies between the two. This needs a backtest of whether the peer residual ranks next-day o→c better. Related: S-030 (intraday multi-factor residual).
• Evidence: 2026-10-23 15:00 report §2.3, §3.2.

S-047 — Alt-manager context symbols and a crowded peer-print flag for cluster 8 · PROPOSAL (extends S-004 / S-019 / S-024)

• Rule: add ARES, APO, OWL, CG, TPG and BAM as non-tradable context symbols with earnings dates. On any session where ≥ 1 of them reports, flag BX/KKR (and BLK at lower weight) "peer-event risk". When ≥ 2 report on the same morning, treat the session as a crowded night (S-041): at most one cluster-8 position, and no fades of the opening move before 10:30 ET.
• Why:
• Across 13 prints each since 2023-06 (my calc, Yahoo daily), BX's reaction-day sign matched KKR's print 92% of the time, CG's 92% and TPG's 85% (ρ 0.58–0.77).
• Single peer prints don't enlarge BX's move on average (mean |BX| 1.5–2.1% vs a 1.63% baseline).
• Crowded mornings did: 2026-02-05 (ARES −11.2%, KKR −5.4%, TPG −4.6%, OWL −3.6%) saw BX at −5.73%.
• Next live case: Thu 10-29 BMO (ARES, OWL, TPG), which also has GDP/PCE at 8:30, MA/CAT BMO and the GOOGL/MSFT reaction session. Then APO Tue 11-03 BMO (election day), CG Thu 11-05, BAM Fri 11-06, KKR Mon 11-09 BMO.
• Evidence: 2026-10-23 15:00 report §2.4–2.5.
• Added (whiggity, 15:00 §9): add the traded BDCs OBDC, OTF (Blue Owl), ARCC (Ares) and FSK (FS KKR) as context-only canaries for the private-credit sleeve; none is in the TRAI universe. On Oct 22 they rose +2.1% to +4.5% with the group; on 10-23 they were −0.4% to −1.0% vs OWL −3.0% (Yahoo, my calc). Use: a BDC basket drop of more than ~2% pre-10:30 on an alt-manager print day = credit-led move. Unbacktested. Evidence: 2026-10-23 15:00 report §9 item 8.

S-048 — Friday day-2 earnings long_momentum in transitional + choppy regimes · NEEDS DATA (open research question)

• Question: should long_momentum entries be skipped or downsized when (a) the signal day was the name's own earnings reaction day, (b) the trade day is a Friday, and (c) VIX is transitional and the trend choppy?
• Why: TRAI replay (net, my calc):
• Own-reaction-day signals: n 48, mean −0.56%, t −1.19.
• Thursday signal → Friday trade: n 108, −0.16%.
• Thursday signal in transitional + choppy: n 30, −0.75%, hit 36.7%, t −1.69.
• None is significant, and the slices overlap. Today's live cases are BX (+0.32% from the open at 15:12) and SPOT (−0.28%). Score them at the close.
• Evidence: 2026-10-23 15:00 report §4. See also S-005.

S-049 — Factor sign-flip flag (SPOT vs semis) · NEEDS DATA (open research question)

• Question: when a name's correlation to a major factor (here cluster 1 / SMH) flips sign between the long window and the 60-day window, should S-014's net-exposure calculation and the residual model (S-046) use only the recent window, and should the dashboard flag the flip?
• Why: SPOT–SMH daily-return ρ by calendar year (my calc, Yahoo): 2022 +0.67, 2023 +0.33, 2024 +0.33, 2025 +0.40, 2026 YTD −0.17. TRAI's 60d ρ is −0.57 to −0.60 to KLAC/ASML/AMAT. On SMH ≤ −4% days SPOT averaged +2.61% (up 82%, n 11) in 2026, but −2.17% (up 27%, n 51) across 2022–2026 (my calc, Yahoo daily). A SPOT long next to semis short_momentum names is therefore one larger anti-semis bet, not a diversifier. On 10-23 the link was quiet (SMH +0.92%, SPOT +0.22% at 15:12).
• Evidence: 2026-10-23 15:00 report §3.2.
• Added test (whiggity, 15:00 §9): is the negative SPOT–SMH link only a selloff-day effect? Quick look (Yahoo daily, my calc): 2026 ρ on SMH < −2% days −0.22 (n 42) vs −0.10 (n 162) on other days; 2025: +0.18 / +0.29. It is stronger on semis selloffs but not limited to them. Next step: rolling 20/60d conditional beta, and check whether flips persist beyond 4–8 weeks (whiggity's unsourced claim). Evidence: 2026-10-23 15:00 report §9 item 14.
• Added (16:00 report): the same sign flip shows up in staples. XLP–SMH daily ρ: 2022 +0.51, 2023 +0.19, 2024 −0.02, 2025 +0.01, 2026 YTD −0.39; cluster-4 core-semis avg ρ −0.38 to −0.60 (Yahoo, TRAI /api/correlations, my calc). Evidence: 2026-10-23 16:00 report §2.

S-050 — BX vs peer-basket pair after a BX earnings blowout (whiggity) · NEEDS DATA (open research question)

• Question: after a BX print with day-1 residual z > 3 (SPY-only) and the alt-peer basket up > 3% the same day, does long BX / short an equal-weight ARES+OWL+TPG basket, entered at 15:30 ET on day 2 and held through the peers' own T−1 close, earn a positive spread? whiggity's hypothesis (no source): peers underperform BX by 100–250 bp on day 2+.
• Why: on 10-23 (day 2) at 15:21 ET, BX was −1.21% vs a peer average of about −2.0% (~80 bp of relative outperformance). Only 2 prior BX reactions above +4% (2023-01-26, 2024-10-17), both with peers up 1.4–4.9%; BX's day-2 o→c was +2.41% and +0.45%. n is tiny. TRAI can't trade it as built (out-of-universe shorts, multi-day hold), so this is research only. First live case: entry 10-23 15:30, exit Wed 10-28 close (the peers report 10-29 BMO).
• Evidence: 2026-10-23 15:00 report §2.3, §2.6, §9 item 13.

S-051 — Do 8:30 AM macro prints wash out alt-manager earnings alpha in the first hour? (whiggity) · NEEDS DATA (open research question)

• Question: on mornings when alt managers report BMO and a tier-1 macro print lands at 8:30 ET (GDP/PCE/CPI/payrolls), how much of the 9:30–10:30 move in the reporters is explained by the macro factor (SPY/XLF/2y-yield change 8:25→9:35) vs their own earnings? If the macro share is high, S-047's "no fades before 10:30" rule should extend to all cluster-8 entries on those mornings.
• Why: Thu 10-29 stacks ARES/OWL/TPG BMO with advance Q3 GDP and September personal income/PCE at 8:30, the day after the Oct 28 FOMC (market ~92% hold / ~8% hike, Dec hike odds ~78%). The reference case 2026-02-05 (ARES −11.2%, BX −5.7%) was not decomposed. Needs 1-minute bars for the reporters and SPY/XLF/ZT on past co-scheduled mornings.
• Evidence: 2026-10-23 15:00 report §2.5, §9 items 7, 9.
• Added (whiggity, 16:00 §9) → S-057: the same decomposition for MO on Thu 10-29 (release ~7:00, GDP/PCE 8:30). MO trades like a bond proxy, so the pre-8:30 vs post-8:30 split is the first test.

S-052 — Consumer-staples peer-event map with two sub-groups · PROPOSAL (extends S-004 / S-047)

• Rule: on a cluster-4 print day, flag "peer-event risk" only on the reporter's sub-group: tobacco (PM ↔ MO) or food/beverage/HPC (KO, PEP, MDLZ, PG). On a KO print morning, flag PEP, PM and MDLZ; on an MO print morning, flag PM only. Do not fade MO's opening move on its own print day.
• Why: Yahoo daily 2022–2026 (my calc):
• KO prints (n 19): same-sign PM 74%, PEP/MO/XLP 68%; bigger-than-normal moves in PM (1.61% vs 1.11%) and MDLZ (1.37% vs 1.00%).
• MO prints (n 19): only PM follows (74%, ρ 0.54); KO/PG ρ ≈ 0.
• PM prints (n 20): MO 65%; PEP moves opposite (ρ −0.41).
• MO's last 4 gaps averaged 2.73 ATR with 0/4 filled and 75% continuation; KO 2.32 ATR, 0/4 filled (TRAI /api/events).
• Live cases: KO Tue 10-27 BMO; MO Thu 10-29 BMO (same morning as GDP/PCE); MDLZ Thu 10-29 AMC.
• Evidence: 2026-10-23 16:00 report §6.
• Caveat (whiggity consult, 16:00 §9 item 10): the KO→PM same-sign rate (74%) falls to 53% once XLP, DXY and Δ10y are removed (79% with DXY + 10y only), and 3 of the 19 KO dates were also PM print days. Treat PM on KO mornings as sector beta, not a KO-specific read-through. See S-056.

S-053 — Staples fade-shorts after a rotation day, filtered by semis direction · NEEDS DATA (open research question)

• Question: after a day with XLP ≥ +1.5% and SMH ≤ −3%, should staples mean_reversion fade-shorts be taken only when SMH is up premarket?
• Why: 11 such days since 2022 (9 in 2026). Next day, when SMH rose (n 6): staples basket (MO, PEP, KO, PG, PM, MDLZ) o→c −1.03%, XLP −0.51%. When SMH fell (n 5): basket −0.43%, XLP −0.10%. After any XLP ≥ +1.5% day (n 50), XLP's next-day o→c is ≈ 0. 2026-10-23 fits the "SMH up" case (SMH +0.74%, basket −1.6% o→c). Needs premarket SMH data and more episodes.
• Evidence: 2026-10-23 16:00 report §5.

S-054 — Thursday-signal (Friday-trade) mean_reversion fade-shorts · NEEDS DATA (open research question)

• Question: is there a real weekday effect in fade-shorts? TRAI replay (net, my calc): Thursday-signal MR fade-shorts n 99, mean −0.92%, t −4.02; Thursday-signal fade-longs n 28, +0.89%, t 3.05; all MR fade-shorts n 322, −0.26%, t −2.10.
• CORRECTION (17:00 report §5.5): the figures above are sign-flipped for the short side. /api/scorecard replayTrades[].oc is already the signed trade return; the 16:00 calc multiplied by side again. Corrected (net of 5 bp): Thursday-signal MR fade-shorts n 99, +0.76%, hit 69%, t +3.30; Thursday fade-longs n 28, +0.95%, t 3.18; all MR fade-shorts n 322, +0.10%, t +0.77 (matches TRAI's own ranking, expectancyNet +0.097%). Today's +1.31% live result therefore agrees with the replay. The open question becomes: is the Thursday-signal (Friday-trade) fade-short edge real or episode overlap? Friday-signal (Monday-trade) fade-shorts go the other way: n 36, −0.30%; transitional n 24, −0.71%, t −2.48. Evidence: 2026-10-23 17:00 report §5.4–5.5.
• Why it's open: on 2026-10-23 (a Thursday signal) the 8 live fade-shorts made +1.31% gross open→close (MO +1.89, PEP +1.62, TGT +2.66, INTU +2.29, HD +1.65, LOW +1.59, JNJ +0.74, ISRG −1.95). Before acting, check for overlap with specific episodes, survivorship (see the replay assumptions in /api/scorecard), the daily VWAP proxy, and whether the effect is really "Friday" or "post-shock". Related: S-048 (the Friday long_momentum slice).
• Evidence: 2026-10-23 16:00 report §4.

S-055 — Day-2 mean reversion of staples earnings reactions in high-oil regimes (whiggity) · NEEDS DATA (open research question)

• Question: when Brent (or WTI) is above its 1-year 90th percentile on the reaction day, do cluster-4 staples earnings reactions (PG, KO, PEP, MO, PM, MDLZ, KHC; plus CL/KDP as context) mean-revert on day 2 (o→c against the day-1 direction) more than in normal-oil regimes?
• Why: P&G (commodity headwind ~$1.4B, "raw materials, energy and transportation") and PMI ("energy-related input prices") both flagged energy costs this week. PG's day 2 (10-23) gave back −2.70% (o→c −1.76%) with Brent >$116. whiggity's hypothesis (no source): oil above ~$115 feeds resin/surfactant/diesel costs with a 1–2 quarter lag, so markets fade staples beats. Needs Yahoo daily 2015+ for the reporters plus BZ=F/CL=F, and a 1-year rolling percentile.
• Evidence: 2026-10-23 16:00 report §3.1, §9 items 5–6.

S-056 — Re-test "PM follows KO" after controlling for DXY, 10y and XLP (whiggity) · NEEDS DATA (quick look done)

• Question: is PM's 74% same-sign rate on KO print days a KO read-through, or a common factor?
• Quick look (Yahoo daily 2022-01 → 2026-10, n 19, my calc): I regressed on non-print days and compared residual signs on KO print days. Raw 74%. DXY + Δ10y removed: 79% (the dollar is not the driver; PM DXY beta −0.33). XLP + DXY + Δ10y removed: 53% (residual ρ 0.21). Excluding the 3 days that were also PM prints: raw 75%, residual 50% (n 16). The follow is the staples factor.
• Next: extend to all cluster-4 pairs (PEP, MDLZ, MO) and use intraday bars to see whether PM's move on KO mornings starts at KO's ~7 AM release or at the 9:30 open. If the residual stays ~50%, downgrade S-052's KO→PM flag to a plain XLP-beta flag.
• Evidence: 2026-10-23 16:00 report §6.2, §9 item 10.

S-057 — Does 8:30 macro dominate MO's earnings-day opening gap? (whiggity; extends S-051) · NEEDS DATA (open research question)

• Question: on earnings days that coincide with an 8:30 ET tier-1 macro print (GDP/PCE/CPI/payrolls), split the premarket gap into release→8:29 and 8:30→9:30 windows. Does the macro surprise (Δ10y, SPY) explain more of MO's opening gap than the EPS surprise?
• Why: MO releases ~7:00 ET on Thu 10-29; advance Q3 GDP + September PCE come at 8:30. whiggity's claim (unverified): MO trades as a bond proxy, so a hot PCE/10y can override an EPS beat by the open. MO's last 4 gaps averaged 2.73 ATR with 0/4 filled (TRAI). Needs 1-minute premarket bars for MO, PM, XLP, SPY and ZN/ZT on past co-scheduled mornings.
• Evidence: 2026-10-23 16:00 report §6.3, §9 (open questions).

S-058 — US gasoline prices and the PM-minus-MO earnings-week spread (whiggity) · NEEDS DATA (open research question)

• Question: does the 60-day change in US retail gasoline prices (EIA weekly regular, or RBOB as a proxy) predict the PM − MO return spread over MO's earnings week?
• Why: MO is 100% US (discretionary income pressure on adult smokers is cited in its own releases). PM is ~93% international (Q3 US = $845M of $11.4B, PMI release). whiggity's hypothesis (no source): gasoline and FX spikes break the PM/MO "tobacco factor". n is ~19 MO prints since 2022, so treat the result as descriptive.
• Evidence: 2026-10-23 16:00 report §9 items 4, 10.

S-059 — Commodity-driver check before fading energy (and gold-miner) moves · PROPOSAL (extends S-013)

• Rule: skip mean_reversion fade-shorts on cluster-2 energy names (XOM, CVX, COP, OXY, EOG, DVN, MPC, PSX) when the signal-day move came with front-month WTI or Brent ≥ +2.5%, unless crude has traded below the signal-day settle by the 10:00 ET entry check. Mirror for gold miners (NEM): skip fade-longs after a gold-down day unless gold futures are above the prior settle by 10:00. Use the correct contract month (Dec Brent BZZ26 until its 10-30 expiry; Yahoo BZ=F has already rolled).
• Why: TRAI replay (2026-04-23 → 10-21, net of 5 bp, my calc): energy MR fade-shorts in transitional regimes n 22, −0.97%/trade, hit 23%, t −3.28; clustered by day 10 days, −0.94%, t −2.72, 8 of 10 negative. On 7 of those 10 days WTI was up ≥ 2.7%, and on 7 of 10 it rose again the next day. All-regime energy fade-shorts n 29, −0.34%. The long-history (2016+) next-day fade after big oil days is small (COP −0.55%, XOM −0.29% o→c) and ~0 in 2026 (n 12). First live case: XOM/CVX/COP fade-shorts on the 10-23 postclose run, with Brent at a 6-month high and Iran-strike headlines into the weekend.
• Caveats: daily VWAP proxy, survivorship, n small, overlapping names per day. The crude-reversal threshold is untested.
• Evidence: 2026-10-23 17:00 report §3.1, §5.4, §6.

S-060 — Commodity context symbols and peer-event map for energy and gold · PROPOSAL (data; extends S-004 / S-024)

• Rule: add context symbols (not tradable): Brent by explicit contract month with a roll calendar, WTI front month, gold futures, GDX, AEM, B (Barrick). Add peer-event flags: XOM/CVX print → cluster-2 peers (COP, OXY, EOG, DVN; ρ 0.75–0.86); AEM / Barrick print → NEM; FCX ↔ NEM (cluster 16).
• Why: XOM and CVX both report Fri 10-30 BMO (the same day Dec Brent expires). AEM reports Wed 10-28 after the close and Barrick on 11-09, but neither is in the universe, so TRAI can't flag NEM. AEM print days lift NEM's |move| (2.20% vs 1.64%), but residual ρ is −0.13, so this is a sector-factor flag, not a read-through.
• Evidence: 2026-10-23 17:00 report §4.5, §5.2, §5.4.

S-061 — Tag missed earnings reactions before fading them (NEM case) · PROPOSAL (data quality; extends S-008 / S-025)

• Rule: when /api/events has next = null or no event for a name that moved > 1.5 ATR with relVol > 1.3, check a second public calendar or the company's IR/8-K for a print in the prior 24 hours. If one is found, tag the signal as an earnings-reaction fade (apply S-006 and S-025) and show it on the dashboard.
• Why: NEM reported Q3 on 10-22 after the close (beat; Q4 AISC guided to $1,900/oz; downgrade the next morning) and fell −4.59% on 10-23. /api/events still has NEM at n 3, next = null, so the postclose run treated it as a plain fade-long. NEM's history after ≤ −3% reactions (n 15) shows a flat next day (+0.63% o→c, 53% up).
• Evidence: 2026-10-23 17:00 report §4.1, §5.4. First flagged in S-008.

S-062 — Oil-up / gold-down "double squeeze" on gold-miner fade-longs · NEEDS DATA (open research question)

• Question: on days when WTI rises ≥ 2% and gold falls ≥ 1% (as on 10-23), do gold-miner fade-longs (NEM, plus GDX/AEM as context) underperform the next session vs other fade-longs?
• Why: NEM's own guidance quantifies both legs: Brent ±$10 → $110M of cost, gold ±$100 → $505M, against a $70 Brent / $4,500 gold plan (Newmont, 2026-10-22). GDX–WTI ρ turned negative in 2026 (−0.22). TRAI's NEM/FCX fade-long sample is too small (n 6) to answer it; needs Yahoo daily 2016+ for NEM/GDX/AEM with CL/GC futures.
• Evidence: 2026-10-23 17:00 report §3.2, §4.1, §5.4.

S-063 — Gold miners that beat but sell off > 4%: GLD-hedged t+1..t+5 drift · NEEDS DATA (open research question; from whiggity, §10)

• Question: after a gold miner beats EPS but closes the reaction day down > 4%, does the stock outperform GLD from the day-0 close to t+5? If so, add a small score to a fade-long when the gold factor has stabilised (and require S-059's commodity check).
• Why: NEM on 10-23 beat EPS (and revenue) yet fell −4.59%. NEM-only quick look (Yahoo 2016–2026, my calc): n 5, mean +2.05%, median +0.23%, 3 of 5 up. Too small. Pool NEM, AEM, B, WPM, FNV (GDX top 5) for n, and control for gold's day-0 move.
• Evidence: 17:00 report §10 (Q1), §4.1.

S-064 — 52-week high in the 5 sessions before an earnings print → reaction-day gap · NEEDS DATA (open research question; from whiggity, §10)

• Question: across the TRAI universe, do names that print a 52-week high within 5 sessions before earnings gap differently (sign, |gap|, continuation) than other prints? If so, add an "extended into print" flag to the event layer (/api/events).
• Why: CVX made its highest daily high in Yahoo data since 2016 on 10-23 ($233.74), five sessions before its 10-30 BMO print. CVX/XOM quick look: CVX n 4 (gaps +0.84, −3.97, +1.51, −1.23%), XOM n 7 (mean −0.38% vs all-print −0.43%). No pattern at this n. A universe-wide test gives n in the hundreds.
• Evidence: 17:00 report §10 (Q3, item 3).

S-065 — Oil-factor model for cluster 2: WTI, crack and SPY betas; COP − XOM residual spread · NEEDS DATA (open research question; from whiggity, §10)

• Question: a rolling 60-day regression of cluster-2 names on ΔWTI, Δ3-2-1 crack and ΔSPY. (a) Does it explain fade-short losses better than WTI alone (S-059)? (b) Does a COP − XOM residual spread beyond 2σ mean-revert net of costs?
• Why: E&Ps carry a higher crude beta than the integrateds (2026: OXY 0.41, COP 0.34 vs XOM 0.28, CVX 0.25), and integrateds are exposed to refining margins. On 10-23 the front-month 3-2-1 crack fell ~12% while WTI rose +3.99%, and XOM/CVX lagged COP/DVN. Needs a contract-aligned crack series (RB/HO/CL with a roll calendar, see S-060). TRAI does not currently trade pairs, so (b) is research only.
• Evidence: 17:00 report §10 (items 6, 10; Q4).

S-066 — Day-after-big-up guard for long_momentum (≥ +10% day or earnings-reaction day) · PROPOSAL (extends S-005)

• Rule: when a long_momentum candidate's signal day was a ≥ +10% close-to-close move or an earnings-reaction day, don't buy at the open. Require opening-range confirmation (hold above the 30-minute OR high, or VWAP at 10:00), or take half size.
• Why: across 17 AI-infra names (Yahoo daily 2016 to 10-29, my calc), the session after a ≥ +10% day averaged −0.27% open→close (n 322). Since 2024: −1.00% (n 161, 40% up, t −2.68), with day 2 at +0.47%. TRAI replay (04-23 → 10-21, net of 5bp + impact): long_momentum on earnings-reaction-day signals averaged −0.56% (n 48, hit 44%, t −1.19) vs +0.13% for non-event signals (n 337). On 10-29, ORCL (day after +10.73%) lost −3.17% open→close, while VRT (+3.36%) and STX (+1.67%) held. A filter, not a ban.
• Caveats: the Yahoo base rate is close-to-close-triggered and ignores TRAI's other filters. It overlaps S-005, so test both together.
• Evidence: 2026-10-29 16:00 report §5, §6.2.

S-067 — After-hours peer-print gap check for cluster-1 longs at 9:25 · PROPOSAL (extends S-004)

• Rule: when a cluster-1 member or a hyperscaler customer (INTC, AMZN, GOOGL, MSFT, META, NVDA…) moves ≥ 5% after hours, re-pull the indicated gap for every cluster-1 long candidate at 9:25. Skip any whose indicated gap is < −2% (or < −0.5 ATR), and count the rest as one position under S-001.
• Why: on 10-29 after the close, INTC was −18% (Q4 guide $15–16B vs ~$17.1B, GM 38%) and AMZN −7 to −9%, despite AWS +39% and capex maintained. By ~4:12 PM, VRT was already −3.9%, ORCL −3.1%, DELL −3.5%, MU −2.9% and STX −2.0%. TRAI ρ to INTC is 0.50–0.71 for MU/MRVL/SMCI/ETN/STX/ORCL/VRT. The premarket run's indicated gap is computed at ~8:53 AM and can't see what happened between the 8:53 run and the open. The 8:53 indicated gaps were accurate on 10-29 (NOW −18.66% vs −19.47% actual).
• Caveats: after-hours quotes are thin and still move overnight. Use the 9:25 indicated gap, not the 4 PM print.
• Evidence: 2026-10-29 16:00 report §4 (after the close), §5, §6.5.

S-068 — Freshness guard on API layers, plus the NOW missed print · PROPOSAL (data quality; extends S-061 / S-008)

• Rule: (a) every run checks asOf for /api/scorecard, /api/correlations, /api/regime and /api/macro. If any is more than one session old, show a dashboard warning and fall back to conservative defaults (e.g., the S-001 cluster cap from the last good matrix, flagged). (b) Extend S-061: if a name's /api/events "next" date has passed with no recorded event, or the next date has jumped to a later estimate, re-query a second calendar or the company IR site.
• Why: on 10-29 all four layers still showed asOf 2026-10-22, five sessions stale through FOMC and the biggest earnings week of the quarter. NOW reported 10-28 AMC, but /api/events has no 10-28 event (history ends 07-22; next = 2027-01-27 estimate). The 8:53 run still flagged NOW short_momentum with a −18.66% indicated gap, but the event layer couldn't tag it as an earnings-reaction trade.
• Evidence: 2026-10-29 16:00 report header, §4 (NOW).

S-069 — Hyperscaler "funding stress" event map for AI-infra suppliers · NEEDS DATA (open research question)

• Question: when a hyperscaler/AI spender reports, does its FCF or financing surprise (negative FCF vs consensus, equity/debt raises) predict next-day returns of cluster-1 suppliers (VRT, ETN, ANET, MU, STX, DELL, SMCI) better than its capex-guide surprise?
• Why: this week no spender cut capex (GOOGL $195–205B, AMZN ~$220B, META $135–145B). Yet GOOGL −12.64% (FCF −$9.07B vs −$2.70B expected), META's ~$97.5B raise, ORCL's $20B raise cap with BBB- and CDS 287bp, and INTC's $23B equity all hit the tape. Suppliers with their own beat-and-raise (VRT, PWR) held, while balance-sheet-dependent names (ORCL, CRWV −4.99%, MRVL) fell. Needs consensus FCF and capex history per print (2023+), which TRAI does not store.
• Evidence: 2026-10-29 16:00 report §3, §6.1.

S-070 — whiggity's hypotheses: CLV-conditioned continuation, VRT→peer lead-lag, hardware–software spread · NEEDS DATA (open research question; from whiggity, §10)

• Question: (a) do ≥ +10% movers that close near the high (CLV > 0.85) continue the next day, versus the negative base rate in S-066? TRAI has rangePos (10-28: VRT 0.86, STX 0.99, ORCL 0.80, SOFI 0.84), so this can be tested on replay. (b) Do VRT big days lead MOD/NVT/ETN over t+1..t+5? MOD and NVT are out of universe, so this needs Yahoo data. (c) Does the hardware–software spread (avg VRT/ORCL/STX − avg NOW/META) mean-revert after > 2.5σ?
• Why: first pass on (c) (Yahoo 2022+, my calc): spread +12.70% on 10-28 (z 3.72) and +11.88% on 10-29 (z 3.10). After the 22 prior > 2.5σ days, the next-3-day spread averaged +1.01% (55% positive), so no mean reversion (small n, overlapping windows). whiggity's numbers used the wrong session (it treated 10-28 moves as 10-29's), and its other factual claims had several errors (report §10).
• Evidence: 2026-10-29 16:00 report §6.4, §10.

S-071 — Own-earnings guard for mean_reversion candidates · PROPOSAL (extends S-061 / S-066)

• Rule: if a mean_reversion candidate's own earnings release falls between the signal close and the entry open (AMC on the signal day, or BMO on the entry day), drop it or tag it own-print and skip it unless the 9:25 indicated gap agrees with the fade direction. The postclose run should also check /api/events for sessionsToReaction = 1 on every MR candidate.
• Why: the 10-29 postclose run (4:34 PM) picked GILD as a fade-short (+4.03%, z1d 2.28). GILD had released a beat-and-raise at 4:02 PM (FY product sales guide raised to $31.0–31.3B, non-GAAP EPS $2.55 vs ~$2.17) and was +2.2% after hours, above its 1.68% average print gap. A fade assumes a move on ordinary volume with no new information. The replay (04-23 → 10-21) has zero MR trades entered into the same name's own print, so the engine has never been tested on this case. The data to block it already exists: /api/events marks GILD gapRisk: imminent, sessionsToReaction: 1.
• Caveats: n = 0, so this rule rests on first principles rather than evidence. Some prints are non-events (GILD's avg |gap| is only 1.68%), but the information set has still changed.
• Evidence: 2026-10-29 17:00 report §4, §5 (GILD), §8.

S-072 — Bucket-level reaction-day concentration flag, plus prior-evening AMC tagging · PROPOSAL (extends S-066 / S-068)

• Rule: (a) when ≥ 50% of a bucket's candidates are earnings-reaction-day signals, show a dashboard warning and apply S-066 (opening-range confirmation or half size) to the whole bucket, with the S-001 cap counting them as related risk. (b) To tag a reaction day, treat a print as "reaction" if it was BMO on the signal day or AMC on the prior session. Cross-check the prior session against a second calendar, because /api/events missed both CVNA and CMG on 10-28.
• Why: on 10-29 the long_momentum bucket was 4/4 earnings-reaction trades: CVNA and CMG reported 10-28 AMC, MO and DASH 10-29 BMO. Replay reaction-day long_momentum: −0.56%/trade (n 48, hit 44%) vs +0.05% for the whole bucket (n 385). /api/events shows CVNA and CMG with next = null and their last event on 07-29, so the 10-28 prints are missing (the same failure mode as NOW in S-068). Without the second calendar, half the bucket would not even be tagged.
• Caveats: the reaction-day replay subset has t ≈ −1.2, which is not significant. The 50% threshold is arbitrary.
• Evidence: 2026-10-29 17:00 report §3, §7.1.

S-073 — Generalize S-067: post-close after-hours sympathy check for every bucket · PROPOSAL (extends S-067 / S-004)

• Rule: at 9:25, for every candidate, list cluster peers (ρ ≥ 0.55) and named "sector drivers" that printed after hours. Flag the candidate if the peer moved ≥ 3% after hours against the trade (e.g., AMZN/NOW-type software or megacap weakness vs a CRM fade-long). Note it as support if the peer moved ≥ 3% with the trade (e.g., MDLZ +3.15% after hours, ρ 0.60, supporting the MO long).
• Why: on 10-29 after the close, MDLZ was +3.15% (cluster 4, MO ρ 0.60), AMZN −6.3% and CRM −1.1% (CRM–NOW ρ 0.75). For ABBV (BMO Friday), GILD is ρ 0.59. S-067 only covers cluster-1 longs.
• Caveats: after-hours quotes are thin, so use the 9:25 indicated gap. The ρ values are 60d to 10-22 (stale, S-068).
• Evidence: 2026-10-29 17:00 report §5, §7.4.

S-074 — Day 2 after "EPS miss, top-line/guide beat" gap-ups · NEEDS DATA (open research question)

• Question: do gap-ups that come from a headline EPS miss together with a revenue or guide beat (CVNA: adj. EPS $0.39 vs $0.44 but revenue +42%; CMG: EPS $0.27 vs $0.29 but comps +2.1% vs ~1.5%) do worse on day 2+ than gap-ups from a clean beat? Both drew analyst target cuts.
• Why: both names are 10-30 long_momentum candidates. TRAI /api/events stores surprisePct (EPS only) per event but not revenue or guide surprise. CVNA's last 4 prints averaged −1.25% day-1 in gap direction, and CMG's July print had a day 2–5 drift of −12.5%. Needs a revenue-surprise field (e.g., from a second calendar) to test.
• Evidence: 2026-10-29 17:00 report §3, §7.1.

S-075 — whiggity's 17:00 test ideas: realized/implied + CLV fade, staple day-2 drift, co-reporter gap divergence, crypto residual, gap retention across macro prints · NEEDS DATA (open research question; from whiggity, 17:00 report §11)

• Questions:
• (a) After an earnings reaction with realized/implied move > 1.25 and close-location (CLV/rangePos) < 0.5, does a short on a break of the next session's opening-range low have positive expectancy?
• (b) Do low-beta staples with an earnings-day move > 3× ATR drift positively on day 2, so that buying a VWAP pullback works? (Ties to S-066 and S-074; MO on 10-29 is the live case.)
• (c) When two highly correlated names report in the same session (XOM–CVX, ρ 0.86), does their gap divergence mean-revert 30–50% by 11:00 ET?
• (d) Does the part of COIN/MSTR's gap that overnight BTC returns don't explain fade in the first hour?
• (e) How much of a premarket gap survives an 8:30 ET macro print?
• Why: these are plausible and testable, but whiggity gave no evidence for any of them. Its factual calendar claims had several errors: PCE on Friday, "FOMC Nov 5/6", UMich final on Friday, AAPL tonight. (a) needs implied-move history, which TRAI doesn't store. The only dated figure found was CVNA ±13.4% (EarningsWatcher), giving realized/implied ≈ 1.17, so it doesn't trigger. (c)–(e) need intraday bars, and (d) also needs BTC data.
• Evidence: 2026-10-29 17:00 report §11.

S-076 — Rotation-day flag: "funded" vs "sold" side for every candidate · PROPOSAL (extends S-073)

• Rule: when the premarket QQQ − DIA spread is beyond ±2pp, tag every bucket candidate by whether its sector is on the bid ("funded") or offered ("sold") side of the rotation, and don't take mean_reversion fades against the funded side (shorts in the bid sector, longs in the sold sector) without opening-range confirmation.
• Why: on 10-30 at 09:17 QQQ was −2.60% vs DIA −0.10% (SPY −1.03%). The MR bucket's three shorts (GILD, HD, LOW) were in bid sectors (healthcare, rate-sensitives) and its one long (CRM) in the sold sector (software: TWLO −21.9%, DDOG −4.1%, IGV −1.2%).
• Caveats: n = 1 day; the 2pp threshold is arbitrary. whiggity's test idea S-081(1) is the way to calibrate it.
• Evidence: 2026-10-30 09:00 report §2, §7.1, §11.

S-077 — Skip MR fade-shorts on own beat-and-raise gaps > 2× typical when the sector is leading · PROPOSAL (extends S-071)

• Rule: drop a mean_reversion fade-short if (a) the name's own print sits between signal and entry, (b) the print was a beat-and-raise, (c) the 9:25 indicated gap is > 2× the name's average |print gap| (TRAI /api/events), and (d) the sector is leading premarket (S-076 "funded" side).
• Why: GILD on 10-30: beat-and-raise 10-29 AMC, five PT hikes overnight, indicated +4.76% vs 1.68% avg |gap| (2.8×), healthcare bid (REGN +6.9%, LLY/AMGN/MRK +2.6–2.9%). The 8:53 premarket run still listed it as a fade-short.
• Caveats: replay has n = 0 own-print MR entries (S-071). Fading new information is outside what the MR bucket was built for, so this rests on first principles.
• Evidence: 2026-10-30 09:00 report §2, §6 (GILD), §8.

S-078 — Treat HD and LOW as one position · PROPOSAL (extends S-001)

• Rule: when HD and LOW are both in a bucket on the same side, count them as one position (one risk unit split, or pick the one with the larger z1d).
• Why: 60d ρ 0.92 (TRAI, stale to 10-22), same macro driver (10y yield), and on 10-30 both were fade-shorts gapping up again (+1.22% / +1.07% at ~09:13 / 08:10). Two shorts double the same rate bet.
• Evidence: 2026-10-30 09:00 report §2, §6; earlier HD/LOW macro-factor note in S-013.

S-079 — /api/events missed GILD's 10-29 AMC print (4th miss this week) · PROPOSAL (data quality; extends S-068 / S-072)

• Rule: after each postclose and premarket run, for every bucket candidate check whether a print fell in the last 2 sessions on a second calendar (or company IR) when /api/events shows next = null or a passed next; tag as own-print if so (feeds S-071/S-077), and log the miss.
• Why: on 10-30 /api/events shows GILD next = null, n = 3, i.e. the 10-29 AMC beat-and-raise is not recorded. Same failure mode as NOW (10-28), CVNA and CMG (10-28) earlier this week.
• Evidence: 2026-10-30 09:00 report §6 (GILD), §9.
• Revision 2026-10-30 10:00 (PROPOSAL, draft): the "misses" are a structural blind spot, not random calendar errors. At 12:17 UTC every 10-29 AMC reporter (INTC, AMZN, RBLX, PINS, MSTR, COIN, SBUX, RIVN, MDLZ, GILD) had no 10-29 event and next null or rolled to the next quarter, while 10-28 AMC reporters (CVNA, CMG, NOW, MSFT, GOOGL, F) were by then recorded. Cause (lib/research/events.ts): history is written only once the reaction-day bar exists, and next counts as upcoming only if nextDate >= today (line 97). So an AMC print is invisible from the print until the reaction-day close, which is exactly the premarket-run window. Revised rule: treat nextDate === previous trading day && timing === 'AMC' as "reacting today" (sessionsToReaction 0, gapRisk imminent), and cache the last-seen next per ticker so a Yahoo roll-forward can't erase it. The second-calendar cross-check stays as a backstop. Evidence: 2026-10-30 10:00 report §7.

S-080 — Re-label CVNA's sector ("Crypto/High-beta" isn't supported) · PROPOSAL (data quality)

• Rule: replace CVNA's "Crypto/High-beta" tag with "Consumer Disc./High-beta" (or derive the label from correlation clusters per S-001) so crypto-shock logic doesn't misfire.
• Why: 60d ρ to COIN 0.26, MSTR 0.27 (TRAI). Yahoo 60d to 10-29 (my calc): β vs QQQ 0.68 (ρ 0.18), vs SPY 1.88. On the Binance-theft morning COIN/MSTR were −7.5%/−7.3% while CVNA was −1.0%, close to its SPY-beta-implied ≈ −2%.
• Evidence: 2026-10-30 09:00 report §5, §11.

S-081 — whiggity's 09:00 test ideas: dispersion-conditioned PEAD vs fade, Dow-member catch-down, beta-adjusted premarket residual, month-end Friday profile · NEEDS DATA (open research question; from whiggity, 09:00 report §11)

• Questions:
• (1) When premarket r(QQQ) − r(DIA) < −2pp, do defensive beat-and-raise names gapping > +4% continue intraday rather than fade? (Calibrates S-076/S-077; GILD/REGN/MDLZ are the 10-30 cases.)
• (2) When a DJIA software/tech member (CRM, MSFT, IBM) gaps down < 1% while peers (IGV/TWLO/AMZN) gap < −5%, what is its 9:30–10:00 return? (CRM fade-long on 10-30.)
• (3) Do high-beta names with a large positive premarket residual (r_pre − β·r_SPY,pre, TRAI beta60) and no own catalyst underperform in the first hour? (CVNA's 10-30 residual was only ≈ +1pp, not whiggity's +5.5pp.)
• (4) On month-end Fridays with NQ gapping < −2%, what are the 9:30–10:00, 10:00–14:00 and 14:00–16:00 return profiles?
• Why: plausible and testable, but whiggity gave no citations and its factual claims had errors again (UMich final "today" — it was Oct 23; PCE "today" — it was 10-29; "10y near cycle high" — it fell >5bp; CVNA NQ-β 2.5–3.0 — measured 0.68). (2)–(4) need intraday bars; TRAI's replay is daily-bar only.
• Evidence: 2026-10-30 09:00 report §11.

S-082 — 10:00 re-check: drop long_momentum candidates trading below both OR30 low and VWAP · PROPOSAL (extends S-066 / S-072)

• Rule: at 10:00 ET, re-score every long_momentum candidate against its 9:30–10:00 opening range and session VWAP. If it is below the OR30 low and below VWAP, drop it for the day unless it reclaims the OR. Mirror the rule for fade-shorts: arm the short only once the name trades below its OR30 low (the 09:00 "OR failure" condition), and skip it while it is above the OR30 high.
• Why: on 10-30 at ~10:13 ET, CVNA (−2.06%, OR30 −1.76…+0.76%, −1.24% vs VWAP) and MO (+1.29%, OR30 +1.38…+3.99%, −1.35% vs VWAP) had both failed, while BA was the only long above its OR high. The HD/LOW fade-shorts met their OR-failure condition (HD +0.03% vs OR low +0.12%; LOW +0.58% vs +0.72%), and GE was below its OR. TRAI's replay uses daily bars, so the bucket currently has no intraday confirmation step.
• Caveats: n = 1 day. OR30 and VWAP come from unofficial Yahoo 5-min bars. Needs an intraday replay to calibrate, which TRAI doesn't have yet (see S-081(4)).
• Evidence: 2026-10-30 10:00 report §2, §9.

S-083 — Day-2 "gap-and-fade" on defensive post-earnings names · NEEDS DATA (open research question)

• Question: when a low/negative-beta defensive name (staples, telecom) gaps up on day 2 after its print in a risk-off tape, how often does it open at the high and close below its open? Should long_momentum skip such names when the gap is larger than ~1 ATR-equivalent of the name's normal day?
• Why: MO (beta60 −0.77) opened +1.75% on 10-30, day 2 after its 10-29 BMO print, and fell through its OR low. PG/KO/PEP likewise traded below their opens while still up on the day. By contrast, the telecom cluster (TMUS day 2 after its 10-28 AMC print; TMUS faded its whole reaction-day gap on 10-29) broke above its OR30 highs on 10-30. Mixed evidence, n = 1.
• Evidence: 2026-10-30 10:00 report §2, §3, §9.2.

S-084 — whiggity's 10:00 test ideas (dispersion-conditioned ORL/VWAP failures, post-hyperscaler SaaS drift into own print, month-end no-news defensive squeezes, AAPL RS as a QQQ stabilizer) · NEEDS DATA (open research question; from whiggity, 10:00 report §10)

• Questions (all from whiggity, untested):
• (a) When r(DIA) − r(QQQ) > +1.5pp in the first 45 minutes, do names below both their 15-min opening-range low and VWAP recover less in the afternoon (13:00–16:00) than other names? Do Dow members above their 15-min ORH show more persistence? (10-30: spread ≈ +2.3pp at 10:23.)
• (b) For cloud/SaaS names with a ≤ −6% sympathy drop on a hyperscaler's earnings reaction day, within 5 sessions of their own report: is the drift from the Friday (reaction-day) close to the close before their own report positive-Sharpe? (10-30 cases: DDOG −10.1% before 11-04 BMO; NET −9.0% before 11-05 AMC.)
• (c) On month-end Fridays, do no-news defensive sector squeezes (10-30: the telecom cluster) have a smaller high-to-VWAP drawdown from 10:30 to 13:30 and more acceleration into the MOC than news-driven moves (10-30: GILD, LLY)?
• (d) Does AAPL's beta-adjusted relative strength vs QQQ from 10:15 to 11:00 predict afternoon QQQ stabilization on the session before a Monday AMC AAPL print? (10-30: AAPL −0.87% vs QQQ −3.07% at ~10:23.)
• Why: these are plausible and directly testable, and (a) would calibrate S-082. whiggity gave no data. In the same reply its factual claims had errors (UMich "final today" again; PCE "today", when it was 10-29; AMZN capex "raised" at Q3, when it was held). So treat these as hypotheses only.
• Data needed: intraday (≤5-min) bars with VWAP for the universe, plus earnings dates, a month-end calendar and a news/no-news tag. TRAI's replay is daily-bar only (same blocker as S-081(2)–(4)).
• Evidence: 2026-10-30 10:00 report §10, §12.

S-085 — Cluster-level earnings-window flag · PROPOSAL (extends S-078 / S-001)

• Rule: when ≥3 names in one TRAI correlation cluster have elevated or imminent gap risk within the next 4 sessions, treat the whole cluster as one gap-risk exposure: cap combined candidate weight at one position's risk and skip overnight holds in cluster members on the eve of each print.
• Why: cluster 1 (avg ρ 0.588) has ON (11-02), AMD/ANET/SMCI (11-03), ARM/QCOM (11-04) and MCHP (11-05) reporting in consecutive nights, with TRAI last-4 avg |gap| 5.7–10.0% for most of them. On 10-30 one name's print (INTC) moved the whole cluster −3% to −10%.
• Caveats: sympathy size per print is untested; the 60d clusters are stale (S-068).
• Evidence: 2026-10-30 11:00 report §4, §6.3.
• Added evidence (11-02 10:00): cluster 5 meets the condition this week: PLTR 11-02 AMC, DDOG 11-04 BMO, NET 11-05 AMC (DDOG–NET ρ 0.80, PLTR–DDOG 0.67). On 11-02 the cluster was the weakest software group on day 2 (DDOG −4.6%, NET −4.4% at ~10:15 ET). Cluster 12 (APP, QCOM) has two prints on 11-04 AMC. 2026-11-02 10:00 report §2f, §6.

S-086 — Pre-print sympathy drawdown event study · NEEDS DATA (open research question)

• Question: when a name falls ≥8% in sympathy with a peer's print within 3 sessions before its own print, does its own reaction gap differ (sign, size, fill rate) from its unconditional history? Does the stabilization of the reporting name (INTC above VWAP by late morning) while upcoming reporters keep falling predict anything for the next session?
• Why: on 10-30, AMD (−9.4%, prints 11-03), QCOM (−10.3%, 11-04), ARM (−10.5%, 11-04), ON (−4.8%, 11-02) and MCHP (−4.3%, 11-05) were falling or at lows at 11:05 while INTC itself and the PC OEMs (DELL, HPQ) were bouncing.
• Data needed: TRAI /api/events history (already has gaps per print) joined with daily bars to tag pre-print sympathy drops.
• Evidence: 2026-10-30 11:00 report §2, §6.1–6.2.

S-087 — whiggity's 11:00 test ideas: sympathy-drawdown gap regression, post-shock implied vs realized moves, INTC→AMD guide dispersion, Friday capitulation · NEEDS DATA (open research question; from whiggity, 11:00 report §8)

• Questions (from whiggity, untested): (1) own earnings gap vs T−3…T sympathy return and its SMH residual, conditioned on guide surprise (overlaps S-086); (2) front-week implied move vs realized gap for reporters right after a sector shock with SMH 5-day RV above its 90th percentile; (3) P(AMD guide surprise > 0 | INTC guide surprise < −1.5σ), split by Intel data-center growth; (4) on Fridays with SMH < −5% by 11:00, correlation of the 9:30–11:05 return with 11:05–16:00, and Monday open/close reversion.
• Why: testable and relevant to next week's cluster-1 print run (S-085). whiggity's mechanism claims came without sources, and one of its figures (prior-quarter INTC revenue) conflicted with RTE, so treat these as hypotheses.
• Data needed: (1) and (3) TRAI /api/events history plus consensus guidance data TRAI doesn't have; (2) options implied vols (TRAI has an /options page; history coverage unknown); (4) intraday bars (same blocker as S-081/S-084).
• Evidence: 2026-10-30 11:00 report §8.

S-088 — Calendar-day gap-risk window for weekend-spanning prints · PROPOSAL (extends S-001 / S-085)

• Rule: in /api/events, add a calendarDaysToReaction field and flag weekendExposure: true when a name's next reaction session falls after a weekend or holiday. On Fridays, exclude names that print Monday (BMO or AMC) from new momentum entries, or size them at the earnings-window cut.
• Why: on 10-30 the API reports PLTR, AAPL, ON and VRTX as tradingDaysUntil 1, but a Friday entry carries two weekend days plus Monday before the print. PLTR's last-4 avg |gap| is 2.23 ATR with 0% fill.
• Evidence: 2026-10-30 12:00 report §1, §7.

S-089 — Does a large pre-print sympathy drop shrink the name's own print gap? · NEEDS DATA (open research question; extends S-086)

• Question: when a name falls ≥2× its own average earnings |gap| (in ATR) in the 3 sessions before printing, is its own reaction smaller, larger, or reversed vs history?
• Why: ON fell −5.9% on 10-30 against a 0.52-ATR average print gap; PLTR fell −4.9% against 2.23 ATR.
• Data needed: TRAI /api/events history joined with daily bars (same as S-086).
• Evidence: 2026-10-30 12:00 report §5, §9.

S-090 — whiggity's 12:00 test ideas: Friday-capitulation gap skew, IV-rich gap-and-go, defensive pre-print fade · NEEDS DATA (open research question; from whiggity, 12:00 report §8)

• Questions (from whiggity, untested): (1) when a name drops >4% and closes below VWAP on the session before its print during a sector rout (SMH < −5% or IGV < −3%), is its post-print gap skew more positive than after flat or up pre-print days? (overlaps S-086/S-089); (2) for names with 0% historical gap fill (PLTR), does day-1 open-to-close continuation weaken when the options-implied move is more than 1.1× the trailing 4-print average |gap|?; (3) when a low-gap name (avg |gap| < 1.5%) outperforms QQQ by more than 5 pts in the 2 sessions before printing (VRTX 10-30: +3.4% vs −3.2%), is its 3-day post-print abnormal return negative?
• Note: whiggity's 4th idea assumed ON reports BMO Monday; onsemi IR says AMC, so it was dropped.
• Data needed: TRAI /api/events history plus daily bars for (1) and (3), and historical implied moves for (2) (TRAI /api/options history coverage unknown).
• Evidence: 2026-10-30 12:00 report §8.

S-091 — Intraday regime re-check (12:00 and 14:00 ET) · PROPOSAL (extends S-068)

• Rule: recompute the regime label and riskPerTradePct from live ^VIX/^VIX9D/^VIX3M at 12:00 and 14:00 ET and publish to /api/latest (or a light /api/regime-live). Show "regime as of" on the dashboard.
• Why: on 10-30 the dashboard still showed the 8:53 AM run (VIX 18.59, VIX9D/VIX 0.854) while the live tape at 12:58 was VIX 20.85 (high 20.94) and VIX9D/VIX 0.991 (high 0.997). That is one tick from the stress label and 1.15 pts from the risk-halving line.
• Evidence: 2026-10-30 13:00 report §2.

S-092 — Align risk halving with the stress label (code vs Glossary mismatch) · PROPOSAL (engine/docs consistency)

• Rule: either halve riskPerTradePct whenever the label is stress (VIX > 22 or VIX9D/VIX > 1), or change the Glossary/Checklist text to say only VIX > 22 halves risk.
• Why: lib/engine.ts sets stress on either trigger (L154) but halves risk only on VIX > 22 (L180). components/Glossary.tsx says "In stress, risk per trade is halved". An inversion-only stress day would show stress with full 0.25% risk.
• Evidence: 2026-10-30 13:00 report §2, §8.

S-093 — "Haven failure" days: gold ≤ −2% with QQQ ≤ −2% · NEEDS DATA (open research question)

• Question: on days when gold and growth equities fall together (liquidation and hike-repricing), is the next-day open→close for long_momentum candidates worse, and do cluster-16 names (NEM/FCX) behave like cluster 1 rather than as hedges?
• Why: 10-30 is the live case: gold −3.15%, QQQ −3.6%, NEM −5.5%, FCX −5.4%, HIMS/MARA/RIOT −9–10%. For gold ≤ −3% days generally (n 25, 2016–2026), NEM's next-day gap is a coin-flip (mean −0.26%), with close-to-close median +0.72% (my calc).
• Data needed: daily bars for GC=F/QQQ joined with /api/scorecard replay by bucket.
• Evidence: 2026-10-30 13:00 report §3.

S-094: Continuous risk scalar instead of the VIX > 22 step · PROPOSAL (from whiggity; extends S-091/S-092)

• Rule (to test): riskPerTradePct = 0.25% × f(z, s), where z = VIX's intraday z-score vs its trailing 20-day distribution and s = max(0, VIX9D/VIX − 0.95). Scale down smoothly from 0.25% toward 0.125% instead of halving at VIX > 22.
• Why: on 10-30, VIX9D/VIX reached 0.997 with VIX 20.9. The step rule gave full risk while the front of the curve was flat. whiggity suggests the ratio may be inflated by event density (ISM 11-02, JOLTS + midterms 11-03, NFP 11-06; calendar verified), which a scalar would treat more gracefully than a binary trigger. The event-density explanation itself is unverified.
• Test: replay /api/scorecard by bucket (2016–2026) under step vs scalar sizing; compare drawdown and Sharpe; split by pre-event weeks.
• Evidence: 2026-10-30 13:00 report §2, §9.

S-095: Cross-asset shock tests from the 10-30 tape · NEEDS DATA (from whiggity, all untested)

• (1) On QQQ ≤ −1.5% days with a positive GLD/SLV–QQQ 20d correlation, do GDX/NEM mean-revert over 1–3 days? (pairs with S-093)
• (2) 4-factor miner beta (copper, gold, crude, SPY) for NEM/FCX/GDX, to separate the cost channel (oil ↑ → AISC ↑) from the rates channel.
• (3) Friday CL1 moves > +2.5% on geopolitical headlines: hold vs fade at the Sunday open.
• (4) In Hormuz-type shocks, does an XOP/JETS pair beat outright USO? (10-30: XOP +1.8%, JETS −3.2%, USO +2.1%.)
• (5) SPOT idiosyncratic continuation on VIX-spike days (extends S-046).
• Data needed: Yahoo daily bars (GC=F, SI=F, CL=F, HG=F, GLD, SLV, GDX, NEM, FCX, XOP, JETS, USO, SPOT, ^VIX); Sunday-open futures data for (3).
• Evidence: 2026-10-30 13:00 report §9.

S-096: Rate-beta tag and a shared "duration" exposure · PROPOSAL (extends S-078 / S-085 / S-088)

• Rule: compute each candidate's 2-factor rate beta: daily return on SPY and on Δ10y (%/10bp), over a 60d or YTD window. Show it on the dashboard. Count candidates with |rate beta| > 1.5%/10bp as one shared duration exposure for sizing. Add scheduled rate events to the gap-risk window for those names, the same way earnings are flagged today: ISM, Treasury refunding (estimates Mon 15:00, statement Wed 08:30), JOLTS, NFP, CPI/PCE, FOMC, and Freddie Mac PMMS (Thu 12:00).
• Why: on 10-30 the rate betas were HD −1.99 and LOW −2.07 %/10bp (t −8.8 / −8.0, 2026 YTD, my calc), as large as XHB/ITB. CVNA was −2.00 (t −3.6). The engine's sector labels miss this: HD/LOW are "Consumer Disc." and CVNA is "Crypto/High-beta". Big banks have ~0 rate beta, so "rate-sensitive" is not a sector property. A 10bp move after Monday's ISM or the refunding estimates is worth ~±2% on HD/LOW, about one ATR.
• Evidence: 2026-10-30 14:00 report §4, §7, §8a.

S-097: Friday-signal effect by bucket · NEEDS DATA (open research question)

• Question: does the signal weekday matter? In the replay, Friday-signal (Monday-traded) long-momentum picks made o→c +0.57% (n 121, 16 Fridays, hit 58%) vs +0.01% Mon–Thu. MR fade-shorts made −0.25% (n 36) vs +0.20%, and MR fade-longs −0.47% (n 33) vs +0.34% (TRAI /api/scorecard, 2026-04-23 to 10-21, gross).
• Caveats: six months, mostly a transitional/risk-on regime; survivorship bias (today's universe); only 9–20 Fridays per bucket.
• Data needed: a longer replay (Yahoo daily 2016+), split by vx regime and by the Monday macro calendar.
• Evidence: 2026-10-30 14:00 report §8.

S-098: Bucket edge on stress-labelled signal days · NEEDS DATA (open research question; interacts with S-092 / S-094)

• Question: before stress cuts risk (S-092) or a scalar replaces the step (S-094), check whether edge changes on stress days. Replay stress-labelled signal days: MR fade-short o→c +1.00% (n 28, hit 75%); long momentum +1.87% (n 9); MR fade-long +0.49% (n 35); short momentum +0.15% (n 25). All four exceed their all-days means (+0.15%, +0.18%, +0.22%, −0.41%). But there are only 8 signal days, and most fall in the 2026-06-05…06-17 episode.
• Why now: on 10-30, VIX9D/VIX crossed 1.0 at 13:50 (1.008 at 13:59) with VIX 21.79. Tonight's postclose run is likely the first stress label since 10-16.
• Data needed: replay on 2016+ daily bars with the engine's regime rules applied historically.
• Evidence: 2026-10-30 14:00 report §2.

S-099: whiggity's 14:00 rate-complex tests: rotation-rally follow-through, curve-shape-conditioned rate beta, post-print leader vs peer pair · NEEDS DATA (open research question; from whiggity, 14:00 report §9)

• Questions (from whiggity, untested):
• (1) After HD/LOW gain ≥ 3% on a day when the 2y rises and QQQ/SMH fall ≥ 1.5% (a rotation rally, not a rates rally), what are the 1/3/5-day forward returns, conditional on a close in the top quartile of the day's range? This complements the 14:00 §7 study, which conditioned on the 10y falling.
• (2) Is the rate beta of the duration names (HD/LOW/XHB, utilities, MO, CVNA) different under bear-flattening vs bear-steepening days, and when WTI > $95? Extends S-096, which uses a single Δ10y beta.
• (3) A 3–5 day pair: post-earnings leader long vs non-reporting peers short (e.g. CVNA vs KMX/ALLY), in hawkish-Fed regimes. Generalizes S-050. Live case 10-30: CVNA +1.65%, KMX −1.41%, ALLY −4.19% (~14:20 ET).
• Not added (duplicates): Friday VIX9D/VIX ≈ 1 pre-macro-week Monday gaps (S-094 plus the 14:00 §8 study); Friday gold/copper/SMH-down, WTI-up Monday gaps by beta decile (S-093/S-095).
• Caveat: whiggity gave no sources this consult, and three of its checkable facts were wrong (LOW date, DASH date, GILD net income), while the HD, CVNA and GILD dates were right. Treat the mechanisms as hypotheses only.
• Data needed: Yahoo daily bars (HD, LOW, XHB, ITB, XLU, DUK, SO, NEE, MO, CVNA, KMX, ALLY, QQQ, SMH, CL=F), a 2y/10y yield history (Yahoo lacks a 2y index; use FRED DGS2/DGS10), and TRAI /api/events history for (3). Intraday high/low for (1) is available from daily bars.
• Evidence: 2026-10-30 14:00 report §9.

S-100: Down-weight long-momentum picks whose signal day is an earnings reaction day · PROPOSAL (extends S-074 / S-085)

• Rule: if a long_momentum candidate's signal date equals its /api/events reactionDate, half-size it, or require a next-day OR30 break before entry.
• Why: in the replay (2026-04-23 to 10-21, gross), earnings-reaction LM picks made −0.42% next-day o→c (n 48, median −0.28%, hit 44%) vs +0.27% for non-earnings LM picks (n 337). Thursday-signal earnings picks (traded Friday) made −1.29% (n 19); Friday-signal ones made +0.92% (n 9). On 2026-10-30, the live case went the other way: CVNA/CMG/DASH/MO were all up open→now on a −2.1% SPY day.
• Caveats: six months, small subsets, survivorship (today's universe), gross of costs; one contrary day does not refute it.
• Data needed: a 2016+ replay using Yahoo daily bars plus historical earnings dates.
• Evidence: 2026-10-30 15:00 report §4(c).

S-101: "Held the print through a down tape" follow-through · NEEDS DATA (open research question)

• Question: do earnings winners (gap > 3%, reaction c/c > 3%) that still close up on a day 2 when SPY is ≤ −1% drift better over days 3–5 than other print winners?
• Why: in TRAI's 2y events (n 125 winners), the day-3 gap is a coin flip in every subset (43–48% up), and LM-like winners have a day-3 c/c median of −0.59%. The only positive subset is day-2 SPY ≤ −1% (n 12; day-3 c/c median +3.11%), and the exact "held up" case is n 4 (median +1.57%). That is too small to use.
• Data needed: 10y earnings dates; control for market beta (CVNA/DASH β ~2 intraday vs MO −0.95).
• Evidence: 2026-10-30 15:00 report §4(b), §5.

S-102: Same-timestamp VIX9D/VIX in the regime label, plus a "borderline" flag · PROPOSAL (data quality; extends S-091 / S-092 / S-094)

• Problem: lib/engine.ts L146–150 takes the last daily close of ^VIX and ^VIX9D. In the premarket run, Yahoo's ^VIX daily bar already carries the live pre-open value, but ^VIX9D still shows the prior close. So the ratio compares two timestamps. In all six premarket runs in /api/history (10-22 → 10-29), vix9d equals the prior postclose value while vix has moved.
• Evidence: the only premarket stress label in the history (10-27 8:53, ratio 1.008) came from VIX falling overnight (18.39 → 17.78) below a stale VIX9D (17.92); the postclose runs on either side were transitional. On 10-30 8:53 the ratio was 0.854 (VIX 18.59 live vs VIX9D 15.87 stale) with Nasdaq futures −2.6%. By 13:50 the live ratio was > 1. At today's close the ratio was 0.99–1.004, depending on which Yahoo endpoint you read.
• Rule:
• (1) Premarket run: compute r9 from prior closes for both (or from VIX's prior close if live VIX9D is unavailable), and show live VIX separately.
• (2) Label |r9 − 1| < 0.01 as "borderline" rather than flipping stress/transitional.
• (3) Re-evaluate at 9:35 with live values (S-091).
• Monday 11-02 risk: Friday's VIX9D close (20.94 in the official 10-30 postclose run) will be the denominator, so a calmer Monday pre-open would print a spurious stress label.
• Evidence link: 2026-10-30 16:00 report §3.
• 2026-11-02 confirmation: the 8:58 AM premarket run printed stress (VIX9D/VIX 1.015 = Friday's final VIX9D 19.25 over live VIX 18.96). Same-timestamp Friday ratio was 0.991 (transitional). Sizing was unaffected (VIX < 22). Evidence: 2026-11-02 09:00 report §4.

S-103: Turn-of-month tag for first-trading-day picks · NEEDS DATA (open research question)

• Question: does the first trading day of a month change bucket edge?
• Data so far: SPY's first-day gap, 2016–2026 (n 128), was 63% up (mean +0.10%), with o→c ≈ flat (+0.05%). After a last-day-of-month ≤ −1.5% (n 9), the gap was 8/9 up (mean +0.56%); QQQ c/c +0.80% (my calc, Yahoo daily).
• Test: replay /api/scorecard by bucket, first trading day vs other days; check whether LM fade-longs / MR fade-shorts that open into a turn-of-month gap behave differently. Combine with S-097 (Friday-signal effect), since 11-02 is both.
• Evidence: 2026-10-30 16:00 report §6(b).

S-104: Measured sympathy map before flagging out-of-universe prints · NEEDS DATA (extends S-085)

• Question: which non-universe BMO/AMC prints actually move universe names?
• Rule (to test): flag a universe name for sympathy gap risk only when its same-day excess return vs SPY matched the reporter's sign on ≥ 7 of the last 9 prints.
• Evidence: on CHKP's 9 inferred reaction days (2024-07 to 2026-07; dates from volume spikes, not IR-verified), sign agreement was PANW 4/9, CRWD 3/9, ZS 5/9, NET 3/9, and only FTNT 8/9 (not in the universe). CHKP's own moves were up to −19.6%. A big CHKP move therefore says little about the universe security names.
• Data needed: verified earnings dates (IR or 10y calendar) for frequent out-of-universe reporters (CHKP, FTNT, MAR, HLT, PEG, ETR, CIFR, EL), plus Yahoo daily bars.
• Intraday extension (from whiggity, 16:00 §9): measure FTNT/PANW (and CRWD/ZS/NET) first-30-minute beta to CHKP's premarket gap when |gap| ≥ 4%. Needs intraday history beyond Yahoo's ~60 days of 5-min bars; 11-02 BMO is the first live case.
• Evidence: 2026-10-30 16:00 report §6(a).

S-105: Friday VIX-spike reversal → Monday morning QQQ/SMH · NEEDS DATA (open research question; from whiggity, 16:00 report §9; extends S-094)

• Question: on Fridays where VIX spiked intraday and then reversed, R_rev = (VIX high − close)/(VIX high − prior close) > 0.5, with VIX9D/VIX > 1 at the close, what is the Fri close → Mon 10:30 ET return of QQQ and SMH, vs Fridays with a spike that held (R_rev < 0.5)?
• Why: 10-30 is a live case: VIX 17.20 → high 23.93 → 19.47 on Yahoo's final bar (R_rev 0.66; 0.48 using the 4:21 PM run's 20.68). The 14:00 report §8 found SPY's Monday median +0.52% (77% up) after VIX-spike Fridays without conditioning on the reversal.
• Caveat: the reversal is sensitive to which VIX print is used (Yahoo's daily bar differed from the run's value by 1.2 pts on 10-30). Mon 10:30 also straddles the 10:00 ISM on 11-02.
• Data needed: ^VIX, ^VIX9D daily OHLC (Yahoo, 2011+); QQQ/SMH 30-min bars for the Monday 10:30 mark (intraday history beyond Yahoo's 60 days, or use Monday open/close as a proxy).
• Evidence: 2026-10-30 16:00 report §2, §3, §9.

S-106: Mean reversion vs distance to earnings · NEEDS DATA (open research question; from whiggity, 16:00 report §9; extends S-085 and the pre-earnings mega-cap MR question)

• Question: do mean_reversion longs with a 3-day z < −1.5 work differently when the next print is > 10 sessions away vs ≤ 4 sessions away (both before the print, no overlap with the reaction day)?
• Why: the 10-30 4:21 PM MR list is print-adjacent: PLTR (Mon AMC), SHOP/ETN (Tue BMO), ANET (Tue AMC), NET (Thu AMC). Earlier replay: universe pre-earnings MR n 27, −0.29% mean (see the mega-cap MR item). Pre-print drawdowns may carry information (S-086) rather than noise.
• Test: replay /api/scorecard MR picks, bucketed by sessions-to-print from /api/events (≤ 4, 5–10, > 10), next-day o→c and 3-day; gross, then with a cost haircut.
• Data needed: historical earnings dates (≥ 5y) for the universe; Yahoo daily bars.
• Evidence: 2026-10-30 16:00 report §9, §10.
• Live case (11-02): PLTR (prints that evening) and SHOP (next BMO) are MR fade-longs on print eve. At ~10:15 ET both were above VWAP: PLTR +0.9% with cluster-5 peers −4%, SHOP +1.3%. Score these at the close as the first ≤ 1-session observations. 2026-11-02 10:00 report §6.
• Live case (11-02 12:00): ANET, a fade-long one session before its Tue AMC print, was −1.44% at 12:10 ET and below VWAP. Score at the close. 2026-11-02 12:00 report §2.

S-107: DST-switch verification for the EDT/EST cron twins · PROPOSAL (ops; no cron change)

• Facts: vercel.json crons are UTC. Each job has an EDT twin and an EST twin, and cronGate / app/api/cron/route.ts run only the twin whose tz matches New York's current offset (lib/stats.ts etZone). A simulation against the real code for Mon 2026-11-02 runs premarket at 13:45Z (8:45 EST) and postclose at 21:20Z (16:20 EST) and skips the EDT twins (12:45Z / 20:20Z). So ET run times do not shift after DST ends on Sun 11-01 (timeanddate, 2026-10-02).
• Risk: the twins have never been exercised across a live switch, and Vercel Hobby can start crons late within the hour.
• Proposal:
• (1) Monday 11-02: check Vercel logs for skip … tz=edt but New York is on EST at ~12:45Z and a premarket run ID stamped ~T13-4x…Z in /api/latest before 9:30 ET; same for postclose (~21:20Z).
• (2) Add a missed-run check: if no premarket run exists by 9:15 ET (or postclose by 17:00 ET) on a trading day, surface a dashboard banner and allow a force=1 manual run.
• (3) Show ET times (not UTC stamps) on the dashboard and in /api/history, so the 1-hour shift in run-ID UTC stamps is not misread as a schedule change.
• Repeat the check at the next switch (Sun 2027-03-14, DST starts).
• Re-verified at 17:00 (whiggity consult):
• I simulated the real gate code with 0, 8 and 59 minutes of start delay. Exactly one twin per job runs on Mon 11-02, at unchanged ET times: premarket 8:45 (8:53 with the usual delay), postclose 16:20, events 7:30, options 15:30, daily 16:40, intraday 17:05.
• The GitHub Actions daily overview (35 21/35 22 UTC + cron.sh ≥ 17:15 ET gate) stays at 17:35 ET.
• There are no hardcoded ET offsets: etZone/etToIso use Intl America/New_York.
• No cron change is needed for DST. (2026-10-30 17:00 report §1)
• Also note: foreign releases move in ET: Caixin PMI is 20:45 EST on Sundays (was 21:45 EDT), and Europe-to-NY offsets return to 5h.
• Evidence: 2026-10-30 16:00 report §9.

S-108: Post-close data-freshness guard for Yahoo daily bars · PROPOSAL (data quality; extends S-102)

• Problem: the official 2026-10-30 postclose run (4:21 PM ET) used stale Yahoo range=6mo&interval=1d bars:
• 22 of 156 names were off by more than 0.5% vs the final close (12 by more than 1%; COIN +3.9%, SNOW +2.8%). The stale values match 5-min bars from 12:05–15:40.
• SPY 751.60 vs 755.72; VIX 20.68 / VIX9D 20.94 vs a final 19.42 / 19.25.
• As a result the label flipped to stress (1.013) instead of transitional (0.991), and 7 of the 17 MR names would not qualify on final data.
• All 12 earlier runs (10-22 → 10-29) matched final closes (max diff 0.14%).
• Rule (proposal):
• (1) In fetchChart/toDaily for postclose runs, compare the last daily close with meta.regularMarketPrice and the last regular-session 5-min close. If they differ by more than 0.05%, refetch from the other host. If the mismatch persists, rebuild the day's OHLC from the 5-min bars.
• (2) Store staleCount and the max diff in the run JSON.
• (3) If staleCount > 0 after the retry, schedule or allow an automatic re-run at ~16:50 ET.
• Evidence: 2026-10-30 17:00 report §2–3.

S-109: Start the postclose cron at ≥ 16:35 ET · PROPOSAL (ops; complements S-108 and S-107)

• Rule: change the EDT twin 20 20 → 35 20 and the EST twin 20 21 → 35 21. Keep the 16–17 ET window.
• Why: the only stale run started at 4:21 PM. Every 4:34 PM start was clean (n 6). The VIX complex settles at 16:15, and Yahoo's daily bars appear to lag after that. The evidence is thin (n 1 failure), so S-108 is the primary fix.
• Evidence: 2026-10-30 17:00 report §2.

S-110: Stale-run flag and scorecard fallback · PROPOSAL (scorecard hygiene; extends S-098)

• Problem: storedRunOutcomes (lib/research/scorecard.ts L237) prefers postclose > manual > premarket picks for a session date. The stale 10-30 postclose run will therefore be graded as the live record. That includes MR names that did not qualify on final data, tagged vx: stress, which pollutes S-098's stress-day statistics.
• Rule:
• (1) Allow dataQuality: 'stale' or supersededBy on a run.
• (2) The scorecard skips flagged runs and falls back to the next run for that date.
• (3) Regime-conditioned statistics exclude flagged runs.
• Optional: a one-time forced re-run for 10-30 (/api/run?session=postclose, behind auth), if the 6 PM session approves.
• Evidence: 2026-10-30 17:00 report §2, §6.

S-111: Earnings-proximity tag for mean_reversion picks in the engine · PROPOSAL (from whiggity; extends S-071 / S-088 / S-106)

• Problem: lib/engine.ts and lib/execute.ts never read /api/events, so MR picks are not tagged or excluded by distance to their own print. S-071 covers only a print between the signal close and the entry open. S-088 covers Friday momentum entries on Monday reporters.
• Example: the corrected 10-30 MR list includes PLTR, which prints Mon 11-02 AMC (Palantir IR; Motley Fool, 2026-10-30), and SHOP, which prints Tue 11-03 BMO (Shopify IR, 2026-10-06).
• Rule (proposal):
• (1) The engine joins /api/events and stores sessionsToReaction on each candidate.
• (2) MR candidates whose own reaction session is ≤ 2 sessions after the entry day are tagged print-adjacent. The dashboard shows them and sizes them at half risk.
• (3) If S-106's replay confirms negative edge (earlier pre-earnings MR: n 27, mean −0.29%), exclude them instead.
• Evidence: 2026-10-30 17:00 report §7; 16:00 report §9–10.

S-112: Day-of-week-adjusted VIX9D/VIX regime threshold · PROPOSAL (from whiggity, verified by my calc; extends S-094 / S-102)

• Finding (Yahoo ^VIX/^VIX9D daily closes, 2016-10-31 → 2026-10-30, n 2,514, my calc):
• mean ratio: Mon 0.953, Tue 0.949, Wed 0.935, Thu 0.933, Fri 0.903;
• share of days > 1: Mon 29%, Fri 14%;
• paired changes: Fri − Thu −0.031 (t −13.8); Mon − Fri +0.044 (t +15.4).
• Why it matters: a fixed 1.0 line under-calls stress on Fridays and over-calls it on Mondays. Friday 10-30's final 0.991 was the 83rd percentile among Fridays, ≈ 1.022 after adjustment.
• Rule (proposal):
• (1) Compare r9 with a weekday-specific baseline. Adjusted ratio = r9 − (mean_dow − mean_all), using rolling 5-year means.
• (2) Or use a continuous score: the percentile of r9 within its weekday.
• (3) Feed the score into S-094's continuous risk scalar rather than flipping labels.
• Combine with S-102's borderline band.
• Caveat: the cause (weekend theta in 9-day options) is whiggity's explanation and was not tested. Check whether the Friday dip shrinks in weeks with a holiday Monday.
• Evidence: 2026-10-30 17:00 report §9.

S-113: Timestamp gate on the last daily bar (≥ 15:59 ET for equities, ≥ 16:15 ET for the VIX complex) · PROPOSAL (from whiggity; complements S-108 / S-109)

• Rule (proposal): in postclose runs, reject any symbol whose Yahoo meta.regularMarketTime (or last 5-min bar) is before 15:59 ET for equities and ETFs, or before 16:15 ET for ^VIX/^VIX9D/^VIX3M. Retry, then fall back to S-108's 5-min reconstruction, and count the symbol in staleCount.
• Why: Cboe disseminates VIX in RTH "between 09:31 a.m. ET and 4:15 p.m. ET" (Cboe VIX methodology v6.0, rev. 2026-02-26). The 10-30 4:21 PM run carried a ~15:35 VIX. whiggity's "Yahoo is unreliable before ~4:45 PM" is unverified: n 1 stale at 4:21 and n 6 clean at 4:34. So keep S-109's ≥ 16:35 start and rely on this gate, not a later cron.
• Evidence: 2026-10-30 17:00 report §8 and §2; S-108 and S-109 (above).

S-114: Cluster-penalized sizing for same-side MR candidates · PROPOSAL (from whiggity; extends S-078 / S-085)

• Rule (proposal): when ≥ 2 same-side candidates in one bucket share a TRAI correlation cluster, size the group to the risk of one position, scaled by 1/√(n_eff). Here n_eff = n / (1 + (n−1)·ρ̄), with ρ̄ the mean pairwise 60d ρ. Show the group on the dashboard as one bet.
• Why: on the corrected 10-30 MR list, ASML/ANET/ETN/AMAT are all cluster 1, with pairwise ρ 0.62–0.80 (mean 0.73, so n_eff ≈ 1.3 for 4 names; /api/correlations, asOf 10-30). whiggity's inclusion of PLTR is contradicted: it is in software cluster 5, with ρ 0.14–0.32 to the four. GS/JPM (cluster 2) and HOOD (ρ 0.60 to GS) form a second group.
• Evidence: 2026-10-30 17:00 report §10, §3.

S-115: whiggity's 17:00 test ideas: dual-listed pre-open arbitrage, Monday crypto context, first-day c→o vs o→c · NEEDS DATA (open research question; from whiggity, 17:00 report consult)

• (a) Dual-listed names (unverified): is US open→close MR edge smaller for names with a pre-US-open home session (ASML on Euronext 09:00–17:30 CET, i.e. 03:00–11:30 EST from 11-02; Euronext hours) than for US-only names? Test: per-name MR replay o→c, dual-listed vs not.
• (b) Monday crypto context (verified relationship): HOOD's weekend open gap vs BTC (Fri 16:00 → Mon 9:30 ET): ρ 0.67, β 0.76; controlling for QQQ, β 0.40 (t 5.4, n 104; my calc, Yahoo 2024-11 → 2026-10). The beta is not larger than the weekday overnight beta (0.63 controlled). Proposal to test: show the BTC weekend move next to cluster-7 names (HOOD, COIN) in the Monday premarket run.
• (c) First trading day (fold into S-103): split bucket replays into close→open and open→close on the first trading day of each month.
• (d) PSX/refiners around OPEC+ (open): cracks weaken when crude rises relative to products (CME crack-spread primer). But Δcrack vs crude was ρ 0.04 over the past year, and PSX's crude β is +0.24 (t 6.6; my calc). Test PSX/MPC gaps after OPEC+ decision weekends, split by crude gap sign and crack change.
• Evidence: 2026-10-30 17:00 report §2–4, §10.

S-116: Contrary-overnight-catalyst check for momentum candidates · PROPOSAL (extends S-004 / S-005)

• Rule (proposal): in the premarket run, flag any long- (short-) momentum candidate whose pre-market move is ≤ −1.5% (≥ +1.5%) against the signal direction, or whose sector driver moved against it overnight (e.g., crude ≥ 3% against an energy long). Flagged names need the normal post-open confirmation (above OR30 and VWAP) before entry; no entry on the open.
• Why: on 11-02 two of six LM names gapped against Friday's signal for identifiable overnight reasons: PSX −1.7% (Brent −4% on Iran–Qatar talks) and TMUS −2.8% (SpaceX–Boost Mobile report). On WTI −4% days (n 63, 2022–26), PSX's average open→close was −0.84% with a 35% hit rate (full-day conditioning; see S-117).
• Evidence: 2026-11-02 09:00 report §3, §5d, §7.

S-117: Energy open→close conditional on the pre-open crude move · NEEDS DATA (open research question)

• Question: when crude is down ≥ 3% from the prior settle at 09:25 ET, do energy names (PSX/MPC/XOM/CVX/COP/EOG/OXY/SLB) keep falling after the open, or fill?
• Why open: the 11-02 study conditioned on the full-day WTI move (n 63; energy o→c −0.56% to −1.01%), which leaks crude's intraday move into the condition. Needs intraday CL futures history (pre-open snapshot) to test fairly.
• Evidence: 2026-11-02 09:00 report §3.

S-118: MAR print → BKNG/ABNB read-through · NEEDS DATA (open research question)

• Question: does Marriott's BMO RevPAR surprise/guide change predict the sign of BKNG's and ABNB's next earnings reaction or the same-day move? 11-02: MAR raised FY RevPAR to 4.5–5.0% from 3.0–3.5% (+4.1% pre-market); BKNG prints Tue 11-03 AMC, ABNB Thu 11-05 AMC.
• Test: IR-verified MAR print dates (last 12), MAR reaction vs BKNG/ABNB same-day and next-print reactions. Not yet run.
• Evidence: 2026-11-02 09:00 report §5b.

S-119: Tag short-momentum signals by cause and split the replay before any re-activation · PROPOSAL (extends S-002 / S-005)

• Rule (proposal): add a signalCause field to each short_momentum (and long_momentum) row:
• own_earnings: reaction day per /api/events
• pre_print: own print within 5 sessions, no own news
• no_event

Report the scorecard replay separately for each cause. Any decision to un-suspend short momentum should rest on the subgroup results, not the pooled −0.45% mean.

• Why:
• The 10-30 SM list mixes causes: RBLX (own print, −18.1%) and DDOG/APP/TEAM (no company news, printing 11-04/11-05).
• On day 2 (11-02 ~10:15 ET) they split. RBLX +2.8% and APP +3.5% reversed; DDOG −4.6% and TEAM −2.2% continued.
• Base rate (my calc, n 20 software/internet ≥10% post-earnings drops, 2025-11 → 2026-10): day-2 o→c +1.04% avg / +0.52% median, 10/20 up, t 1.5. Shorting the flush has no edge, and neither does buying it.
• Pooled replay: n 415, net −0.45%, t −3.62 (/api/scorecard).
• Evidence: 2026-11-02 10:00 report §3, §4.

S-120: Day-2+ drift of "pre-print sympathy losers" vs earnings flushes · NEEDS DATA (open research question; extends the 10-30 10:00 test idea (b) and S-029)

• Question: when a name falls ≥ 8% with no company news and its own print is ≤ 5 sessions away (DDOG, APP, TEAM on 10-30), does it keep underperforming its sector ETF into the print? And does an own-earnings ≥ 10% flush (RBLX) revert on day 2?
• Why: on 11-02 the no-news cluster-5 names kept falling (DDOG −5.4pp vs QQQ at ~10:15 ET) while the flush bounced. That is one morning, n 4.
• Test:
• Yahoo daily for the TRAI Software/Internet names over 2+ years.
• Identify no-news ≥ 8% drops by excluding /api/events reaction days; "no news" is a proxy.
• Measure day-2 o→c and day-2 → T−1 excess vs IGV/QQQ.
• Compare with earnings-flush cases from §4 of the report.
• Evidence: 2026-11-02 10:00 report §2–4.

S-121: "Gap-captured reversion" for MR fade-longs, plus an S-100 re-test · PROPOSAL (extends S-103 / S-115(c) / S-100)

• Rule (proposal):
• (a) For each fade-long, log gapCapture = next-day 09:30 gap ÷ |signal-day move| and the 10:00 price vs VWAP/OR30.
• Split the replay by gapCapture tercile, measuring the TRAI-tradable window (10:00 → 15:55), not the open → close.
• If the top tercile shows no post-10:00 edge, skip or half-size fade-longs that have already captured ≥ ~40% at the open.
• (b) Re-run S-100 on a fixed event-date list.
• Why:
• On 11-02, GS (46% captured), JPM (~60%) and HOOD (~9%) all gapped up. By 11:15 ET they were flat from the open, at VWAP and inside OR30. The reversion happened before TRAI could enter.
• Separately, the current replay (05-01 → 10-29) does not reproduce S-100: earnings-reaction LM averaged +0.38% gross (n 48, hit 58%) vs +0.25% for non-earnings (n 330) (my calc). S-100 had −0.42% vs +0.27% on 04-23 → 10-21. Neither is significant.
• Evidence: 2026-11-02 11:00 report §3, §6.

S-122: Peer-print flag for singleton momentum names · PROPOSAL (extends S-004 / S-085)

• Rule (proposal): if ≥ 2 of a candidate's top-3 60d-ρ peers report within the next 2 sessions, tag it peerPrintRisk on the dashboard and in /api/events. Show the peer dates. Do not change sizing until it is tested (see the question below).
• Why:
• REGN (LM, singleton cluster 50) has its top three correlates printing within ~30 hours: VRTX 11-02 AMC (ρ 0.61), PFE 11-03 BMO (0.60), AMGN 11-03 AMC (0.48).
• S-085 only fires for names inside a multi-name cluster, so singletons miss it.
• Today REGN decoupled from peers (+3.6% vs XBI −2.4%), so the size of the read-through is untested.
• Test: compare next-day o→c and day-2→3 gaps for LM/MR names with and without peer prints, using rolling (not current) correlations.
• Evidence: 2026-11-02 11:00 report §4(c–d).

S-123: HOOD/crypto-cluster fade-longs: split day-2 beta into equity and BTC legs · NEEDS DATA (open research question; extends S-115(b))

• Question: for cluster-7 fade-longs (HOOD, COIN, MSTR), does BTC's 09:30→15:55 return explain day-2 o→c beyond SPY/XLF beta? Should a BTC overnight filter (Fri 16:00 → 09:30) gate entries?
• Why: on 11-02, HOOD was +1.47% at 11:15 ET vs a naive beta-implied ~+2.8% (beta60 4.24 × SPY +0.67%), with BTC −0.66% since Friday's close and COIN −2.26%. HOOD–JPM ρ is 0.08, so HOOD is not a bank trade.
• Test: Yahoo 5-min HOOD/COIN/SPY/XLF plus BTC-USD hourly, on replay fade-long dates.
• Evidence: 2026-11-02 11:00 report §3.

S-124: Day-2 split inside cluster 1 by business line after a semis shock · NEEDS DATA (open research question; extends S-001 / S-011)

• Question: on the session after a cluster-1 shock, do AI compute/memory/equipment names (AMD, MU, SMCI, LRCX) and analog/handset/power names (ON, QCOM, MCHP, TXN) behave as different factors? Should bucket entries and the cluster cap treat them separately for that session?
• Why: on 11-02 at 12:10 ET, after SOX −7.2% on 10-30, all names gapped up ~0.7–2.2% and then split: AMD −3.1%, MU −3.1%, SMCI −3.2%, LRCX −2.2% (all below OR30 lows) vs ON +2.3%, QCOM +1.9%. The short QCOM worked against TRAI; the fade-longs ASML/AMAT/ANET did not bounce. Sourced drivers differ by line: AI-capex doubts (Intel/Amazon) and CXMT DRAM competition for compute/memory; memory cost for networking/system margins (ANET, SMCI).
• Test: 5-min bars on past SMH ≤ −5% days; sub-cluster day-2 o→c and VWAP-relative paths; check whether a 60d hierarchical cut at higher ρ already separates them.
• Evidence: 2026-11-02 12:00 report §2–§3.

S-125: Negative-beta co-movers count as one cluster for the long-momentum cap in stress regimes · PROPOSAL (extends S-001 / S-014 / S-049)

• Rule (proposal): when the regime is stress or VIX9D/VIX > 1, group long-momentum candidates with beta60 < 0 that are all up on the signal day into one "low-beta/anti-AI" cluster for the "1 per correlated cluster" cap. Show it on the dashboard next to the sector flag. Don't change sizing until it is tested.
• Why:
• The 10-30 LM list has CMCSA (beta60 −0.34), TMUS (−0.79) and SPOT (−0.87). Pairwise 60d ρ is only 0.32–0.52, and they sit in two different correlation clusters (9 and 8).
• TRAI's sector flag "Internet/Media ×2" pairs CMCSA+SPOT and misses TMUS. The 60d clusters pair CMCSA+TMUS and miss SPOT.
• On 11-02 all three were up together again (+2.26% / +1.57% / +1.05% at 13:10 ET, all above OR30 and VWAP), with non-candidates CHTR +3.5% and DIS +2.4%, and no company catalyst. They behave as one factor bet against AI hardware.
• Test: on replay LM days, compute the realised same-day ρ of o→c returns among beta60 < 0 picks vs other pairs. Compare the portfolio variance of the top-3 with and without the grouping.
• Evidence: 2026-11-02 13:00 report §2, §4.

S-126: Reaction-day fade-long for BMO reporters that gap down ≥ 2.5% · NEEDS DATA (open research question; extends S-005 / S-106)

• Question: for BMO reporters in the universe that gap down ≥ 2.5% on the print, is 10:00 → 15:55 positive on average, after the opening range forms? Does it hold only for names with positive estimate revisions or a recent pre-print drawdown?
• Why:
• ETN's own history (my calc, Yahoo daily, n 19 prints 2022-02 → 2026-07; pre-2025 dates inferred by volume): after down-gaps ≤ −2.5% (n 7), day-1 o→c averaged +2.47%, 6 of 7 up. Across all prints, the day-1 o→c went against the gap 12/19.
• TRAI's last-4 for ETN shows 0% continuation (/api/events).
• TRAI's aggregate (S-005) had day-1 in-gap-direction −0.16% over n 622, so ETN may be an outlier rather than a rule.
• ETN reports 11-03 BMO with a ±7.48% implied move: a live case.
• Test: use /api/events reaction days (gap, day1Drift) for all BMO names: bucket by gap ≤ −2.5% / −5%, and measure o→c and 10:00 → 15:55 from 5-min bars where available. Gross, then with costs. Research only; no entry rule.
• Evidence: 2026-11-02 13:00 report §6.

Evidence added 2026-11-02 13:00 to existing items

• S-005 / S-100: defensive media/telecom names up ≥ 3% on QQQ ≤ −2% days showed no day-2 drift. Next-day o→c −0.14% mean, +0.06% median, 50% up (n 16, excluding the 10-22 cluster hit by the 10-23 SpaceX story; n 23 incl.: −0.33%) (my calc, Yahoo 5y). Report §5.
• S-106 live case: ETN, a fade-long one session before its own BMO print, was +0.33% at VWAP at 13:10 ET. Its print falls after TRAI's 15:55 flat time, so there is no overnight gap exposure. The eve-of-print session historically averages +0.26% o→c (n 19, my calc). Report §6.

S-127: Pre-print-day move as a predictor of the earnings reaction · REJECTED as a signal (recorded so it isn't re-tested blindly; extends S-005)

• Question: does a ≥ 4% drop (or rise) in the last session before a print predict the reaction gap's direction or size?
• Result (my calc, TRAI `/api/events` history, n 624): pre-day ≤ −4% (n 36): mean gap +1.28%, 56% gap up; flat pre-day (n 464): −0.35%, 49% up. Normalised by each name's other prints, median |gap| ratio is 0.89 in both groups. Bigger raw gaps after big pre-day moves are a volatile-name effect, not a timing signal.
• Use: don't size up or skew direction on names that sold off into their print (AMD/SMCI on 11-02 are live cases).
• Evidence: 2026-11-02 14:00 report §5.

S-128: Flag gap risk by the larger of TRAI's history and the options-implied move · PROPOSAL (extends S-005)

• Rule (proposal): when TRAI's last-4 avg |gap| is less than half of a dated public options-implied move, label gap risk by the implied move in /api/events and on the dashboard, with the source and snapshot date.
• Why: ON prints 11-02 AMC with options pricing ~±10–11%, vs TRAI's last-4 avg |gap| of 2.73% (gaps −1.1/−0.5/−4.3/+5.0%). PLTR (implied ~11.5–13% vs 9.6%) and AAPL (~4% vs 3.65%) are much closer.
• Test: after tonight's prints, compare ON's realised gap with both numbers; across the next two weeks of prints, record implied vs TRAI history vs realised.
• Blocker: needs a free, dated implied-move source; until then the hourly reports record implied moves by hand.
• Evidence: 2026-11-02 14:00 report §3, §8.

S-129: Range-scale the next session after a flat-spot front-end inversion · PROPOSAL (extends S-092 / S-094 / S-112)

• Rule (proposal): when the VIX9D/VIX close is ≥ 1.05 on a day SPY moves < 0.5%, scale the next session's stop/ATR multiple up ~1.5× or riskPerTradePct down by the same factor, regardless of the regime label. Feed it into S-094's continuous scalar, adjusted for weekday (S-112).
• Why (my calc, Yahoo 10y): all such days (n 68): next-day SPY |o→c| 1.10% vs 0.57% for all days, direction ≈ 0 (o→c −0.07%, 54% up). At the same VIX level (19–24): ratio ≥ 1.05 (n 27) → next |o→c| 0.95% vs 0.64% for ratio < 1.0 (n 149). Pre-election analogs 2024-10-29/30/11-04 split both ways on direction.
• Live case: 11-02 at 14:55 ET, VIX 21.21, VIX9D 22.54 (≈ 1.06) with SPY flat into the midterms.
• Evidence: 2026-11-02 15:00 report §6B.

S-130: Tag oil-shock-driven momentum signals and don't chase them on day 2 · PROPOSAL (extends S-116 / S-117)

• Rule (proposal): if crude (CL) moved ≥ 5% on the signal day, tag LM/MR candidates in oil-sensitive groups (airlines, cruise, OTAs, refiners, E&Ps) as macroDriven: oil. Require the normal post-open confirmation (above OR30 and VWAP) and don't size up.
• Why (my calc, WTI ≤ −5% days, 5y, n 39): next-day o→c JETS +0.07% (51% up), DAL +0.15%, RCL +0.17%, XLE −0.03%, PSX −0.25% (41% up); OTAs slightly positive (BKNG +0.40%, EXPE +0.57%, not significant). The move is a one-day gap.
• Evidence: 2026-11-02 15:00 report §6A.

Evidence added 2026-11-02 15:00 to existing items

• S-122 / S-085 live case: PSX (LM, own events row shows no print, own avg |gap| 0.99%) vs MPC (ρ 0.91) printing Tue 11-03 BMO with a TRAI max |gap| of 8.52%. Report §5, §7.
• S-128: BKNG implied ±7.5% vs TRAI 4.46% (ratio 0.6: no trigger, but large). Report §5.

S-131: Use the ≥ 16:30 ET after-hours move as the reaction-gap estimate for AMC reporters · PROPOSAL (extends S-085 / S-128)

• Rule (proposal): for an AMC reporter, record the last extended-hours price at 16:30 ET (or later) vs the close. If \|AH\| ≥ 3%, set the expected reaction gap to ≈ 1.2 × AH in the same direction, and use it (not the last-4 avg \|gap\|) for the event-risk flag and the next morning's open review. Ignore prints before 16:30 and thin first prints. If \|AH\| < 3%, keep the history/implied estimate.
• Why (my calc, n 36 AMC prints in /api/events, 09-10 → 10-29, Yahoo 5-min): 16:30 AH vs gap: sign match 29/36, r 0.85, median \|gap\|/\|AH\| 1.19. With \|AH\| ≥ 3%: 18/20 sign match. At 16:15 it is weaker (27/36, one ≥ 3% sign miss: IBM). Day-1 o→c in the gap direction ≈ −0.45%, so no follow-through edge. Live: VRTX's +4.6% at 16:09 on ~7k shares was −1.1% by 16:32.
• Blocker: needs an AH quote source in the post-close run (Yahoo is unofficial). Small n, one 8-week window.
• Evidence: 2026-11-02 16:00 report §6A.

S-132: Peer read-through: tag ρ ≥ 0.7 peers of a big-gap reporter with an expected gap · PROPOSAL (extends S-122 / S-085)

• Rule (proposal): when a reporter's expected gap (S-131 or a realised BMO gap) is ≥ 5%, tag every LM/MR candidate with 60d ρ ≥ 0.7 to it with an expected gap of 0.26× (ρ 0.7–0.8) or 0.32× (ρ ≥ 0.8) of the reporter's gap, in the same direction. Apply S-122's no-entry-before-10:00 rule if the tag is ≥ 1.5%.
• Why (my calc, reporters with \|gap\| ≥ 5% since 07-01, 103 peer-events / 46 reporter events): 93% same sign; median peer \|gap\| 2.56%; ratio 0.26 (ρ 0.7–0.8, n 68) / 0.32 (ρ ≥ 0.8, n 35). Peer o→c in the reporter's direction is small (+0.18% to +0.43%). Example: TXN −5.74% (07-23) → ON −3.76%, MCHP −4.48%, ADI −2.74%.
• Caveats: ρ look-ahead (today's 60d), no SPY-gap subtraction, pairs not independent.
• Live case: ON AH −5.9% (16:32 ET 11-02) → ADI/MCHP/TXN expected ~−2% [CALC]. None of them is a candidate.
• Evidence: 2026-11-02 16:00 report §5, §6B.

Evidence added 2026-11-02 16:00 to existing items

• S-128: ON (implied ~10–11%, TRAI 2.73%) traded −5.9% to −6.5% AH, confirming the premise live. Tuesday: PFE implied ±5.2% vs TRAI 1.78% (2.9×, fires; PFE–REGN ρ 0.60). Report §3, §4.
• S-129: close values VIX9D/VIX 22.91/21.42 ≈ 1.07, \|SPY\| 0.47%: the trigger fires for Tue 11-03 (Election Day). VIX 21.42 < 22, so S-092 does not fire in code. Report §2, §7.
• S-122: PSX closed below VWAP and below its OR30 low into MPC's BMO print. Report §2.
• Evidence add (2026-11-06 11:00): Mon peer maps OXY→EOG/COP/DVN (ρ≥0.89), KKR→BX 0.85, MARA crypto cluster. report §5.

S-133: Use the latest after-hours print (≥ 18:00 ET, ideally 19:55) as the AMC reaction-gap estimate · PROPOSAL (amends S-131)

• Rule (proposal): keep S-131's 16:30 ET read only as a provisional flag. The gap estimate for an AMC reporter is the last extended-hours price at ≥ 18:00 ET (19:55 preferred) vs the close, scaled ≈ 1.0 (not 1.19). Re-run the event-risk flag and S-132 peer tags off that number.
• Why (my calc, same 36 AMC prints as S-131, Yahoo 5-min): r with the reaction gap rises 0.85 (16:30) → 0.88 (17:00) → 0.91 (17:30) → 0.94 (18:00) → 0.95 (19:55); median |gap − AH| falls 2.57 → 1.54 pp; median |gap|/|AH| falls 1.19 → 0.99. Most AMC calls start 16:30–17:00, after the 16:30 read.
• Live case: ON −5.9% at 16:32 → −2.44% at 17:05 ET during its 5 PM call (Yahoo, Public.com). No in-sample ≥ 3% 16:30 move had halved by 19:55 (0/20), so this is the first such case if it holds.
• Blocker: same as S-131 (AH quote source); the post-close cron runs at ~16:49, so this needs a second, ~20:00 ET pass.
• Evidence: 2026-11-02 17:00 report §3, §6A.

S-134: Tag Tuesday-type BMO peer links only at ρ ≥ 0.7; show 0.5–0.7 as "watch" · PROPOSAL (extends S-132 / S-122)

• Rule (proposal): BMO-print read-through tags on LM/MR candidates require 60d ρ ≥ 0.7 (as S-132). Links of 0.5–0.7 are shown on the dashboard as "watch" with no sizing effect.
• Why (my calc): on 8 MPC/PSX reaction days, XOM/CVX/OXY (ρ 0.50–0.54 to MPC) gapped −2.2% to +1.3% with poor sign agreement (MPC −8.52% on 2025-11-04 → XOM −0.33%, CVX −0.40%). On 7 FISV gaps ≥ 5% (2y), PYPL (ρ 0.29) matched sign 2/7, median ratio 0.07. Small n.
• Live case: Tue 11-03 BMO (MPC, PFE, ETN, SHOP, FISV) vs the 11-02 post-close candidates: highest links PFE→REGN 0.60 and MPC→CVX 0.54. None reaches 0.7.
• Evidence: 2026-11-02 17:00 report §4, §6B.

Evidence added 2026-11-02 17:00 to existing items

• S-111: BKNG is on the 11-02 post-close MR fade-short list and prints Tue 11-03 AMC (sessionsToReaction 2), the live print-adjacent case. Report §2, §7.
• S-128: PFE (S-128 fires, implied 2.9× TRAI history) traded +1.57% AH on 11-02 with no news found, before its Tue BMO print. Report §4.
• S-129: engine values at 16:15 ET: VIX9D/VIX 1.062, |SPY| 0.48%. The trigger holds on the engine's numbers too. Report §2.
• S-132: with ON back to −2.4% AH by 17:05, the reporter estimate falls below S-132's 5% threshold, so the ADI/MCHP/TXN tag would not fire. Report §3.

S-135: 09:00 ET pre-market refresh of every reporter's gap estimate · PROPOSAL (extends S-133 / S-128 / S-132)

• Rule (proposal): at ~09:00 ET, re-read the pre-market price of every AMC (prior evening) and BMO (same morning) reporter in the universe, replace the S-133 after-hours estimate with it, and re-run the event-risk flag and the S-132 peer tags before the 10:00 entry. Never use a snapshot taken inside the call window (≈ 16:30–17:30 ET for most AMC calls) as the estimate.
• Why (my calc, 36 AMC prints, Yahoo 5-min): r with the open gap is 0.98 at 09:00 vs 0.95 at 19:55; 5/36 flipped sign between 19:55 and 09:00. Live case 11-02/03: AAPL +0.2% at 17:00 → −2.8% at 19:55 → −4.1% at 09:00; PLTR +1.2% → −3.2% → −6.2% (AAPL's gross-margin guide came in the call's prepared remarks).
• Blocker: the 08:58 premarket run already exists; it needs extended-hours quotes for reporters (same source issue as S-131).
• Evidence: 2026-11-03 09:00 report §3, §5.

S-136: Overnight extension as a day-1 continuation filter for AMC reporters · NEEDS DATA (open research question; extends S-135)

• Question: when the 09:00 pre-market move extends the 19:55 AH move by ≥ 1 pp in the same direction, does the reaction day's open→close continue?
• First look (my calc): extension ≥ +1 pp: continued 8/11, mean +0.67%; < +1 pp: 11/25, mean −0.81%; linear r 0.02. n 36 on 10 dates. Needs a longer window, the 10:00→15:55 TRAI window instead of open→close, and BMO reporters (09:00 move vs 08:00).
• Live cases to log today: AAPL (+1.3 pp), PLTR (+3.0 pp), ON (flipped −0.5% → +3.2%).
• Related tests (from whiggity, 11-03 09:00 consult): split AMC gaps by whether most of the move formed after the call started vs right after the release (needs call-start times); energy fade-longs with vs without a same-morning diplomacy/supply headline on a > 1% pre-market crude move.
• Evidence: 2026-11-03 09:00 report §5. Script: scripts/research-calcs/2026-11-03-overnight-extension-amc.py.

Evidence added 2026-11-03 09:00 to existing items

• S-133: 19:55 read had the right sign for AAPL, PLTR, VRTX on the 11-02 prints; the 17:00 read (in-call) was wrong-signed for AAPL and PLTR. ON's −6.3% at 16:30 became +3.2% by 09:00: the first in-sample case of a ≥ 3% 16:30 move reversing. Report §3.
• S-134: FISV +9.8% pre-market on its BMO print; PYPL (ρ .29) +0.15%. MPC −2.2% vs CVX/OXY/XOM −1.0% to −1.3%, moving with crude. Report §6.
• S-130 / S-117: energy fade-longs OXY/XOM/CVX open a further −1.0% to −1.3% with crude −1.4% pre-market on day 2 of the oil shock. Report §2, §7.
• S-119: semis +3% pre-market rebound runs against the (suspended) short-momentum names AMD/ORCL/CSCO; tag as "sector-rebound day" in the SM replay. Report §4.

S-137: Big up-gap LM reaction / day-2 after own AMC beat — allow vs skip under S-005 / S-006 / S-066 · PROPOSAL (extends S-005 / S-006 / S-066)

• Rule (proposal): when a long_momentum candidate's signal day (or the prior AMC session) is its own earnings beat that gaps / opens ≥ +15% on the reaction day:
• Do not add a drift premium (S-005 stands).
• Do not fade the gap same-day (S-006 stands when |gap| > 2 ATR).
• Apply S-066's day-after-big-up guard for any new LM entry on day 2 / day 3 unless the name re-qualifies after 10:00 on fresh RelVol/OR30/VWAP confirmation and residual is still positive vs the post-gap open (not vs the pre-print close).
• Log the reaction-day open→15:55 and day-3 o→c for the APP-class case file.
• Why (live case): APP printed AMC 11-04 (rev $2.185B +55%, EPS $4.30, Q4 guide $2.365–2.395B), indicated +23.9% into 11-05, opened ~+28.8% and traded ~+29.5% by 15:08 ET. By 15:00 the move is done for a flat-by-3:55 book; chasing into the close is not an S-005 edge.
• Test: collect AMC beats with reaction open ≥ +15% in the TRAI universe over 2y; measure day-2 10:00→15:55 and day-3 o→c vs the no-earnings LM base rate.
• Evidence: 2026-11-05 15:00 report §3–4, §7.

S-138: Pre-print same-day weakness into AMC as a gap-risk modifier (not a sign predictor) · PROPOSAL / NEEDS DATA (extends S-128 / S-005; distinct from rejected S-127)

• Rule (proposal): on an AMC reporter's print day, if the same-day last (by ~15:00 ET) is ≤ −8% vs the prior close, set the overnight gap-risk flag to max(TRAI last-4 avg |gap|, options-implied if known, |pre-print move|). Show the pre-print move on /api/events and the dashboard. Do not use the pre-print sign to predict the reaction-gap sign (S-127 stays rejected).
• Why (live cases): MCHP ~−12 to −15% into tonight's Q2FY27 call (avg |gap| only 5.7%); NET ~−13% into tonight's print after DDOG −14% (avg |gap| 11.6%). The pre-print move already exceeds MCHP's historical avg |gap|, so history alone understates overnight risk.
• Test: for AMC reporters with same-day ≤ −8% into the print (2y), compare realised |gap| vs last-4 avg |gap| vs |pre-print|; also check whether |pre-print| ≥ avg |gap| raises |gap| overrun rate.
• Caveat: S-127 rejected pre-print as a directional predictor; this item only resizes the risk flag.
• Evidence: 2026-11-05 15:00 report §6–7.
• Evidence add (2026-11-06 11:00): MARA Thu −14.4% and HIMS Thu −9.8% into Mon AMC — size modifiers (not sign). Fri HIMS bounce does not clear overnight risk. report §4, §6.

S-139: Cluster-5 peer gap-risk after a ≥ 10% guide-crush into a ρ ≥ 0.7 peer's print night · PROPOSAL (extends S-132 / S-122 / S-004)

• Rule (proposal): when a cluster-5 name gaps / closes ≤ −10% on a guide-driven print, and a ρ ≥ 0.7 peer prints AMC within 1–2 sessions, tag that peer with expected gap ≥ 0.26× the crushed name's reaction (S-132 coefficients) or at least the peer's own last-4 avg |gap|, whichever is larger, and apply S-122's no-new-entry-before-10:00 on the reaction morning for LM/MR candidates in the same cluster.
• Why (live case): DDOG −14.42% on 11-04 (Q3 beat but Q4 guide $1.19–1.20B; secondary sources cite largest-AI-customer contractual minimum) into NET AMC 11-05 (DDOG–NET ρ 0.78; NET avg |gap| 11.6%). NET itself is already −13% pre-print. Peers ZS/CRWD/PANW (ρ 0.74–0.76 to NET) inherit Friday open risk.
• Test: cluster-5 (and analog high-ρ software) pairs where reporter |reactionCC| ≥ 10% and peer prints within 2 sessions; measure peer |gap| and sign agreement.
• Evidence: 2026-11-05 15:00 report §5–7.

Evidence added 2026-11-05 15:00 to existing items

• S-005 / S-066: APP is the live big-up-gap LM reaction-day case; move largely complete by 15:00 — no drift premium into the flat. Report §4, §7.
• S-006: APP ~+29% reaction day — do not fade same-day. Report §4.
• S-116: QCOM was LM on 11-04 close with indicatedGapPct −11.8% into its own handset-miss print — contrary overnight catalyst confirmed real. Report §4.
• S-127 (rejected): MCHP/NET pre-print weakness is logged as a risk-size input (S-138), not re-opened as a sign predictor.
• S-132: NET print night after DDOG −14% is the live ρ ≥ 0.7 peer case (0.78). Report §5–6.

S-140: BMO energy/utility reaction → same-day SM/LM conflict check with overnight AMC peers · PROPOSAL / NEEDS DATA (extends S-122 / S-132)

• Rule (proposal): when a cluster-4 energy name or cluster-16 utility prints BMO, record the same-day reaction by ~15:55 and tag overnight AMC peers (ρ ≥ 0.7) with a peer-backdrop flag: muted/negative BMO reaction raises overnight gap-risk weight for AMC peers even if the BMO print was a fundamental beat; strong BMO reaction does the opposite. Separately, if an LM name in the same regulated cluster (e.g. NEE) is live, require an explicit conflict note that the BMO peer print did not invalidate the LM signal before treating next-day continuation as clean.
• Why (live case): COP BMO beat (adj EPS $3.27, dividend +11%) but closed ≈ −1.1% into EOG/DVN AMC the same night (COP–EOG 0.938). DUK/SO quiet greens left NEE’s 11-04 LM signal without a negative regulated-utility print shock.
• Test: cluster-4 BMO prints followed by same-day AMC peer prints (2y); compare peer |gap| when BMO reaction ≤ −1% vs ≥ +1% despite beat.
• Evidence: 2026-11-05 16:00 report §3–6, §8.

S-141: Multi-day crush into own BMO while SM suspended — resize gap-risk; do not reactivate SM · PROPOSAL / NEEDS DATA (extends S-138 / S-005; SM policy)

• Rule (proposal): if a name is in the suspended short_momentum residual list and falls ≥ 10% over the 1–2 sessions into its own BMO print, set overnight/open gap-risk flag to max(TRAI last-4 avg |gap|, |multi-day move|, options-implied if known) and apply S-122-style no-new-entry-before-10:00 for ρ ≥ 0.7 peers on the reaction morning. Explicitly: do not treat the SM residual as a short reactivation signal while the bucket is suspended.
• Why (live case): VST 11-04 −13.8% (residZ −6.84, SM suspended) then 11-05 −6.1% → ~−19% two-day into Fri 11-06 BMO, while TRAI avg |gap| is only 1.49%. CEG ρ 0.85 inherits Friday open risk.
• Test: SM-listed (or residual) names with ≥ 10% 1–2 day drawdown into own BMO over 2y; measure realised |gap| vs last-4 avg; also measure what a naive SM reactivation would have done (expected negative — documents why suspension stays).
• Evidence: 2026-11-05 16:00 report §7–8.

S-142: Leftover AMC pre-print modifier beyond the −8% software cases · PROPOSAL / NEEDS DATA (extends S-138)

• Rule (proposal): apply S-138’s gap-risk modifier (size flag by max(history, |pre-print|); never as a sign predictor) to non-software AMC leftovers when |pre-print| ≥ 0.75× the name’s last-4 avg |gap| or |pre-print| ≥ 5%, whichever triggers first. Log mild bids (e.g. DKNG +1.8%) separately as “pre-print support” without shrinking the historical |gap| flag.
• Why (live cases): AFRM −7.8% into 5.1% avg |gap|; ABNB −3.5% into 5.9%; DKNG +1.8% into 5.8%; EOG −1.6% / DVN −2.9% into 1.5%/2.9% energy avgs — same night as the 15:00 MCHP/NET software crush cases that motivated S-138.
• Test: AMC reporters outside semis/software with |pre-print| ≥ 5% or ≥ 0.75× avg |gap|; compare realised |gap| overrun rate vs baseline.
• Evidence: 2026-11-05 16:00 report §6, §8.

Evidence added 2026-11-05 16:00 to existing items

• S-138: AFRM −7.8% pre-print is a near-threshold non-software test case; VST multi-day crush into BMO is handled under S-141 rather than overloading S-138’s AMC wording. Report §6–7.
• S-132: CEG as VST ρ 0.85 peer into Fri BMO; ABNB–BKNG 0.81 into ABNB AMC. Report §6–7.
• SM suspension policy: COP/XOM/OXY/VST remain listed residuals — close wrap explicitly rejects reactivation after COP beat and VST day-2. Report §8.

Evidence added 2026-11-05 17:00 EST — deduplicated, research only

• S-132 / S-139 (with S-133 / S-135 timing): split-catalyst peer test. NET's issuer-guide raise and TEAM's also-raised FY guide but negative AH path must be recorded separately by economic exposure, guidance revision, and observation time. NET–ZS rho 0.76 does not automatically validate ZS's MR long; sparse ZS/DDOG AH bars did not inherit NET's positive sign. Compare to event-proximity-only baseline. Do not treat TEAM as a guide-cut case. S-139's guide-crush premise cannot be inferred solely from price direction.
• S-138: pre-print weakness is gap-risk context only. NET, TEAM, MCHP's divergent initial AH signs do not validate a sign predictor or an exact expected move based on n=4 history.
• S-012 / S-046 / S-098: extend existing residual/regime research rather than create another filter ID. Stress MR ZS raw z1d −2.23 versus residualZ −0.25 is a case for comparing raw-z extremes against idiosyncratic-residual MR under the 10:00 AM EST entry, 1 ATR stop / 2 ATR target, 60-minute time stop live rule. Historical o→c proxy results are not that test's outcome. Use availability-correct regime labels, costs, out-of-sample checks, day clustering and multiple-testing caution.
• S-068 / S-102 / S-108: retain source asOf/generatedAt, signal-session date and actual decision availability per feature. Latest stress is Nov 5 4:49:57 PM EST (0 stale); standalone regime/macro/correlations/scorecard remain Nov 4. Do not merge those clocks, relabel stale composite regime as current, or backfill postclose stress into same-day 10 AM decisions. This is evidence/provenance refinement, not a claim these proposals are newly implemented.
• Validation / suspension: live-rule 148 picks over only 9 days, mean net R −0.0620, day-bootstrap 90% CI [−0.1378,+0.0265]; MR only 8 days. No established edge, no SM reactivation, no overnight order or regular-session opportunity today. Whiggity was unavailable (site connection error twice, including one built-in Retry); no research reply or attributed suggestions.
• Source: 2026-11-05 17:00 report, especially §2, §4, §6–9. No new proposal ID and no live-engine/app changes.
• Evidence add (2026-11-06 11:00): non-software leftovers MARA/HIMS into Mon AMC mega-night; same modifier logic. report §4, §6.

S-143: After multi-day crush into BMO, issuer forward-guide raise → asymmetric peer gap-risk · PROPOSAL / NEEDS DATA (extends S-141 / S-132 / S-004)

• Rule (proposal): when S-141's multi-day crush (≥10% over 1–2 sessions into own BMO while SM suspended) is followed by an issuer raise of forward EBITDA/FCFbG (or clear equivalent growth/cash guide), keep ρ≥0.7 peer gap-risk sized for |gap| overrun into/through the peer's print, but do not auto-fade the reporting name on the reaction morning solely because of the pre-print crush. Log reporter reaction and peer co-move separately.
• Why (live case): VST ~−19% two-day into Fri 11-06 BMO, then raised/narrowed 2026 Ongoing Ops Adj EBITDA to $7.1–$7.5B and initiated 2027 $7.7–$8.5B (Vistra IR, 2026-11-06); TRAI indicatedGapPct +5.86% with SM still suspended. CEG ρ 0.845 prints Mon 11-09 BMO.
• Distinct from: S-141 (crush→resize + no SM reactivation); S-077 (MR fade-short skip on own beat-and-raise); S-006 (don't fade huge gaps same-day).
• Test: SM-residual crush→BMO cases split by guide raise vs cut/in-line; measure reporter open→10:00 / open→close and ρ≥0.7 peer |gap| vs baseline. Costs, day clustering, OOS, multiple-testing caution.
• Evidence: 2026-11-06 09:00 report §3, §7.

Evidence added 2026-11-06 09:00 EST — Fri open under stress

• S-141: VST live case completed the print: actual Ongoing Ops Adj EBITDA raise/narrow + 2027 initiation after the multi-day crush; SM remains suspended — bounce/indicated +5.86% is not reactivation evidence. Report §3, §5, §7.
• S-132 / S-004: CEG cluster-23 avgRho 0.845 peer into Mon 11-09 BMO after VST's positive revision morning; TRAI CEG indicatedGapPct +3.02% on the 8:58 run. Guide raise does not automatically clear peer |gap| sizing (feeds S-143 test). Report §6–7.
• S-006: MRNA/VST large indicated/secondary opens — do not same-day fade without data; n=4 histories fragile. Report §5, §7.
• SM suspension policy: semis SM residuals (TSM/KLAC/AMAT et al.) showing +7–9% indicated gaps after SMH −11.21% Thu — still suspended. Report §5.
• Validation: liveRules n 196 / 10 days, mean R −0.0524, enoughDays false; TARGET 0. No claimed new edge. Whiggity consult available (MRNA-focused; VST left UNV there — filled from issuer IR). Screening/research only.
• Source: 2026-11-06 09:00 report. No live-engine/app changes.

S-144: Under STRESS + VIX9D/VIX>1, regime-split before sizing MR fade-longs of non-earnings residual downs · PROPOSAL / NEEDS DATA (extends S-012 / S-046 / S-098)

• Rule (proposal): when the live regime label is stress and VIX9D/VIX > 1 (front-end inversion), do not size mean_reversion fade-longs on non-earnings residual downs (HON/TMO/AMGN class) until a regime-split backtest vs NORMAL (and vs stress without inversion) shows positive expectancy under the live rule (10:00 AM EST entry, 1 ATR stop / 2 ATR target, 60-minute time stop), with costs, day clustering, OOS, and multiple-testing caution.
• Why (live case): Fri 2026-11-06 8:58 AM ET run — stress, VIX 26.92, VIX9D/VIX 1.213, risk 0.125%; MR 8 fade-longs HON/TMO/BLK/AMGN/ZS/COIN/BAC/MSTR. Early regular-session prints ~10:10–10:31 mixed (HON/TMO soft green; ZS flipped soft; COIN/MSTR red). Scorecard liveRules all mean R −0.052, TARGET 0/196, enoughDays false.
• Distinct from: S-006 (earnings big-gap fade ban); S-012 (generic residual-reversion replay); S-141 (SM crush policy).
• Whiggity Auditor handoffs to test (not findings): condition fade-long sample on VIX9D/VIX>1; residualZ≤−3 under STRESS vs NORMAL 60-min stats; check look-ahead in regime flag availability at 10:00.
• Evidence: 2026-11-06 10:00 report §5, §7–8.

Evidence added 2026-11-06 10:00 EST — VST webcast / CEG peer / MR fades / AMC leftovers

• S-006: live Fri unrefilled earnings gaps — NET ~+11–12% day-1 after beat/raise; TEAM ~−15–17% after outlook scare; MCHP still deeply dislocated after soft sequential guide; APP day-2 still elevated after Thu +31%. Do not same-day MR-fade. Report §6–7.
• S-004 / S-132 / S-141 / S-143: VST webcast-hour ~+7% with CEG sympathy ~+4% (ρ 0.845) into Mon 11-09 BMO; SM remains suspended. Friday sympathy ≠ Monday open print (Auditor open question). Report §3–4, §7–8.
• S-014: same AMC night split — edge/security beat-raise (NET) vs collab/DC outlook scare (TEAM); opposite day-1 signs. Do not treat TRAI "Software" as one factor. Report §6–7.
• S-005 / S-137: APP day-2 ~+5% with LM empty — continuation is not a live LM signal; no drift premium. Report §6.
• S-144 (new): STRESS + inversion MR fade-long regime split — NEEDS DATA; mixed early Fri prints + negative liveRules. Whiggity tool budget exhausted after topic 1; Auditor/yoo handoffs logged as open questions only. Report §5, §7–8.
• yoo handoffs (open): peer-reports-within-1-session flag for CEG; APP day-2 in MR universe? — NEEDS DATA, not truths.
• Source: 2026-11-06 10:00 report. No live-engine/app changes.

S-145: Pre-weekend sizing for Monday AMC mega-|gap| names (TTD class) · PROPOSAL / NEEDS DATA (extends S-006 / S-038 / S-138; calendar-distinct from same-day S-138)

• Rule (proposal): when Friday is the last regular session before a Monday AMC print whose TRAI last-4 avg |gap| is ≥ ~10% or ≥ ~2 ATR (live case: TTD avg |gap| 13.56% / 2.58 ATR into Mon 2026-11-09), tag Friday afternoon / weekend inventory with overnight gap-risk sized to max(hist avg |gap|, event-timed options-implied if available under S-045). Do not treat Friday soft path as a validated fade into the print (S-127 stays rejected). After the Mon reaction gap prints, apply S-006 if |gap| ≥ 2 ATR.
• Why (live case): Fri 2026-11-06 mid-session TTD ~−1.4% into the largest near-term |gap| in /api/events upcoming; weekend news risk sits between Fri cash and Mon AMC wire — S-138’s same-day AMC pre-print modifier does not cover that calendar gap.
• Also on the same Mon night (context, not automatic mega-gap class): HIMS avg |gap| 6.85%, OXY 4.61% (1.50 ATR), MARA 6.13% with Thu −14.4% pre-print (S-138/S-142 size modifiers); KKR/CEG are BMO Mon (sessionsToReaction 1).
• Test: Friday→Monday AMC reporters with hist avg |gap| ≥ 10% or ≥ 2 ATR (2y); compare weekend |open gap| overrun vs baseline and vs same-week non-mega AMC names; costs, day clustering, OOS, multiple-testing caution.
• Evidence: 2026-11-06 11:00 report §4–6.

Evidence adds — 2026-11-06 11:00 (Mon gap stack)

• S-006: TTD hist fill 25% with avg |gap| 13.56% / max 27.16% — reinforce no same-day MR-fade when Mon reaction ≥ 2 ATR. Report §4, §6.
• S-004 / S-122 / S-132: Mon crowded stack — TTD/HIMS/OXY/MARA AMC + KKR/CEG BMO; peer maps OXY→EOG/COP/DVN (ρ ≥ 0.89), KKR→BX (0.85), MARA→RIOT/MSTR/COIN (cluster 8). Report §4–5.
• S-045: do not invent impliedVsHist for TTD/HIMS/OXY/MARA this hour; keep event-timed expiry gate (prior KKR non-event-days case). Report §8.
• S-138 / S-142: MARA Thu −14.4% ≫ hist avg |gap| 6.1%; HIMS Thu −9.8% ≥ hist 6.9% — size modifiers into Mon AMC, not sign predictors; Fri HIMS bounce does not clear overnight risk. Report §4, §6.
• S-008 (calendar quality): BRK-B TRAI timing AMC on 2026-11-07 mismatches issuer Saturday ~7:00 a.m. CT Internet post (Berkshire PDF); not Fri AH risk; hist avg |gap| only 0.72%. Report §3.
• S-145 (new): pre-weekend sizing for Mon AMC mega-|gap| (TTD class) — PROPOSAL / NEEDS DATA. Whiggity consult filled (budget exhausted after 4 calls; TTD/KKR primaries + BRK calendar conflict confirmed; Auditor/yoo handoffs only). Report §7.
• Whiggity (11:00 fill): confirmed TTD AMC/call [V-P]; KKR call 10:00 + Tue 08:00 conference [V-P]; flagged TRAI BRK-B Saturday/AMC mapping (issuer Sat 7:00 a.m. CT — already §3); HIMS/OXY/MARA left UNV in consult but issuer-filled in report; implied-move UNV. Handoffs only — no new edge.
• Source: 2026-11-06 11:00 report. No live-engine/app changes.

S-146: Split TRAI "Industrials" SM/MR risk into rails / farm / AI-power electrical · PROPOSAL (extends S-001 / S-014)

• Rule (proposal): when clusterFlag or sector label is Industrials, do not count UNP (cluster 18), DE (singleton), HON (cluster 22), and GEV/ETN/CAT (cluster 1) as one correlated risk unit. Cap exposure using 60d correlation clusters; optional hand map: rails (UNP/UPS), farm (DE), automation (HON/GE), AI-power electrical (GEV/ETN/CAT/VRT).
• Why (live case): Fri 2026-11-06 ~12:10 ET — ETN ~+3.9%, VRT ~+3.3%, ANET ~+6.9%, CAT ~+1.2%, GEV ~+0.6% vs UNP ~−1.9% on STB merger friction, while SM residuals shared "Industrials ×5" on the 8:58 run.
• Test: on SM/MR signal days, compare same-day and next-open returns of sector-flagged industrials split by cluster id; measure book P&L when the engine allows one name from rails and one from cluster-1 electrical vs forcing a single industrials slot. Costs, day clustering, OOS, multiple-testing caution.
• Evidence: 2026-11-06 12:00 report §2–6, §8.

Evidence adds — 2026-11-06 12:00 (industrials split)

• S-001 / S-014: Industrials sector flag over-aggregates; midday path split rails vs cluster-1 electricals. Report §2–3.
• S-144: HON ~+1.6% midday after Thu residual crush, no company news; still after 10:00 window; liveRules unchanged negative. Also: midday VIX ~23.61 / VIX9D/VIX ~1.07 vs 8:58 snapshot VIX 26.92 / 1.213 — regime clock matters. Report §2, §4, §8.
• S-141: GEV HVDC press release while SM residual — do not reactivate SM. Report §6.
• S-146 (new): split Industrials risk by correlation cluster — PROPOSAL. Report §8.
• Source: 2026-11-06 12:00 report. No live-engine/app changes.

S-147: Split TRAI "Consumer Staples" SM/MR risk into tobacco / food-HPC / retail · PROPOSAL (extends S-001 / S-014 / S-146)

• Rule (proposal): when clusterFlag or sector label is Consumer Staples, do not count PM/MO (cluster 19), PG/MDLZ/KO/PEP/KHC (cluster 4 food/HPC core), and WMT/COST (cluster 13) as one correlated risk unit. Cap exposure using 60d correlation clusters; optional hand map: tobacco (PM/MO ± BTI watch), food/bev/HPC (cluster 4 staples subset), retail compounders (WMT/COST).
• Companion test (same family): after a staples residual crush (e.g. |residualZ| ≥ 3 on PM or sector-flag SM staples), when next-session SPY > 0 and VIX > 22, measure continuation vs fade of PM/MO vs XLP vs cluster-4 names over the live-rule 10:00→11:00 window (and open→close proxy). Do not treat afternoon paths as that test's outcome.
• Why (live case): Fri 2026-11-06 ~13:13 ET — PM ~−1.1%, MO ~−1.3%, BTI ~−3.1%, XLP ~−0.7% while SPY ~+0.5% after Thu PM residualZ −5 / −4.63%; SM tag still grouped PM with MDLZ/PG/WMT.
• Test: on SM/MR signal days, compare same-day and next-open returns of sector-flagged staples split by cluster id; plus the SPY-up / VIX>22 continuation-vs-fade slice. Costs, day clustering, OOS, multiple-testing caution.
• Evidence: 2026-11-06 13:00 report §2–6, §8.

Evidence adds — 2026-11-06 13:00 (staples PM residual crush)

• S-001 / S-014: Consumer Staples sector flag over-aggregates; midday path splits cluster-19 tobacco vs cluster-4 food/HPC vs cluster-13 retail. Report §2–3.
• S-144: PM midday ~−1.1% continuation after Thu residZ −5 (no verified US earnings reprint) under SPY green + VIX>22 — contrasts noon HON bounce; still after 10:00 window; liveRules unchanged negative. Midday VIX ~25.7 / VIX9D/VIX ~1.10 vs 8:58 26.92 / 1.213. Report §2, §4, §8.
• S-141 / SM policy: do not reactivate SM on staples residuals or Japan local IR (bonds by IQOS expansion). Report §4.
• S-147 (new): split Consumer Staples risk by correlation cluster + SPY-up/VIX>22 continuation-vs-fade companion test — PROPOSAL. Report §8.
• Source: 2026-11-06 13:00 report. No live-engine/app changes. Whiggity partial UI only (2 searches, no final answer); Vectura $198M headline rejected as 2024 closed deal, not Nov 2026 news.

S-148: Split cluster-1 "memory/storage" risk into HBM/DRAM silicon vs HDD twin (WDC–STX) · PROPOSAL (extends S-001 / S-014 / S-146 / S-147)

• Rule (proposal): when sizing SM/MR or cluster caps involving MU, WDC, STX (SNDK external watch), do not treat them as one residual book. Cap WDC+STX as a ρ≥0.9 twin (0.95 asOf 2026-11-05); treat MU with other HBM/DRAM / semis factor names separately even though all currently sit in cluster 1.
• Companion tests: (1) after a ≥15% same-day crush in MU and WDC/STX, measure next-session MU−WDC / MU−STX return spreads under VIX>22; (2) on TSMC-or-peer DRAM rumor days, test whether WDC/STX co-move with MU or decouple (hypothesis: decouple). Costs, day clustering, OOS, multiple-testing caution.
• Why (live case): Fri 2026-11-06 ~14:05 ET — MU ~+5.6% while WDC/STX ~−2.2% after Thu ~−17% each; TSMC 3D-DRAM rumor is MU/TSM narrative, mis-applied to HDD in some secondary headlines.
• Evidence: 2026-11-06 14:00 report §2–5, §8.

Evidence adds — 2026-11-06 14:00 (memory/storage MU bounce vs WDC–STX fail)

• S-001 / S-014: cluster-1 now bundles MU with WDC/STX; product split still required inside the cluster (HDD twin ρ 0.95 vs HBM/DRAM). Report §3.
• S-004 / S-132: AMAT/CSCO 11-12 AMC remain cluster-1 peer gap-risk; do not size from Fri afternoon bounce. Report §6.
• S-005 / S-006 / S-141: no day-2 LM claim on MU; no SM reactivation on failed WDC/STX rebound. Report §8.
• S-144: midday VIX ~24.5 / VIX9D/VIX ~1.08 (still stress/inverted vs 8:58 26.92 / 1.213); MU/WDC/STX were not on the 8:58 MR list. Report §2.
• S-148 (new): split HBM/DRAM vs HDD twin inside cluster-1 — PROPOSAL. Report §8.
• Source: 2026-11-06 14:00 report. No live-engine/app changes. Whiggity folded late: budget after 4 rounds, no [V-P]; multi-catalyst framing + F5 cost-spillover footnote kept; F10 rejected; closes confirmed via Yahoo. Report §7.

S-149: Split TRAI "Healthcare" SM/MR risk into cluster-5 pharma vs REGN singleton vs TMO tools vs MRNA/MRK print cluster · PROPOSAL (extends S-001 / S-014 / S-146 / S-147 / S-148)

• Rule (proposal): when clusterFlag or sector label is Healthcare, do not count cluster-5 large pharma (PFE/LLY/GILD/VRTX/AMGN/ABBV/…), REGN (cluster 38 singleton), TMO (cluster 48 tools singleton), and MRNA/MRK (cluster 20) as one correlated residual book. Cap exposure using 60d correlation clusters; optional hand map: large-pharma core (cluster 5), Regeneron singleton, life-science tools (TMO), mRNA/print pair (MRNA–MRK), fintech-adjacent telehealth (HIMS cluster 16 — separate calendar risk).
• Companion tests: (1) after a same-day healthcare SM crush under VIX>22, measure next-session cluster-5 mean vs REGN vs TMO vs MRNA return spreads; (2) on MRNA (or peer) BMO overrun days (|gap| ≥ 2× hist avg), test whether cluster-5 residuals co-move with MRNA or decouple (hypothesis: decouple); (3) look-ahead hygiene on overrun tags — EDGAR/wire acceptance ≤ live-rule clock (today 06:47 EST vs 08:58 premarket). Costs, day clustering, OOS, multiple-testing caution.
• Why (live case): Fri 2026-11-06 ~15:09 ET — MRNA ~+10%, TMO ~+2.3%, cluster-5 roughly flat (PFE still soft), REGN soft green after Thu crush; SM still tags Healthcare ×5 including REGN singleton.
• Evidence: 2026-11-06 15:00 report §2–6, §8.

Evidence adds — 2026-11-06 15:00 (healthcare split MRNA BMO vs cluster-5/TMO)

• S-001 / S-014: Healthcare sector / Healthcare ×5 flag over-aggregates; afternoon path splits cluster-5 vs TMO tools vs MRNA print vs REGN singleton. Report §3–6.
• S-006 / S-143: MRNA ~+10% positive overrun after BMO beat/raise — do not same-day fade; do not treat as peer inheritance into suspended SM residuals. Report §4–5.
• S-141 / SM policy: do not reactivate short momentum on Healthcare ×5 after Fri flat/bounce paths. Report §8.
• S-144: TMO/AMGN remain on MR fade-long list under STRESS + inversion (afternoon VIX9D/VIX ~1.06); TMO bounce descriptive only after 10:00 — NEEDS DATA unchanged. Report §2, §4.
• S-145: HIMS Mon AMC unchanged — not re-litigated; Fri HIMS bounce ≠ cleared overnight risk. Report §1.
• S-005: no day-1 LM claim on MRNA overrun; LM bucket empty.
• MRNA fold (whiggity): YE cash raise includes convert proceeds (not pure ops); GSK patent cross-license "no payments" (2026-10-28) cited from Newswire [V-P]; 06:47 EST acceptance vs BMO label — look-ahead open for auditor. Report §4, §7.
• S-149 (new): split Healthcare risk by correlation cluster — PROPOSAL. Report §8.
• Source: 2026-11-06 15:00 report. No live-engine/app changes. Whiggity folded late: budget after 4 rounds, MRNA leg only; F1–F10 kept after Newswire/IR verify (incl. GSK no-pay settle + 06:47 acceptance nuance); F12–F14 TRAI/Yahoo independently verified; non-MRNA [UNV]; Actionable Fri NO. Report §7.

S-150: Split TRAI "Financials" SM/MR risk into cluster-2 banks/managers vs V–MA payments twin vs cluster-8 crypto-beta · PROPOSAL (extends S-001 / S-014 / S-146 / S-147 / S-148 / S-149)

• Rule (proposal): when clusterFlag or sector label is Financials, do not count cluster-2 banks/brokers/managers (JPM/BAC/C/GS/MS/BLK/KKR/BX/…), V–MA payments twin (ρ≥~0.9; currently cluster 3 with media/travel), and cluster-8 crypto-beta (COIN/MSTR/HOOD/MARA/RIOT) as one correlated residual book. Cap exposure using 60d correlation clusters; optional hand map: money-center/broker core (cluster 2), payments twin (V–MA ρ gate), crypto-beta (cluster 8), BRK conglomerate (cluster 13 — separate Sat filing calendar).
• Companion tests: (1) after a STRESS-day financials SM/MR crush under VIX>22, measure next-session cluster-2 mean vs V/MA vs COIN/MSTR/HOOD return spreads; (2) payments ρ gate — require V–MA co-move (or MA confirmation) before sizing a V residual as "financials"; same for GS–MS inside brokers; (3) COIN residual vs crypto β — hit rate of COIN MR fade-long when residualZ>1 but BTC 1d <0 / ETF outflow day (hypothesis: residual alone insufficient); (4) Mon contamination from BRK Sat + KKR BMO on cluster-2 alt-manager names (look-ahead / event contamination — Auditor). Costs, day clustering, OOS, multiple-testing caution.
• Why (live case): Fri 2026-11-06 close — BLK ~+0.6% vs BAC −0.7% / COIN −1.5%; V flat vs MA −0.5%; HOOD +3.2% vs COIN/MSTR soft; SM still tags Financials ×3 as JPM+V+C across two correlation homes.
• Evidence: 2026-11-06 16:00 report §2–6, §8.

Evidence adds — 2026-11-06 16:00 (financials STRESS BLK/BAC/COIN vs JPM/V/C)

• S-001 / S-014: Financials sector / Financials ×3 flag over-aggregates; close path splits cluster-2 vs payments twin vs crypto cluster-8. Report §3–6.
• S-144: BLK/BAC/COIN remain on MR fade-long list under STRESS + inversion (close VIX9D/VIX ~1.08); paths mixed (BLK bounce, BAC/COIN soft) — descriptive only after 10:00 — NEEDS DATA unchanged. Report §2, §4.
• S-141 / SM policy: do not reactivate short momentum on Financials ×3 after Fri mixed paths. Report §8.
• S-145: KKR Mon BMO + BRK Sat ~7am CT filing unchanged — not re-litigated; Mon gap stack still governs weekend prep. Report §4.
• S-012: COIN residualZ +1.68 with crypto home + Fri BTC soft — caution that residual ≠ BTC-orthogonal (feeds S-150 companion 3). Report §4, §7.
• S-005: no LM claim; LM bucket empty.
• S-150 (new): split Financials risk by correlation cluster + payments/GS–MS ρ gates + COIN residual vs BTC β — PROPOSAL. Report §8.
• Source: 2026-11-06 16:00 report. No live-engine/app changes. Whiggity folded: budget after 4 rounds, KKR/BRK calendar leg primarily; F1–F3/F5–F6 kept after BusinessWire/PDF verify (BRK TZ resolved to Central); F10 rejected; BLK/BAC/JPM/V/C issuer [UNV] by whiggity (we covered BAC Boston + secondaries independently); Actionable Fri NO. Report §7.

S-151: Split TRAI "Software" SM/MR/LM risk into cluster-11 collab vs cluster-7 security/edge vs cluster-9 productivity vs APP singleton · PROPOSAL (extends S-001 / S-014 / S-146 / S-147 / S-148 / S-149 / S-150)

• Rule (proposal): when clusterFlag or sector label is Software, do not count cluster 11 (TEAM/CRM/NOW), cluster 7 (NET/DDOG/ZS/CRWD/PANW/PLTR/SNOW), cluster 9 (INTU/WDAY/ADBE/TTD), and APP (cluster 27 singleton) as one correlated residual book. Cap exposure using 60d correlation clusters; optional hand map: collab/seat SaaS (11), security/observability/edge (7), productivity/adtech (9), adtech-AI singleton (APP). Keep ORCL with semis cluster-1 (exclude from Software risk unit).
• Companion tests: (1) after a Software SM crush under VIX>22, measure next-session cluster-11 mean vs cluster-7 vs cluster-9 vs APP return spreads; (2) on same-night dual AMC (TEAM+NET class), test day-1 c/c sign agreement (hypothesis: disagree when guide quality differs); (3) travel/gaming adjacency — do not inherit ABNB LM or DKNG crush into Software caps; test ABNB–BKNG same-day break on ABNB print days. Costs, day clustering, OOS, multiple-testing caution.
• Why (live case): Fri 2026-11-06 post-close — TEAM −16.15% SM vs APP LM +9.86% vs NET +9.91% (rangePos 0.18) vs CRM/INTU/WDAY soft −3–4%; ABNB LM +11.67% vs DKNG −21.30% vs BKNG −4.77%.
• Evidence: 2026-11-06 17:00 report §2–5, §8.

Evidence adds — 2026-11-06 17:00 (software/SaaS bifurcation TEAM vs APP/NET; ABNB vs DKNG)

• S-001 / S-014: Software sector flag over-aggregates; postclose path splits cluster-11 vs cluster-7 vs cluster-9 vs APP singleton. Report §2–3.
• S-005 / S-066 / S-137: APP day-2 LM + NET day-1 gap-fade (rangePos 0.18) — descriptive only; no Monday LM seed / no chase. Report §4.
• S-006: TEAM −16% and DKNG −21% mega earnings gaps — do not same-day MR-fade. Report §4.
• S-141 / SM policy: sole SM name = TEAM; do not reactivate short momentum after outlook-scare crush. Report §4–5.
• S-144: postclose still STRESS (VIX 25.64, VIX9D/VIX 1.060); MR bucket empty — no fade-long claim. Report §2.
• S-145: Mon gap stack unchanged — not re-litigated. Report §1.
• S-118: ABNB print day vs BKNG −4.8% — live twin-break case. Report §4.
• S-151 (new): split Software risk by correlation cluster + dual-AMC sign-disagreement + travel/gaming adjacency — PROPOSAL. Report §5, §8.
• Source: 2026-11-06 17:00 report. No live-engine/app changes. Whiggity skipped (no computerUse in executor). Actionable Fri NO.