Jul 30, 2026

Regime & Verdict#

• Regime: CHOP · ret5 +0.480% · ret10 −1.200% · dd15 −2.830% · directional_tradable=TRUE · s1_standdown=FALSE
• Vol-state: VIX 17.09, −17.3% on the day — the 07-29 spike to 20.66 fully round-tripped in a single session. SPY GEX −$366.9M FULLY_NEGATIVE; QQQ NEGATIVE (spot 683.77 vs zero-gamma 729.4). Dealer-short-gamma on both = vol-expansion conditioning. VRP FAIR both indices (SPY +0.0126, QQQ −0.0057) — nothing rich to sell. Front-end backwardation SPY 1.296× / QQQ 1.445×.
• Breadth: DIVERGENT. SPY closed +1.68% on only 39.17% green (197 adv / 304 dec); UW bullish_pct 35.7%, trend CHOPPY/TRANSITIONAL. Today's bounce was narrow and mega-cap/Tech-led, not breadth-confirmed. QQQ +3.30% on the day but still −1.22% ret5 / −3.17% ret10 — the AI-capex repricing leg has not round-tripped.
• Bottom line: no new directional edge tonight — zero new starters. Two names cleared every mechanical lane gate and both were VETOED at gate 4 on fundamentals. One held short (PLNT) is carried at unchanged half size with 2 sessions to a binding exit. Vol book has two clean earnings-crush candidates; the 0DTE sleeve stands down.

Directional Book (excess-scored)#

No new starters. Everything else below is watch or advisory.

Why the two clean signals were killed#

Both cleared liquidity, ETP, earnings-window, catalyst-split and persistence — then failed the fundamentals veto channel, which exists precisely to separate crowding from corroborated conviction.

• AVTR — VETO. Cleanest OI signal of the night (oi_net_5d +101,124, rel_build 2.713, persistence 0.47 = genuinely organic). The 2026-W29 objection against it (single-day block, persistence ~1.0) is resolved. But: Q2 beat +8.58% with FY guidance raised, and five same-day PT raises on 07-30 (WFC $14→16, Stifel $8→15, Evercore $10.5→14, MS $10→14, RBC $9→14), zero cuts, insider MSPR +33.33. Price +22.12% ret5, RSI 75.68. This is the ROST/ALLE pattern exactly.
• HRI — VETO carried and reinforced from 07-29. The recorded re-admit condition was specific: the analyst tape must turn. It did the opposite — KeyBanc raised $165→$185 and BNP raised $160→$165 on 07-29, zero cuts. Condition NOT MET. Separately, ret5 −10.62% means an entry would short a completed collapse, and the lane's edge is measured on the general population of heavy-call-build names, not on names already down double digits.

Recorded for fairness: HRI's PE ~99× and D/E 4.18 are genuinely stretched, and the −14.6% unwind off the 07-28 high reads as a valuation reset — a real reason the short could still work. It is not enough to overturn a gate whose bar is "do the fundamentals contradict the thesis."

Vol Book (non-directional, delta-neutral, 0 directional points)#

Earnings IV-crush — SELL VOL. Note the screen inverted the usual assumption: most of the IV-rank ≥ 80 list is IV-cheap vs its own realized post-print history (TWLO 2.7% implied vs 13.5% realized; AKAM 2.8% vs 14.9%; AMD 3.5% vs 9.1%; PLTR 2.2% vs 12.4%) — IV rank is a level, not an edge. Four names are actually rich:

All defined-risk, wings set at the realized-max print, never at the implied move. ATM spreads run ~16–24% on these monthlies; no measured cost constant exists for this sleeve, so net_expectancy is left null rather than fabricated.

0DTE-VRP — STAND DOWN, both indices. The script returned GO_PREMIUM_SELL_INTRADAY with positive net math (SPY gross +0.246% → net +0.146%/day; QQQ +0.376% → net +0.276%/day) and was overridden: front-end backwardation on both (1.296× / 1.445×) + dealer-short-gamma on both + a fresh ~21% one-day VIX round-trip is the unsampled-left-tail combination the tail cap exists for. The 60-day sample behind those averages contains no analog.

Maintenance item (not patched tonight): scripts/zerodte_setup.py misclassified QQQ as long-gamma — its recommend() path sums net gamma over 0–45DTE instead of using the spot-vs-zero-gamma-level convention the CLI gex command applies, so it picked the quiet/mean-reverting branch. Logged, not fixed: a code edit tonight would trip the regression gate mid-scan.

Watch / Stood-down#

• OI_FADE mechanism failures — OLLI, INFY. Both fail the lane's own multi-day-build test: persistence 0.974 (OLLI: +6,672 of a +6,850 build landed in ONE session, today) and 0.993 (INFY: a ~108K-contract single-session block). A same-day mega-block on a dealer-short-gamma bounce day is at least as consistent with systematic hedging as with crowd positioning. Re-check 07-31 — if the build persists a second session, they re-enter. No fundamentals gate was run on either, correctly: they fail upstream of gate 4, and this repo gates only names about to be sized.
• MOM_SHORT — watch-only (SNPS, WSO, LHX, VICI). Capped by CLAUDE.md invariant #6: the lane's −0.03% mean is a known-negative recorded exception. WSO is the archetype of the lane's risk (RSI 22.2, ret2 −15.45%). BLDR/QXO/PEG earnings-blocked.
• MOM_LONG — 108-name near-high basket, watch-only BY THE RUBRIC (+0.18% mean is below the +0.3% MEDIUM threshold; median −1.06%, tail-driven). Cohort data has been supportive three sessions running, but the validated prior is what sizes.
• S2 (advisory, +0.06%) — 40 survivors; top PLD, FCX, DHI, CFG. The lane's own read: one-sided concentration is a binary event gate with IC ≈ 0, not a ranking signal — do not read the biggest print as the best candidate. Five vetoed on same-day/recent earnings (ITW, PNC, CBRE, GFL, ACGL).
• S4 (advisory, +0.44%) — 37 candidates, 86% assessed LIVE PCR rather than stale 07-29 fear. Top: QTWO (147.0), RGEN (82.2), JEF (81.3). Caution: 11 names show ivrank_chg_5d < −25 — hedging premium already partly harvested.

Risk#

• Correlation clusters: none. Max pairwise 0.15 across the measurable set — nothing near the 0.70 collapse threshold. PLNT is the sole open position.
• Event calendar (10 td): PLTR 08-03 · AMD 08-04 · NFP 08-07 (Tier-1) · SMCI 08-11 · CPI 08-12 (Tier-1). No FOMC. PLNT's binding 08-03 exit lands clear of its own 08-06 print (3 trading days) and ahead of both Tier-1 macro events.
• Tail caps applied: MOM_SHORT watch-only (invariant #6); MOM_LONG basket-only; S2/S4 advisory-only; 0DTE stood down; UCTT excluded on a breached left tail.
• CHOP half-cap ceiling: non-binding — nothing cleared above half in the first place. PLNT was already at half from entry.
• PLNT stop is NOT vestigial. Tested rather than asserted: over 133 two-day windows only 2 (~1.5%) cleared the +6.29% needed to reach $59.50. Low-probability but not arithmetically unreachable — the vestigial precedent describes a winning position whose stop has been outrun, which is the opposite geometry. Kept live and unchanged; not tightened further, since with 2 sessions left a tighter stop adds whipsaw without shortening a resolution the h10 exit already caps.

Two process findings#

1. Gate-1 dissent — raised by risk-sizer, DECLINED by the orchestrator. It argued the tape shape is a textbook V-rebound and asked to treat s1_standdown as FIRED for MOM_SHORT despite the mechanical FALSE. Declined: regime_check.py is the authoritative gate the retro harness scores against, and hand-flipping the recorded flag would corrupt the exact field /calibration-audit uses to grade whether the crash gate works. The caution is real and recorded — Phase A raised it independently — but it changes nothing operationally, since MOM_SHORT is already unsized under invariant #6. If the shape recurs, codify a proper up-thrust detector via /calibration-audit rather than re-litigating nightly.
2. MOM_LONG lane bug, second consecutive night. The lane again returned ROST inside its own basket; ROST carries a standing fundamentals VETO/DROP from 2026-07-20. It still does not run prior_verdicts.py against its own output. Excluded by the orchestrator both nights. Flagged 07-29, unfixed — worth a maintenance ticket.

Data note#

Preflight initially failed (exit 1): the truth-set parquets ended 07-29, one session behind the trade date. Rebuilt prices / returns / features through 2026-07-30 before any lane ran; re-run went clean (exit 0). This is the failure that produced an inverted regime read on 2026-07-24 — no lane read a stale panel tonight.