Jul 31, 2026

Regime & Verdict#

• Regime: CHOP · ret5 +1.100% · ret10 +0.500% · dd15 −1.860% (scripts/regime_check.py, authoritative)
• Vol-state: VIX 15.99 (−6.4% d/d; −22.6% off the 07-29 spike of 20.66). Forward term structure CONTANGO on both indices (VIX3M 19.02 > VIX 15.99; 3–6DTE→30DTE curve rising on both). The 07-30 backwardation has cleared.
• Dealer gamma is split: SPY POSITIVE but pinned on the flip (spot 747.06 vs zero-gamma 746.93, total_gex +$1.19B) — long-gamma by a coin-flip's width. QQQ FULLY_NEGATIVE (short gamma at every sampled strike). Vol-state conditioning only; no direction read from this.
• Breadth contradicts the tape, second straight night: pct_green 43.94% (221 adv / 281 dec), UW bullish_pct 34.1%, Technology the largest sector outflow (−$187.7M) — while SPY grinds higher and sits above both the 20- and 50-SMA.
• directional_tradable = TRUE · s1_standdown = FALSE (mechanical, recorded as measured)
• Bottom line: No new directional starters. Third consecutive session at zero for OI_FADE. One held short (PLNT) carried to its binding exit Monday. Vol book active on three earnings-crush names + both 0DTE indices.

V-rebound caution (recorded, NOT applied as a flag flip)#

Phase A independently re-raised it: the week is structurally a V-shape (SPY −2.40% ret5 on 07-29 → +1.68% on 07-30 → +0.72% today). The s1_standdown guard mechanically cooled off one day after the shape that would have fired it — +1.10% ret5 sits below the +1.5%-off-a-dip trigger. Combined with the breadth divergence, this reads as a narrow, unconfirmed bounce sitting on top of a real V.

Handled exactly as the 07-30 precedent: the mechanical flag is recorded FALSE and not hand-flipped, because regime_check.py is the field /calibration-audit grades the crash gate against, and hand-editing it would corrupt that measurement. The caution is carried as sizing discretion instead. It is moot operationally tonight — nothing new was sized, and MOM_SHORT is already unsized under invariant #6.

---

Directional Book (excess-scored)#

No new starters. Zero names cleared the full stack.

PLNT — HOLD to Monday's close#

Entry 2026-07-20 @ $55.20 · mark $55.91 · gross −1.29% · since-entry SHORT excess −0.53%.

Computed directly against SPY. held_book.py prints gross (since-entry) beside excess (today-only, +0.85%) on adjacent lines — two different horizons. Do not read that column as since-entry.

• Signal is decaying but has not crossed. Pre-registered exit leg (b) is "first genuinely NEGATIVE call_net": today call_net = +20, still nominally positive. But puts opened +355, so net OI printed −335 — the first negative net-OI print of the hold.
• Not exiting early. Jumping an unconfirmed pre-registered trigger is a discretionary override of a mechanical rule. The h10 timer binds Monday regardless; if leg (b) confirms intraday Monday, both resolve on the same session.
• $59.50 stop is LIVE, not vestigial. held_book.py flagged it VESTIGIAL (6.42% away, beyond entry). Overridden on the same reasoning as 07-30: the vestigial precedent describes a position that has run >10% in its favor so its old stop can no longer bind. PLNT is adverse (gross −1.29%) — opposite geometry. With one session left it is near-moot either way.
• Event-clean: PLNT reports 08-06, three sessions after the 08-03 exit. NFP (08-07) and CPI (08-12) both land after exit.

---

Vol Book (non-directional, advisory, 0 directional points)#

Earnings IV-crush — SELL VOL, all three carried positions re-confirmed. Ranked by implied-move vs each name's own realized post-print history, not IV rank.

None tripped the UCTT exclusion condition (max-realized > implied). Zero new genuinely-rich names from ~27 screened — every one screened IV-cheap vs its own history.

0DTE-VRP. The 07-30 stand-down rested on a three-condition triad (backwardation + dealer-short-gamma + VIX spike). Only 1 of 3 is present tonight.

• SPY: GO, half size. Iron fly/short straddle ≈747, wings ≈ ±0.79% (≈741.4–752.5).
• QQQ: GO but reduced (≤0.5×) and widened. Wings ≈ ±2.01% (≈674.2–701.8). Dealer short-gamma confirmed by both the CLI and the script tonight — they agree, unlike 07-30.

net_expectancy NULL on all five entries. The gross figures (SPY +0.248%/day, QQQ +0.382%/day) come from a 60-day sample containing no vol-shock day, and no measured half-spread cost constant exists in this repo. Fabricating a net number would violate the net-of-cost discipline it pretends to satisfy.

Caveat on QQQ: its VRP is negative (−3.23%) — realized 25.88% ran above implied 22.65%. Options never repriced this week's whip. That is an argument against selling QQQ premium that is independent of the gamma sign, and it is why the size cut matters.

---

Watch / Stood-down#

OI_FADE — zero starters, third straight session.

• AVTR — cleanest mechanical profile in the panel (rel_build 2.657, persistence 0.416 organic, catalyst LIVE, earnings 11-04 clear). VETO SUSTAINED. Re-admit condition re-tested tonight; neither half met. A 07-31-datelined PT-raise item was a lagged republication of the same 07-30 actions, not new coverage; FY2026 guidance is still the freshly-raised figure.
• HRI — VETO SUSTAINED, both halves unmet. Additionally worsening: persistence 0.847 sits on the 0.85 fail line, and it printed +2.89% today, so the "shorting a completed collapse into a bounce tail" objection is stronger, not weaker.
• INFY — WATCH, mechanism fail unchanged. Tonight's +3,568/+2,812 add is noise on a block: the 07-30 print alone is 98.4% of the 5-day net, persistence still 0.984.
• OLLI — DROPPED. call_net went genuinely negative (−7) — calls closing, a real lane invalidation, not just the persistence complaint.
• Cut: QGEN/SOLV/DBX/TPG (earnings inside h10) · KBR (RESOLVED_PRE_PRINT — build died at its own 07-30 catalyst) · MTUM (ETF) · APAM ($37.7M ADV) · SHAZ (listing age 41 days post reverse-merger — 30d OI base not meaningful).

MOM_SHORT — watch-only capped (invariant #6). Clean after the corrected earnings gate: CRH, LII (+ CLBK, PATK, both fail-closed on the parquet gap). Earnings-blocked: ECHO (08-03), PEG (08-04), FIGR (08-13), QXO (08-13).

MOM_LONG — basket, watch-only. +0.18% mean is below the +0.3% threshold; median −1.06%, tail-driven. 191 names, of which only 106 pass the h10 earnings gate.

S2 liquidity-reversion (advisory, +0.06%) — 58 names. Top: WFC, VZ, ITW, CMS, TD, CRH, RY, TRP, BNS, TSCO. CRH appears in both S2 and MOM_SHORT — recorded as evidence-type diversification, never summed (additive confluence realized −8.2%, 0-for-6).

S4 sentiment-contrarian (advisory) — JEF, DKS, DPZ (the only names both correlation-checkable and not premium-harvested) · VMC, EQIX, SWKS (premium already harvested, ivrank_chg_5d −30.2/−39.7/−43.6) · PUMP, TENB, VIK, ESI, LBRT, HUBB, CSX, EIX, AGNC, GFI, KMX, NSC, GDDY (fail-closed, absent from prices.parquet).

QTWO self-excluded cleanly: PCR collapsed 147.00 → 0.30 post-print.

---

Risk#

Correlation clusters. Only pairwise corr ≥ 0.70 found: VMC ↔ CRH. risk-sizer reported 0.77; my independent recomputation on 62 days of daily returns gives 0.871. The window/method differs and I did not reconcile them — but both sit above the 0.70 threshold, so the verdict (collapse to one position if both were ever sized) is unchanged either way. Moot tonight; neither is sized. Recorded because the two figures should agree and don't.

Event calendar. ISM Mfg 08-03 (Mon) · NFP 08-07 (Tier-1) · CPI 08-12 (Tier-1) · PPI 08-13. A fresh h10 window opened tonight runs 08-01→08-15 and therefore captures both NFP and CPI; an h5 window captures NFP on its boundary. This is a live argument against opening anything tonight, and it applies on top of the lane verdicts. PLNT's exit lands clear of both.

Tail caps. MOM_SHORT watch-only cap holds. Vol-short sized for the unsampled left tail with net_expectancy NULL rather than fabricated.

Hedge note. No net directional exposure beyond one half-size short resolving Monday. No hedge required.

---

Process findings#

1. data/prices.parquet coverage gap is now blocking real decisions — fix this first. The parquet carries 783 tickers vs features.parquet's 2,460, so the correlation-cluster gate is only computable on the narrower panel. Tonight 13 of 19 S4 names (including top-ranked PUMP) and 2 of 4 clean MOM_SHORT names were uncomputable. Treated fail-closed by both the orchestrator and risk-sizer independently. This is the fifth consecutive session this class has surfaced (07-28 HOG · 07-29 HRI/PLNT · 07-30 LHX/VICI · tonight the S4 majority) and the first on which it touched a lane that was a live sizing candidate — it is no longer moot. The gate does not error when a ticker is missing; it silently returns nothing.

2. The momentum lane's earnings gate is broken — it reported "0 dropped" and that is false. Independent re-run: 4 of 8 MOM_SHORT names (ECHO, PEG, FIGR, QXO) and 85 of 191 MOM_LONG names report inside the h10 window. PEG and QXO were already earnings-blocked on 07-30, so the lane regressed on names it had previously excluded correctly. Same class as the ROST prior_verdicts bug — the lane runs the script and then misreports the result. Nothing live depended on it (both legs watch-only), but this is a measuring-instrument failure and the third consecutive night of a momentum-lane hygiene defect. The prior_verdicts half is genuinely fixed — 37 blocked names dropped, ROST correctly absent, zero ETP leakage.

3. front-end-iv-ratio needs a near_dte_actual check every time it is read. Tonight it reported backwardation on both indices (SPY 1.61, QQQ 1.70) with near_dte_actual=0 — that is today's expiring 0DTE contract against the 30-day, a same-day-expiry inflation artifact, not a forward signal. The genuine forward curve is in contango. This casts doubt on the 07-30 0DTE stand-down, which cited "front-end backwardation on both (1.296× / 1.445×)" as one of its three conditions. I cannot retroactively verify that night's near_dte_actual, so I am not claiming the 07-30 call was wrong — but the metric it rested on has a known artifact mode that was not checked. Worth resolving in /calibration-audit.

4. fundamentals-gate deliberately NOT spawned on JEF/DKS/DPZ, against risk-sizer's action item. Phase D rule 4 gates only names about to be sized; S4 resolved to advisory/watch, so gating them would contradict the same discipline that correctly left OLLI ungated on 07-30. Recorded as an explicit adjudication rather than an omission. If S4 is ever elevated to starter, these three need the gate first.

5. S4 tier-vs-confidence conflict resolved on the sizing map, not by re-litigating the rubric. Phase C names +0.3–1% "MEDIUM", which reads as sizeable; the envelope's actual sizing map reads +0.3–1% → starter with half reserved for ≥+1% sign-stable, and separately advisory lane → watch. So half-cap was never reachable for S4 at +0.44% on magnitude grounds, independent of the LOW-MOD confidence tag. The Phase C tier names and the sizing map use overlapping language for different things — worth reconciling in the spec so this is not re-derived nightly.

6. scripts/zerodte_setup.py QQQ gamma-sign bug did not bite tonight — script and CLI agreed on QQQ's negative sign. The bug (summing 0–45DTE net gamma instead of spot-vs-zero-gamma-level) is still unfixed and still logged from 07-30; it simply did not surface. Do not treat this as evidence it is resolved.