Aug 10, 2026

Regime & Verdict#

• Regime: UPTREND · ret5 +2.030% · ret10 +4.590% · dd15 −1.300% (regime_check.py, arbitration exit 0 — Phase A's claim reproduced against the panel)
• vol-state: VIX 15.46 (LOW tercile), normal contango VIX9D 12.77 < VIX 15.46 < VIX3M 18.98. SPY dealer gamma NEGATIVE, spot 772.87 pinned at the 773.02 zero-gamma flip — whipsaw risk at this level, not suppression. (Aggregate GEX nets +$1.84B; the regime call is spot-vs-flip. The two stay separate.)
• breadth: pct_green 45.7% (230 adv / 270 decl), options-flow bullish 34.9%. Divergence flagged — a firm UPTREND tape with sub-50% breadth and net-bearish flow. Rotation OUT of Technology (−$279.9M), INTO Financials / Healthcare / Energy. Advisory, not a veto.
• directional_tradable = True · s1_standdown = False The crash guard was live all five sessions 08-03→08-07 (thrust peaked ret5 +5.53% on 08-05) and expired only today as ret5 decayed +5.53% → +3.51% → +2.03%. The guard lifting is not a green light — the deceleration is recent and shallow.
• Event risk inside the h10 window: CPI 2026-08-12, PPI 2026-08-13 — both Tier-1, both landing on days 2–3 of any window opened this week. FOMC 09-16/17 is outside.

Bottom line: no directional edge sized tonight — a NINTH consecutive zero-starter session, and the book stays flat for a sixth. One name (COO) survived the entire mechanical stack and was cut to WATCH on tier math. Two others died on hygiene, one of them via a boundary defect this scan found and fixed the read on.

Directional Book (excess-scored)#

Tier math, stated plainly: OI_FADE's measured excess is +0.55% mean, which sits in the MEDIUM band (+0.3–1%), not HIGH (≥ +1%). MEDIUM implies a starter. The fundamentals CAUTION takes one tier off → WATCH. The event-risk CAUTION would have imposed the same −1 independently. COO was never reachable for a half-cap HIGH size; it did not clear the +1% bar. Invariant #4 was not the binding constraint — tier math was.

COO's mechanical case was genuinely clean, which is why it got this far: 5d net call-OI build +4,619 (call +4,615 / put −4), rel_build 1.029, persistence 0.549, and a build that is really multi-day rather than one block — dailies +148 / +2,534 / +1,002 / +13 / +18 / +1,052 across 08-03→08-10, re-derived directly rather than taken from the lane. Next print 2026-09-09, clear of the whole window.

The countervailing read is real: 4/4 EPS beat streak and insider MSPR +40.33 (net buying) both contradict a short, while −42.6% YoY EPS growth on a ~63× trailing P/E and UBS initiating Neutral at a $75 PT below the $76.32 spot both confirm it. Short float 4.08% / 2.82 days-to-cover — no squeeze fighting an entry. A split verdict, correctly resolved as CAUTION rather than VETO. Healthcare being tonight's rotation-IN sector is a further headwind for a Healthcare short (carried in regime_fit 0.35, not as a gate).

Killed upstream — recorded so they are not re-proposed#

• EWTX — CUT. catalyst_split.py reported "LIVE 83.0%", and that number is an artifact. See Data Quality below. Genuine post-print continuation is +387 of +8,490 = 4.6% — the build died at its catalyst. This is the SSNC pattern the script exists to catch, and it very nearly passed. EWTX also carried a prior blocking verdict and a documented next_earnings_date staleness flag from the 08-06 scan.
• AVT — VETOED (fundamentals). The full triad fights the short: 4/4 accelerating beats (+2.26% → +9.15% → +11.01% → +27.50%), revenue +24.5% YoY, EPS +46.1% YoY, insider MSPR +52.37, Truist PT raised to $110 (+13% above spot) the same day as the beat, and the stock −2.6% off its 52-week high printing fresh highs rather than fading. The OI lane flagged this itself as the AVTR pattern — shorting a completed gap that held, with fundamentals confirming the new level — and held it out of sizing before the gate confirmed the kill.

Vol Book (non-directional)#

0DTE-VRP — STAND ASIDE, both legs, on economics rather than mechanics.

The pooled headline (mean_pnl_open_net_pct +0.26%) is a MID/HIGH-VIX artifact. Tonight's actual bucket is LOW (VIX 15.46, tercile cuts 16.4/17.9), and the conditional expectancy is negative net of cost on both names. Two independent reasons converge on each leg — gamma regime and conditional economics.

Earnings IV-crush — advisory only, never sized (no backtested crush harness in this repo; the case rests on structure alone):

Macro overlap caveat: CSCO and ENVX print on CPI day; CAVA and CAH have PPI inside their exit window. Single-name IV will not fully revert to a quiet post-print baseline while a Tier-1 macro print is still pending — that argues for wider wings and lower size, not tighter ones.

Watch / Stood-down#

• MOM_LONG — BASKET_WATCH, never sizes. 240-name near-52w-high basket; top sector Financial Services 70 names (29.2%), inside the 1/3 cap. Lane at −1.12% mean, n=1081 tonight. The full-universe refresh turned this lane outright negative, corroborating the forward BH(0.10)-negative read; 0 of 52 resolved forward calls have ever sized.
• MOM_SHORT — watch-only capped. 7 Yahoo-verified near-52w-low names: PEG, AMRZ, CMS, VICI, GME, AGCO, RBA (POST and GPI dropped on split/adjustment artifacts — exactly the false-short class the verification step exists for). Lane at −1.26% mean, n=805, hit−base only +0.031. First session since 07-29 that s1_standdown is not the binding reason — the standing invariant-#6 cap is. Note the utilities pair (PEG, CMS) plus RBA/AGCO industrials for correlation if this lane ever promotes.
• S2 — advisory only. 48 clean survivors. Lane +0.0007, n=1138, hit−base −0.058 — winning less often than its own base rate. Regime is actively unfavourable: S2 wants weak tapes and flight-to-liquidity, and this is a strong uptrend.
• S4 — advisory only. 17 clean survivors: LYG, GGG, RHP, ACA, LH, INCY, IMAX, VSNT, LBRT, BWA, EXEL, JXN, GFI, LNC, QSR, CSGP, MDLN. Lane +0.0039, n=1160 — the only lane positive both historically and forward, and still advisory. The extreme PCR tail is real (LYG 1023.6 on 8,189 puts vs 8 calls).

Risk#

• Correlation clusters: none live — the book is flat and COO is unsized. Nothing to collapse.
• Held book: analyses/scan/2026-08-07/conviction_2026-08-07.json → 0 open positions, verified by reading the file directly rather than trusting the tool's empty result (the held_book EXIT-substring false-empty was explicitly ruled out).
• Event calendar: CPI 08-12, PPI 08-13 inside the horizon; FOMC 09-16/17 outside.
• Tail caps applied: none needed — nothing sized. OI_FADE is not in the tail-capped lane list.
• Hedge note: no exposure to hedge.
• Standing forward caveat (informational, not a cap): OI_FADE's decay question is still unresolved. The −3.73% new-evidence flag reversed to +1.99% / hit−base +0.67, but on 9 units across only 3 exit-days — ANECDOTE-level N, durable in neither direction. 17 open h10 rows mature 08-10→08-21.

Data Quality#

catalyst_split.py:98 — post-catalyst boundary is inclusive of the print day. Load-bearing tonight.

post = sum(d["call_net"] - d["put_net"] for d in daily if d["date"] >= cat)

The >= counts the print date itself as post-catalyst. But the OI panel's row dated D measures last_date = D−1 → curr_date = D — it is the change settled into the morning of D, i.e. positioning established at or before the print, never after it. Verified against the panel for EWTX: the 08-06 file carries last_date 2026-08-05, curr_date 2026-08-06.

Effect on EWTX: +6,668 of the +7,047 "post-catalyst" build is the print day itself.

The two readings are opposite, and the inclusive one would have put a dead build into the sized book. Any fix must patch both catalyst_split.py and any inline re-implementation in retro_harness.py — the harness imports nothing and re-implements the gates, so patching only the lane script makes the live engine more permissive than the instrument grading it.

Carried forward, unfixed tonight (also affects the same class of check):

• earnings_gate.py one-sided bound — the window check compares ned.isoformat() <= end with no lower-bound check (first recorded 2026-08-06). Nothing turned on it tonight; both names were fail-closed either way.
• prices.parquet unpriced names (22/2,471) — ACLX, AL, APLS, BK, CFLT, CTRA, CUK, DAWN, EXAS, FOLD, HOLX, IAC, NGD, OS, PSTG, SEE, SLNO, TERN, THR, TPH, VRE, VSCO. These fail closed downstream.
• QQQ zero-gamma level field looked stale in the Phase A read (220.46 against a 721.2 spot); the 0DTE script's own flip calc disagreed with the CLI. Treated as fail-closed rather than resolved.

Housekeeping#

Truth set was 3 sessions stale at preflight (parquets ended 08-07) and was rebuilt in order — prices → returns → features → weekly_features — before any lane ran. universe.json was deliberately not rebuilt: refreshing the spine mid-cycle would change the universe the 2026-08-08 baseline was measured against, and a baseline compared across two universes is not a comparison. Preflight re-run clean (exit 0).

Regression gate (retro_harness.py --all, panel now 84 days, 2,471-ticker spine) — no code changed tonight, only data, so this is panel growth by construction:

All five within ±0.0004 on a ~15-row increment, two of them up. No regression — and with zero code delta, the "old code, new data" isolation check is satisfied trivially.