Aug 14, 2026

Regime & Verdict#

• Regime: UPTREND · ret5 +0.400% · ret10 +3.920% · dd15 −2.350% · directional_tradable=TRUE · s1_standdown=FALSE
• Vol-state: VIX 14.25 (LOW tercile), falling from 16.50 on 08-04. SPY GEX positive (+$2.02B), spot pinned essentially at the flip (776.06 vs zero-gamma 774.67). QQQ GEX positive (+$652M) but spot within 0.07% of its flip, and the two gamma reads disagree. SPY VRP FAIR (iv30d 11.76% vs realized 12.49%); QQQ realized (23.24%) running hotter than implied (18.17%).
• Breadth: 250 advancers / 245 decliners, pct_green 49.7%, on a day SPY itself closed −0.20%. Structurally healthy: 246 names within 5% of 52w highs vs 64 within 5% of lows (~3.8:1).
• Bottom line: No directional edge today. ZERO new starters; the book stays FLAT for a 5th consecutive session. The one name carried at WATCH (NRG) was CUT on a fired invalidation.

Two regime facts that constrain everything below#

1. dd15 is not a drawdown. dd15 = min(last 15 closes) / close_10_back − 1. Tonight's −2.350% is driven by the 07-29 dip to 729.46 measured against the 07-31 anchor (747.03) — an excursion behind us, not current distance from highs. Recomputed by hand: −2.352%. SPY is 0.4% off its high, at 98.0% of its 52w range, having printed a fresh 52w high (779.37) on 08-13. This metric was misread in the opposite direction on 08-12; it is stated explicitly here to stop the error recurring in either direction.

2. The regime mechanically expires inside the horizon. Holding price flat and rolling the 10-session window forward:

Anything entered tonight on h10 (window 08-14 → ~08-28) spends roughly 9 of 10 sessions outside the regime that would justify it. No further price movement is required for this; it is pure window arithmetic off the 08-03→08-07 rally rolling out.

Directional Book (excess-scored)#

Empty. No candidate reached a sizing band.

Six names cleared the full OI_FADE mechanical stack (STUB, FDX, GEN, INSW, BSY, ITRI) — liquidity, earnings gate, catalyst_split, issue_type, news. All six failed at Phase C, before the gate stack was even reached, on one decisive fact:

Not one of them clears the raw top-15 cohort the +0.59% prior was measured on. Verified independently by the orchestrator on the 1,503-name features spine the harness itself uses:

validated_excess is therefore not established for any of them, so none can reach MEDIUM (+0.3–1%) or HIGH (≥+1%). This is the same selection-rule discount that has held COO and NRG at WATCH for four consecutive nights. Citing +0.59% here would be citing a number measured on a different cohort.

Two names carry an additional, independent problem:

• INSW — cut, not watched. At 97.9% of its 52w range it is effectively at a fresh high. Shorting it directly opposes the strongest cross-sectional factor in the study (pct_52w_range, rank-IC t=+6.8) — the same factor whose long leg would flag this name. The lane itself flagged it as closest to the ENS breakout-chase failure mode, and its $50.3M ADV sits a hair over the floor.
• STUB — the NRG setup exactly. 10.6% of range, −71% off its high, $8.08, heavy call accumulation (23,441 fresh calls today) on a bounce after a bad print. Beaten-down name + heavy call build + short = precisely what just cost this book its only carried position.

Cohort-level caution (new tonight)#

Tonight's call-OI building is not froth at the highs — it is bottom-fishing in broken names. Of the raw oi_net_5d top-15, 8 sit below 30% of their 52w range: MSTR 3.8%, TTD 3.0%, MARA 15.1%, NFLX 21.2%, TSLA 22.3%, ACHR 22.5%, META 25.2%, SPCX 29.1%.

That makes an OI_FADE short in this cohort structurally the same trade as MOM_SHORT — the one lane knowingly carrying negative excess (−0.0128) — while carrying the V-rebound risk NRG just realized. Treated as a cohort-level caution, not six independent names.

Vol Book (non-directional)#

Delta-neutral, quoted net of cost, 0 directional points.

0DTE-VRP — STAND ASIDE, both symbols.

The pooled headline says sell premium on both. It is the wrong number: pooled net_expectancy averages across all VIX terciles, and tonight sits in the LOW bucket where the conditional net is negative for SPY and ~breakeven-negative for QQQ. QQQ fails independently on gamma-sign disagreement (aggregate sign drives range/wing-width; spot-vs-flip drives the sell decision — they answer different questions and disagree here, so it fails closed).

Earnings IV-crush — ROST, advisory only. Prints 08-20 (IV-rank 93.2, backwardation, front/back 4.46). ATM 245-strike straddle on the 08-21 bracketing expiry: call $9.00 + put $8.95 = $17.95 → 7.31% implied move (median IV 64.5%). Net of a half-spread cross ($1.15): ~6.8% net credit. Wing floor: treat 7.31% as a lower bound and set wings no tighter than ~9.5–10% OTM — TPR realized −16.5% against a ~12.2% raw-IV estimate on 08-13, live evidence that straddle-tight wings breach.

This resolves last night's specific objection (no net-of-cost quote existed) but does not promote the name: the earnings IV-crush stack has no backtested harness, so it stays advisory per the standing 08-10 decision. KC / YRD / VWAV structurally qualify but are not sizable — no clean ATM pair or no usable chain.

⚠️ implied_move_perc in the screener is unusable for large-caps: it reads 0.47% for ROST, 0.60% TJX, 0.43% LOW, all printing next week at IV-rank 71–93. Straddle-derived numbers only.

Watch / Stood-down#

CUT tonight — NRG (OI_FADE short, WATCH 3 nights, never sized, no realized loss). Overdetermined on two independently sufficient grounds:

1. Price invalidation fired — close $126.24 vs standing "close above $124.71" (08-05 swing high, tracked unchanged 4 nights). Closed at the session high: O 120.00 / H 126.25 / C 126.24, +5.42% on a day SPY was −0.20% (~+5.6% day-excess against the short).
2. Mechanical invalidation fired — oi_build.py last net −1,056 (call_net −110): calls genuinely closing. Build died in sequence: +20,900 → +13,239 → +1,997 → +89 → −1,056; rel_build 0.835 → 0.604 → 0.424 → 0.395.

Do not carry to a 4th watch night. COO and WEC remain CUT from 08-13 and were correctly not re-proposed (COO's build partially re-formed but its raw rank is 304; WEC printed call_net −3,516 on its ex-dividend date, persistence 2.374).

OI_FADE watch (4): FDX, GEN, BSY, ITRI — mechanically clean, outside the measured cohort. STUB and INSW excluded above.

MOM_SHORT watch-only (6) — never sizes (baseline −0.0128). All Yahoo-verified genuinely near-52w-low, most having printed that low within three sessions:

MOM_LONG basket/watch (~275) — never sizes (baseline −0.0108), DURABLE-N unmet.

S2 — UNVERIFIED, not advisory-cleared. The lane returned 70 names as having "cleared all hard gates" but ran no news verification, which its charter requires and which prior nights enforced fail-closed. Downgraded to mechanically-screened-but-unverified.

S4 advisory (harness-consistent top-15): RYN, UTZ, LTH, COCO, BMRN, CHRW, LBRT, FDXF, MOH, ECHO, STT, CAR, PGR, CAMT, FITB. Never sizes (+0.0035).

Risk#

• Event cluster inside the horizon. PCE + Q2 GDP 2nd estimate (Wed 08-26), NVDA earnings (Wed 08-26 after close), and Jackson Hole with the new Fed Chair's first keynote (Thu–Sat 08-27→29) all land within three days of each other — right at the h10 exit for anything entered tonight. The mechanical CHOP transition (08-18) precedes that cluster, so the back half of the window is range-bound-and-ambiguous into peak event risk.
• Regime cap. The out-of-regime half-cap overlay is binding in principle (UPTREND for 1 of ~10 sessions) but non-binding in practice — nothing reached a sizing band.
• Correlation. Not load-bearing tonight; no positions. Noted for future OI_FADE cohorts: the raw top-15 is dominated by two tight clusters — crypto-proxies (MSTR, MARA, IREN, WULF, RIOT) and mega-cap tech (NVDA, GOOGL, AAPL, TSLA, META, NFLX, AMZN). Shorting the raw top-15 is close to shorting QQQ-beta plus crypto-beta.
• Hedge note. No book, no hedge required.

Methodology findings logged tonight (follow-ups, not applied)#

Each needs a paired lane+harness fix and a full regression-gate run before shipping; none were applied tonight.

1. S4 live lane is more permissive than the harness grading it. The harness applies (call_volume + put_volume) >= 1000 and LIMIT 15; the live lane applied neither. Result: 67% of its cohort fails the harness's own floor, and 33% had call_volume < 10 (CSL: 1 call, 26 puts, PCR 26; USFD: 3 calls; SLM: 3 calls). Median call volume in the selection cohort was 25 contracts. The report above uses the harness-consistent list.
2. pct_52w_range split artifact is inside the harness, not just the lane. IESC carries UW week_52_high 408.26 against a true 816.51 — an exact 2:1 un-split-adjusted error implying the close is 185% of its 52w high. Correctly read it is 92.6% of high and fails the ≥95% MOM_LONG screen. It appeared in retro_harness.py's own MOM_LONG top-15, so the MOM_LONG −0.0108 (n=1066) baseline is measured on a partly-corrupted cohort.
3. prior_verdicts.py conflates "previously vetoed" with "previously mentioned." HBM was flagged PRIOR BLOCKING VERDICT on the strength of its own prior S2 advisory listings, where it had been explicitly cleared. This dropped one of the three names S2 carried last night on a false exclusion.
4. Lane-reported raw ranks are computed on a different population than the baseline. OI_FADE ranked on its own 1,198-ticker panel; the baseline uses the 1,503-name features spine. Conclusions matched tonight (all six far outside top-15) but the absolute figures differed materially (BSY 389 vs 235, ITRI 465 vs 273).

---

Next /calibration-audit due 2026-08-15 (tomorrow) — first cohort spanning non-overlapping exit-day windows. Count distinct exit-days, not rows.