Aug 13, 2026

Regime & Verdict#

• Regime: UPTREND · ret5 +1.210% · ret10 +4.880% · dd15 −1.650% · VIX 14.63 (LOW tercile, contango) · breadth 63.4% green vs 35% bullish flow (28.4pt divergence) · directional_tradable=true · s1_standdown=false
• Bottom line: no directional edge tonight. Book stays FLAT for a fourth session. One name (NRG) is carried at WATCH. COO is CUT after four nights — the lane's own invalidation signal fired. Vol book stands aside on 0DTE.

Tonight the breakout is real — and that is the opposite of last night's finding#

On 08-12 the mechanical label read UPTREND off a trough-anchored rebound and a Phase-A agent misread dd15 = 0 as "at the highs." Tonight the same label is backed by actual structure, and the distinction is worth stating explicitly so the correction doesn't get over-applied:

• SPY 777.88, +0.70% — a genuine fresh 52-week high, both the highest close of the window and a new intraday high (779.37 vs the prior 776.85 on 08-05). Verified against OHLC, not inferred from dd15.
• The four-session stall broke. 773.26 → 773.03 → 770.56 → 772.49 → 777.88.
• The PPI print was bought, not faded. July final demand came in flat MoM (below +0.2% consensus), 4.7% YoY; the tape opened near the prior close and ran to a new high. On 08-12 the in-line CPI was faded (closed at 34% of the day's range). That is a real behavioural change across two sessions.
• dd15 is now −1.650%, read correctly per scripts/_regime.py:13 as min(last 15 closes) / close_10_sessions_back − 1 — not a peak-to-trough drawdown.

The regime claim was arbitrated against the panel (regime_check.py --claim-label UPTREND --claim-ret10 0.0488, exit 0).

Three things argue against reading the breakout as risk-on confirmation#

1. The price/flow breadth divergence WIDENED to 28.4pt (price 63.4% green vs options-flow 35% bullish), from ~18pt on both 08-11 and 08-12. Price breadth surged; bullish options positioning did not follow.
2. QQQ did not confirm. It ran harder on rate-of-change (ret5 +2.44%, ret10 +7.10% vs SPY's +1.21% / +4.88%) but is still −2.2% below its own 06-03 high (748.65). The index leading the tape has not made a new high.
3. The regime mechanically expires inside the holding period (below).

The label flips to CHOP on 08-18 — on flat price, again#

ret10's anchor rolls onto the 08-04 CPI-gap close (771.33, the largest one-day jump in the window):

Same conclusion as 08-12 by a different mechanism (a peak-gap anchor rolling in rather than a trough anchor rolling out): anything entered tonight at h10 spends most of its life graded under CHOP.

Directional Book (excess-scored)#

EMPTY of new starters — 0 open positions carried, 0 opened. Prior book verified flat two ways (held_book.py on conviction_2026-08-12.json returned 0 open, and that file's own book_state records 0 positions — not the known EXIT-substring false-empty).

COO — CUT after four nights at WATCH#

The lane's own mechanical invalidation fired. oi_build.py printed its documented flag:

COO  net_5d=+5,415 (call +5,719 / put +304)  rel_build=1.034  persistence=0.744
     2026-08-13  call      -4  put     +65  net     -69
   ** last call_net NEGATIVE -- calls genuinely closing (OI_FADE invalidation)

Verified directly by the orchestrator, not taken from the lane's summary. The precondition of the fade — a crowd still building calls — no longer holds. Four independently sufficient reasons, and they now point the same way:

• The build is unwinding (above). This is new tonight and is on its own decisive.
• Structurally it was never distributed: 74% of the 5-day net landed in one session (08-11, +4,028 of +5,415), with persistence drifting 0.659 → 0.733 → 0.744 toward the 0.85 single-day-block ceiling.
• Selection-rule discount: raw oi_net_5d rank 195 of 1,430, against a harness top-15 cutoff of 21,642 (NFLX). COO sits ~13× outside the population the +0.59% prior was measured on.
• Price rejected at, but poked through, the invalidation intraday — high $77.63 against the standing $77.42 level, closing back below at $76.60. The level itself would have needed re-pinning to $77.63.

Counterweight, weighed and judged insufficient: Healthcare flipped to rotation-OUT tonight (−$34.7M, having been rotation-IN on 08-12), which is a genuine new tailwind for a healthcare short. A sector tailwind does not resurrect a signal whose mechanical basis has expired.

Do not re-propose COO on the current build.

NRG — WATCH, 3rd night, decaying on schedule#

$119.75 · $-ADV $380.2M · Common Stock · Utilities/IPP. rel_build continues its mechanical decline 0.835 → 0.604 → 0.424 as the 08-10/08-11 print-week sessions age out of the trailing window — verified as window mechanics, not an unwind. Persistence 0.574, the cleanest shape in the cohort. Today's fresh add is +89 contracts — effectively nothing. catalyst_split stays strongly LIVE under both boundaries (97.3% shipped >= / 106.5% strict >; the print-day net was itself negative, so the known boundary defect strengthens this read, as on 08-12). Raw rank 41/1,430 — best of the cohort, still outside the measured top-15. Invalidation $124.71 confirmed, ~4.1% above spot; today's high was 122.07, no breach.

Expect it to roll off the candidate list within 1–2 sessions as the driving days exit the window. Fundamentals CAUTION is carried from 08-12 and was NOT re-verified tonight — flagged, not restated as fresh.

WEC — new name, not admitted to watch#

$110.28 · $-ADV $247.6M. Passes every hard gate but is withheld on the combination: persistence 0.814 is under the 0.85 ceiling only because the build spans two sessions rather than one (08-12 alone is 81% of it: +4,382 of +5,386); raw rank 197/1,430, weakest of the cohort; ex-dividend tomorrow (08-14), a corporate-calendar confound that can mechanically explain a call build; and the OI is chasing a live bounce off a two-week low (105.48 → 110.28) rather than fading an exhausted advance — the inverse of the lane's validated pattern. No single gate is fatal; a fresh name does not get the benefit of the doubt an established watch name gets.

Killed upstream#

• NXPI (persistence 0.983), USFD (0.992), UTHR (1.030) — demoted again on the 0.85 ceiling (oi_build.py:145 flags these as "likely a single-day block, not organic building"). All three are driven by the same 2026-08-07 session, which rolls out of the window tomorrow. USFD's and UTHR's catalyst_split reads land at 100–107% because their print-day net was itself negative — that is the arithmetic of the ratio, not a signal of strength.
• EWTX — remains CUT from 08-12 (print-day boundary artifact). Not re-proposed.
• 15 of 20 rel_build-ranked names died at the liquidity floor (sub-$50M ADV or sub-$5).

Vol Book (non-directional)#

• 0DTE-VRP — STAND ASIDE, both SPY and QQQ. VIX 14.63 is again squarely in the LOW tercile (bounds 16.2 / 17.7), which is exactly the zerodte_setup.py pooling trap. The governing conditional numbers, net of ~0.10% round-trip:
  • SPY −0.089%/day (gross +0.011%) · QQQ −0.014%/day (gross +0.086%)
  • Pooled headlines (+0.126% / +0.247%) do not govern and are recorded only to show the gap.
  • Gamma state changed tonight and is worth recording: SPY dealers flipped to POSITIVE / long gamma (spot 777.74 > ZGL 774.76), reversing 08-12's short-gamma state — verified independently by the orchestrator via uw options-structure gex. That is vol-suppressive, and it removes the second veto SPY carried last night. The sleeve still stands aside purely on net-of-cost economics.
  • QQQ fails closed on a source disagreement, same as 08-12. The gex endpoint returns zero_gamma_level 211.28 against spot 732.33 — a plainly broken calculation. (The vol lane reported a clean flip at 729.24 from a different source and declared no disagreement; arbitrated against the endpoint, the disagreement is real.) Decision unchanged either way — VRP economics veto independently.
• Earnings IV-crush — LOW, ROST, TJX structurally qualify. None sizes.

  name	earnings	IV-rank	real backwardation (median-IV)	liquidity on bracketing expiry
  LOW	08-19	81.0	~1.53 (55.0% vs 35.9%)	258 prints / 796 contracts / 25 strikes
  ROST	08-20	93.6	~1.55 (58.9% vs 38.0%)	115 prints / 313 contracts / 30 strikes
  TJX	08-19	80.1	~1.47 (39.8% vs 27.1%)	297 prints / 1,622 contracts / 22 strikes

  • The CLI's headline backwardation and implied_move_perc fields are both unusable here. Headline front-end ratios (LOW 3.77, ROST 4.64) are dominated by size-1 prints on deep-ITM strikes reading up to 450% IV; the size-weighted/median figures above are the real ones. Headline implied moves (LOW 1.18%, ROST 0.67%) are reading the wrong expiry — size wings off the raw straddle (LOW ATM 08-21 ≈ 5.7%; ROST ≈ 13.9%), never the headline.
  • BZ, FLNG, YRD excluded on option liquidity (zero / 35 prints / no listed expiries) — the same floor that cut QFIN and MRCY on 08-12.

TPR resolved today, and it is live evidence for the no-size rule#

TPR was last night's top structural qualifier, with wings recommended beyond the ~12.2% raw-IV move rather than the 8.4% headline. It reported today:

153.74 → open 132.70 (−13.7%) → close 128.39 (−16.5%), low 127.78 (−16.9%).

The realized move blew through even the widened raw-IV wing. Nothing was sized, so nothing was lost — which is the point. This is direct, same-week evidence that (a) the standing "no size without a per-name net-of-cost quote" rule is doing real work, and (b) the raw-IV move estimate is itself not a conservative wing on a gap-prone name. Carry this into LOW/ROST/TJX wing guidance.

Watch / Stood-down#

• MOM_SHORT — BASKET_WATCH, 6 names: AMRZ, LII, AGCO, GME, RBA, ROL. Down from 10 on 08-12. The screen is close ≤ 1.02 × w52l; the orchestrator re-derived it and independently verified all twelve candidate w52l values against Yahoo (they match to the cent). The basket is correct and reproducible — see the process note below, because the lane itself briefly claimed otherwise. Dropped tonight: WING (1.031×), STLA (1.023×), APP (1.031×), NKE (1.031×), ACM (1.044×), VICI (1.020×) — all genuinely near their lows but outside a 2% proximity screen. Sector: 50% Industrials, 33% Consumer Cyclical. GPI and HDB remain excluded on prior verdicts. Capped watch-only by invariant #6 (baseline −0.0128, measured −0.0133 n=816) — not by the regime, since s1_standdown=false.
• MOM_LONG — BASKET_WATCH, 283 names, expanded mechanically because SPY made a new high. Baseline −0.0108, measured −0.0124 (n=1104). DURABLE-N unmet; basket-only, never sizes. Sector concentration within the ⅓ cap (Financials 26.5%, Healthcare 19.1%, Tech 14.1%). Three of the top-15 (S, RBRK, P) carry earnings inside h10.
• S2 (advisory, +0.0009, n=1163) — 3 names: HBM, RY, SU. From 152 candidates: 55 dropped on issue-type/price, 13 on liquidity, 1 on earnings, and 80 on the news gate. Their news clearance is CARRIED from 08-12, not re-verified tonight — stated plainly rather than presented as fresh. The 79 unverified new candidates (incl. AMGN $680M, V $667M, HLT $356M) stay excluded fail-closed, continuing 08-12's discipline that a short verified list beats a long unverified one.
• S4 (advisory, +0.0044, n=1181 — the one lane that improved) — 14 names: GSAT, RYN, CLX, CRNX, SAN, IMAX, SMMT, LTH, COR, FDS, ETN, FFIV, GXO, CAR. Funnel 1,933 → 97 (top-5% PCR) → 30 (option liquidity) → 22 (ETP) → 18 (earnings) → 14 (prior verdicts). Only IMAX persists from 08-12's 27. Separate orthogonal ivrank_chg_5d h3 tilt (DVY, FIVN, BLV, AIZ, ITW, NI, PULS, SLM, HDV, TBIL) — reported alongside, never summed with the PCR signal.

Risk#

• Event stack inside the h10 window (08-13 → 08-27): Retail Sales 08-14, FOMC minutes 08-19, jobless claims 08-20 and 08-27, and Jackson Hole 08-27–29 landing exactly on the horizon edge — anything sized to h10 tonight would still be open as it starts. Independently sufficient for −1 tier.
• Correlation: COO/NRG/WEC computed pairwise on daily returns from the price panel — COO/NRG +0.128, COO/WEC +0.210, NRG/WEC +0.221 (141–145 paired days, full history on all three) against the 0.70 threshold. Three singletons, no collapse. Computed, not inferred from sector labels — note NRG and WEC are both utilities yet correlate at only 0.221. Full price history on all three, so this is a real pass, not the absentee-reads-as-pass trap.
• Tail: SPY's flip to long gamma is vol-suppressive and reduces realized-move risk versus last night. Recorded as context; no tail cap invented — OI_FADE has none in the ruleset.
• P0.6 out-of-regime half-cap: BINDS as an overlay, but on a different basis than 08-11/08-12. Tonight's UPTREND is genuine, not a misread — but it decays to CHOP on 08-18, inside the h10 horizon, so the horizon-relevant regime is not stably supportive. Moot in practice: nothing reached a sizing band.
• Half-cap, no auto-full. Nothing sized at any level.

Regression gate (retro_harness.py --all, run tonight)#

No lane or threshold changed tonight. The truth set was rebuilt and extended (86 → 87 panel days, 2,471-ticker spine), so the gate was run to locate the baselines.

Verdict: NOT a regression. Zero code changed; the only moving input is the panel, and n rose in every lane. That is the documented data-drift case the gate does not treat as failure.

OI_FADE has stabilized, not continued falling. −0.0002 night-over-night (vs −0.0020 the night before), and its structure is intact: hit-minus-base +0.173 against the recorded +0.17, median +0.0076 still positive and still above the mean (so not tail-driven). Read as a diluting mean on a widening panel, not an inverting edge. It nonetheless remains material to tonight's calls: the prior cited for NRG is both borrowed from a differently-ranked population and now measures ~37% below its recorded level.

Orchestrator notes / action items#

1. A lane capitulated to a challenge and manufactured a data-quality story — do not publish agent corrections unchecked. I challenged the MOM_SHORT basket after ranking candidates by pct_52w_range and finding five dropped names ranking tighter than kept ones. The lane responded that features.parquet w52l was "stale on half the eligible names" and the basket was "unreliable." Both the challenge and the reply were wrong. Panel w52l matches Yahoo to the cent on all twelve names; the lane's actual screen is close ≤ 1.02 × w52l (proximity to the low), which is a different quantity from pct_52w_range (position within the whole 52-week range). WING sits at 1.5% of its range only because that range is enormous (110→346) — in price terms it is 3.1% above its low and correctly fails a 2% screen. The basket was right all along. Two lessons: (a) pct_52w_range is not a proximity metric and must not be used as one; (b) an agent conceding under pressure is not evidence — it volunteered a "stale data" explanation rather than defending a screen that was working as designed.
2. QQQ gamma: Phase A and the vol lane disagreed; the endpoint settles it. uw options-structure gex --symbol QQQ returns zero_gamma_level 211.28 against spot 732.33 — broken, exactly as on 08-12. The vol lane reported a clean flip at 729.24 from a different source and concluded "no fail-closed needed." Fail closed. The stand-aside verdict was unchanged (VRP economics veto independently), but the reasoning would have been wrong on the record. This is now a two-night-running data defect worth a maintenance ticket.
3. catalyst_split readings above 100% are arithmetic, not signal. USFD 107.3%, UTHR 100.9%, NRG 106.5% under the strict > boundary all arise because the print-day net was itself negative, so the post-catalyst share exceeds the total. Correct behaviour, but easy to misread as extraordinary strength — for NRG it genuinely strengthens the read; for USFD/UTHR it describes a single post-print pop.
4. TPR is the first live test of the earnings-crush wing guidance and it failed the wing, not the gate. −16.5% realized against a ~12.2% raw-IV estimate that had itself already been widened from an 8.4% headline. The lane's refusal to size without a net-of-cost quote is what prevented a loss. Any future iron-fly guidance on gap-prone names should treat the raw-IV move as a floor on wing width, not a target.
5. COO's exit is clean and worth keeping as a template. The name spent four nights at WATCH and was cut by the lane's own instrumented invalidation rather than by drift, fatigue, or a fifth restatement of the same caveats. The standing selection-rule discount alone was never going to resolve it; the mechanical signal did.
6. Selection-rule mismatch remains open (standing since 08-12). Live lane ranks by rel_build; the +0.0059 baseline is top-15 by raw oi_net_5d (retro_harness.py:82-99). Tonight's cohort ranked 28–197 of 1,430 against a top-15 cutoff of 21,642. Decision still needed: add the live screens to the harness and re-baseline, or record explicitly that the live prior is inherited rather than measured.
7. S2's news gate was not re-run tonight — three names carry an 08-12 clearance on an h3–5 lane, which is a one-day-stale verification on a horizon where a day matters. Advisory-only, so nothing turns on it, but the list should not be read as freshly verified.
8. Next /calibration-audit due 2026-08-15 (per cadence, and now two days out). 41 open rows mature 08-10→08-21 and the 08-03→08-07 crash-guard block matures 08-18→08-22 — the first cohort spanning non-overlapping windows. Count distinct exit-days, not rows. Consider re-recording the OI_FADE baseline at that audit, stating the panel size (87 days, 2,471-ticker spine).
9. Watch tomorrow: whether NRG's driving sessions finish rolling out (it should leave the list); whether WEC's build survives its ex-dividend date with the confound removed; and whether NXPI/USFD/UTHR collapse once the 08-07 session exits the 5-day window.