Jul 15, 2026

Regime & Verdict#

• Regime: CHOP (uptrend-leaning). SPY 754.81. ret5d +1.26%, ret10d +1.08% — short of the +1.5% UPTREND cutoff. 15d drawdown −0.08%: SPY is pinned at its short-term ceiling (07-10 high 755.42), not trending.
• directional_tradable = TRUE. s1_standdown = FALSE — the 06-15→06-26 dip (−3.42%) recovered as a 13-session grind (+3.54%), not a 5-day thrust, so the crash/V-rebound guard is OFF and MOM_SHORT/OI_FADE shorts stay live.
• Breadth: pct_green 51.09% (257/246 of 503) — balanced. Options-flow bullish only 35.3% against price at highs: a mild price/flow divergence, advisory distribution tell.
• Sector rotation: OUT of Technology / Industrials / Healthcare; INTO Communication Services / Consumer Cyclical / Financial Services.
• Vol-state (2nd-moment only): VIX 15.67. SPY VRP FAIR (−0.023, iv30d 13.18% vs rv30 15.52%, iv_rank 13.1 — vol is cheap, not rich). QQQ VRP PREMIUM_BUYING (−0.071, iv30d 23.12% vs rv30 30.21% — realized running above implied). SPY dealer gamma NEGATIVE, spot pinned ~0.1% below the zero-gamma flip (~755) — two-way vol-expansion knife's-edge.
• Event calendar (h10 window 07-16 → 07-29): FOMC 07-29 (Tier-1, no SEP) lands exactly on the window close. CPI (07-14) and PPI (07-15) already resolved. Q2 earnings season sits inside the window — this is the binding fact tonight.
• Bottom line: 4 open carry-forward shorts, all held unchanged; ZERO new directional entries. Every new candidate with a tradable signal has earnings inside the h10 window; the only two names with clean calendars are the two lowest-quality names in the cohort. This is the framework's modal "no new directional edge" output, and it is the correct one.

Directional Book (excess-scored)#

Carry-forward — held, not re-sized (no invalidation triggered)#

New entries#

None. See "Stood-down" below for the gate-by-gate reason on each.

---

Key adjudications tonight#

MAT — the pre-registered trigger was tested and did not fire#

The 07-14 conviction note pre-registered a concrete exit: "if tomorrow's OI print is ALSO negative, treat that as confirmation of the reversal and flag for exit." Tonight's net print is +4,129 (after 07-14's −2,269). The build reasserted; 5d net +13,338. The trigger did not confirm, so MAT holds — despite excess deteriorating to −3.99%, the worst in the book. This is the pre-registered rule executing as written, not a discretionary rescue. Consumer Cyclical is rotation-IN tonight = headwind for the short (reflected in regime_fit 0.50).

LNG — first negative print is an early-warning, not an exit (consistency with the MAT precedent)#

LNG's adverse_flow_exit_check narrative says "if the daily OI print turns negative … flag for exit" — and tonight's net print is negative (−1,241). But LNG's formal invalidation field — the binding contract — reads "the 5-day net call-OI build fully unwinds without downside follow-through," and 5d net is +17,983: nowhere near unwound.

These two documents conflict, and the conflict is resolved by precedent that was vindicated tonight: on 07-14 MAT printed its own first negative (−2,269) and was explicitly held as "early-warning only, not a trigger" requiring confirmation. Tonight MAT printed +4,129 — the single negative was noise. Applying a stricter single-print rule to LNG than the one just proven correct on MAT would be inconsistent and would exit the book's best position (+3.87%) on a signal empirically shown to be a false alarm. LNG holds; flagged for confirmation next run — a second consecutive negative print changes the read.

ICE — invalidation is unfit for purpose (spec defect, flagged not fixed)#

ICE now has two consecutive negative net prints (07-14 −116, 07-15 −294), ratio decayed 0.337 → 0.177, and excess has turned adverse (−0.82%). Its stated invalidation has two clauses and neither can do its job:

1. "Close back above the pre-build breakout level" — untestable against ICE's actual path. ICE never fell below any level; it rose monotonically from $122.91 (06-29) to $139.84 tonight. A "close back above" condition can never fire on a name that never went below, no matter how wrong the short gets.
2. "5-day net build fully unwinds" — not met (5d net +5,037).

So ICE holds by the letter of its rule while having no functioning exit. This is logged as a spec defect. Re-specifying it is a rule change and must clear python3 scripts/retro_harness.py --all first — it does not ship tonight.

ORCL — the 07-14 give-back concern was correct (calibration evidence)#

The 07-14 hold-note explicitly flagged give-back risk after ORCL realized +3.10% excess in one day (~3.8× the lane's entire h10 mean). It gave back all of it in one session (+3.10% → +0.03%) on a $127.94 → $132.49 bounce. The invalidation still fires on neither clause (RSI 30.41, close « $144.22), so it holds at starter. The concern was correctly identified and is now a data point for /calibration-audit: front-loaded excess in this lane mean-reverts.

---

Watch / Stood-down (gate-by-gate)#

OI_FADE — the strongest lane produced a real cohort that the event gate disqualifies#

The h10 window is 07-16 → 07-29 (10 trading days). Q2 earnings land inside it for four of five candidates:

DINO deserves its own paragraph because it is the best signal of the night and is being stood down anyway. It is a textbook OI_FADE setup: crowded, accelerating call build into a fresh 52-week high ($83.93, exactly 100% of range). But sizing it would stack four independent risks — (a) shorting a fresh 52w high, i.e. the "fading strength" case where hard rule 2's up-biased-tape discipline binds hardest; (b) earnings 07-28 inside the window; (c) FOMC 07-29 on the window close; (d) CHOP regime. The 07-14 precedent is decisive: LNG was sized starter with only two such risks, and its acceptability was argued explicitly on "earnings 08-06 safely outside the h10 window." DINO has strictly more risk than LNG had, plus the one disqualifier LNG lacked. Consistency therefore puts DINO below starter → watch. Note also a genuine cross-lane conflict: at 100% of 52w range DINO is simultaneously a MOM_LONG basket constituent (long) and an OI_FADE candidate (short). Not summed, per hard rule 2 — noted only; MOM_LONG's median ≈ 0 makes it weak evidence either way.

No fundamentals-gate was spawned tonight, deliberately. Phase D step 4 gates only names about to be sized; nothing is. And fundamentals is a veto channel only, never a promotion mechanism (research/80) — it could not rescue any watch-tier name, only cut it further. Spawning it could not have changed a single disposition.

MOM_SHORT — 20 names, nothing sized#

Top candidate PNR verified at exactly 0.0% of its 52w range ($64.33, at its low) — a clean near-low name — but earnings 07-28 falls inside the h10 window. Same event gate. MOM_SHORT's +0.81% maps to the starter band; an event-gated starter lands at watch. Cluster: IND (CPRT/ORLY/CNM) tagged headwind on Industrials rotation-OUT.

MOM_LONG — 189 names, basket-only by lane design#

+2.35% mean but median ≈ 0 and hit-base −9.3pp: tail-driven, never a per-name HIGH. cluster_id: mom_long_basket_20260715. FIN cluster (11 names) carries HIGH rate-beta correlation into FOMC 07-29. Watch/basket only.

S2 liquidity-reversion — 56 names returned, all watch; lane defect found#

The lane returned 56 "≥90% buy-side concentrated" names with validated_excess_h5 empty for every row — the Phase C score is literally undefined, so nothing can size regardless. But auditing the classifier surfaced a real defect:

A midpoint print is by definition neither buyer- nor seller-initiated — a negotiated cross carrying zero directional information. The lane classifies mid prints as "buys," manufacturing a 90%+ one-sided read out of ~95% directionally-null volume. ADBE's headline "90.1% buy-side" is really 3.2% buy-side and 92.6% nothing.

Under a defensible classifier (price ≥ ask), only 15 of 1,294 liquid names clear 90% — 1.2% of the universe, genuinely extreme — vs the lane's 56. Of the lane's headline names only MNST (90.8% strict) and partially BMY (81.2%) reflect real at-or-above-ask aggression. Genuine strict-qualifying cohort: HEFA, MNST, SCZ, TLH, AAAC, CRNX, SCHQ, FESM, GGB, IGOV, FLDR, GSK, EWW, EWJV, BSBR (mostly ETFs — excluded by lane rule anyway).

Same artifact family as the JPST tiny-base trap and the single-day-print family: a filter that looks selective while selecting noise. Logged as a proposed rule-set fix; does not ship tonight — any threshold change must clear retro_harness.py --all.

S4 sentiment-contrarian — 31 names, advisory-only by lane design → watch; news gate unreliable#

• SLAB is Silicon Laboratories Inc. — the lane called it "Broadcom" and attached Broadcom's EU-antitrust and Apple $30B chip-deal news to it. Broadcom is AVGO; wrong company entirely. Verified against live Yahoo (SLAB → Silicon Laboratories Inc., $218.19). Nothing sized on it (S4 is advisory-only), but that error would have been load-bearing in any lane that could size. S4's catalyst verification is not trustworthy tonight.
• MAT appears in the S4 long cohort (PCR 34.1) while we hold MAT short via OI_FADE — a direct cross-lane contradiction. The S4 lane correctly killed it itself: tagged catalyst_informed because the put-heaviness follows Goldman's SELL downgrade and Roth/Citi target cuts, i.e. the puts are informed, not hedging-overpricing. The lane's own exception logic worked. No conflict materializes.
• ivrank_chg_5d h3 tilt: no fresh data — features.parquet max date is 2026-06-28, stale vs the 07-15 trade date. Not surfaced.

Regime-data integrity note (does not change tonight's gating)#

The momentum and sentiment-contrarian lanes both independently reported ret5d = −2.38%, ret10d = −1.19% — sign-flipped and wrong. Live Yahoo and Phase A both give +1.26% / +1.08%. Two lanes emitting identical wrong figures indicates both hit the known stale-regime fallback (harness-date-stale-fallback: retro_harness --date silently returns stale regime past the truth-set edge). Impact contained: both lanes were passed the Phase A context as binding and their CHOP label coincidentally matched, and gating tonight used the live read throughout. Flagged because it is a live, reproducible bug and the coincidence of the label matching should not be relied on again. Both lanes also carried stale 52-week levels (PNR quoted "$69.93" vs actual $64.33; DAVE "$419.86" vs actual $438.19) — the names still qualified on the live read, but the lane-reported figures should not be trusted.

---

Vol Book (non-directional, delta-neutral, net-of-cost, advisory — 0 directional points)#

0DTE-VRP — HARD STAND-ASIDE, both legs#

scripts/zerodte_setup.py returned GO_PREMIUM_SELL_INTRADAY for both SPY and QQQ. Overridden on the live read — that verdict is built only from VIX-tercile + GEX-regime and has no VRP input at all.

• SPY — stand aside. VRP FAIR (−0.023), iv_rank 13.1: vol is cheap, not rich — not a premium-selling setup by definition. Dealer gamma NEGATIVE with spot ~0.1–0.3% below the zero-gamma flip (~755) = genuine two-way vol-expansion knife's-edge, not the quiet long-gamma case the wing-sizing backtest assumes. That backtest contains no vol-shock day and does not condition on flip-proximity.
• QQQ — stand aside. VRP PREMIUM_BUYING (−0.071): realized is running above implied, so selling premium here sells cheap vol — it fights the lane's own measured edge. Front-end backwardation confirmed two independent ways (front/far ratio 1.091; 0DTE IV 31.2% vs 30d 24.4%), hitting the lane's explicit stand-aside gate.

QQQ GEX tool is broken — confirmed independently#

Phase A flagged zero_gamma_level: 300.35 against spot 717.41 as anomalous. Confirmed and worse: total_gex is negative (−354,621,492) while the tool's own regime label says "POSITIVE" — internally contradictory. Hand-reconstructing the per-strike cumulative sum puts the real local flip at strike ~720–722, i.e. spot 717.41 is just below it, in negative-gamma territory — the same knife's-edge as SPY, not the comfortable long-gamma regime the label implied. This reinforces the QQQ stand-aside rather than offsetting it. Data-quality flag: do not read the QQQ GEX label without reconstructing.

Earnings IV-crush SELL-VOL — panel-confirmed dates only, advisory#

The conflicting web dates for big tech are resolved by the panel: TSLA/GOOG/GOOGL 07-22; MSFT/META 07-29 (outside horizon). None clear the iv_rank ≥ 80 floor anyway (TSLA 40.3, GOOGL 72.7, GOOG 74.8) — excluded on the vol-rank gate regardless of date.

Excluded on options-liquidity despite passing the equity floor: MAN (151 near-term contracts, thinning to 1–7), KALU (2 contracts). Excluded on price/ADV floor: TNL, CCS, FNGR.

Net-of-cost caveat (binding): this repo carries no calibrated per-name net-expectancy backtest for earnings IV-crush. RESEARCH/20 §2.1 and RESEARCH/30 §3.1 validate the mechanism only. Per the permanent-advisory rule, this is quoted as directionally-positive-in-mechanism but advisory-only — not a sized net-expectancy figure. Defined-risk iron fly, wings set outside the implied move with real room, half-cap. TEL note: it is simultaneously an OI_FADE short candidate (stood down) and a vol-crush name — different lanes, not summed; its 07-22 earnings is exactly why the directional short was gated.

---

Risk#

• The dominant risk tonight is calendar, not price. The h10 window (07-16 → 07-29) contains Q2 earnings for essentially every quality name in the OI_FADE and MOM_SHORT cohorts, and closes on FOMC 07-29. Mid-July structurally cannot support a clean 10-day directional book — that, not signal absence, is why nothing new sizes. DINO's signal is real and was stood down purely on this.
• Open book concentration: 4/4 positions are shorts; 3/4 are OI_FADE. That is a single-lane, single-direction book. If the OI_FADE mechanism is regime-impaired in this grind-up tape, all four correlate. No new short was added tonight, which incidentally caps that exposure.
• Carry-forward P&L is bifurcated: LNG +3.87% and ORCL +0.03% vs ICE −0.82% and MAT −3.99%. Net book excess ≈ −0.23% across four starter positions.
• ICE has no functioning exit (spec defect above) — the single highest-priority fix in the queue.
• Rotation: Consumer Cyclical (MAT) and Financial Services (ICE) are both rotation-IN = headwinds for those two shorts, reflected in their regime_fit.
• Every deviation from a lane's implied ceiling tonight was a downgrade. No name was upsized on the strength of any move, signal, or fundamentals verdict. No position was trimmed on profit — no such rule exists and none was invented.

Proposed rule-set fixes (queued, NOT shipped — each requires retro_harness.py --all to pass)#

1. S2 buy/sell classifier: stop counting midpoint prints as buy-side; require price ≥ nbbo_ask for "buy" and add an at_mid_share exclusion. Would cut tonight's 56 → 15.
2. ICE-class invalidation spec: "close back above X" is untestable on monotonically-rising paths. Invalidations need a clause that can fire on the actual path realized.
3. S4 news gate: ticker→company resolution must be verified against a primary source (the SLAB/Broadcom error).
4. Single-day-print concentration floor for OI_FADE (max single-day |build| ÷ |5d net| ≥ ~85%) — still unpinned, still caught only by lane judgment; third consecutive scan flagging it.
5. Stale-regime fallback: lanes calling the regime helper past the truth-set edge get silently wrong ret5/ret10. Should fail loudly.