Jul 29, 2026

Regime & Verdict#

• Regime: PULLBACK (confirmed downtrend) · ret5 −2.400% · ret10 −3.360% · dd15 −3.360%
• Vol-state: vol-AMPLIFYING / short-gamma. ^VIX 18.21 → 20.66 (+13.5%, intraday 17.45–20.88 — a genuine vol-regime-shift day). GEX FULLY_NEGATIVE on both indices: SPY total_gex −$2.78B (7/50 strikes positive), QQQ −$1.18B (2/50 positive). Zero-gamma level not computable. Dealer hedging flows amplify moves, not dampen them.
• Breadth: confirming, not diverging. 32.5% bullish (2,043/6,279), pct_green 36.2%, avg change −0.89%. Red tape and red breadth — no distribution divergence.
• directional_tradable = TRUE · s1_standdown = FALSE
• Tech-led, not broad: SPY 729.46 (−1.54% today) ret5 −2.40%; QQQ ret5 −6.18% / ret10 −7.80%; IWM ret5 −1.78%. The AI-capex repricing that opened 2026-W30 is still the driver.

Bottom line: no new directional starters. The one name that cleared every mechanical lane gate (HRI) was VETOED on fundamentals. The book is a single held short (PLNT) carried at unchanged half size on its own pre-registered exit rule, three sessions from a binding exit. Vol book is thin by design: 0DTE stands aside.

Why s1_standdown is affirmatively FALSE (not merely mechanically silent)#

The guard fires on two mirror conditions and neither holds:

• up-thrust / V-rebound leg (ret5 > +2.5%, or > +1.5% off a ≥2% dip) — N/A, ret5 is −2.40%.
• unconfirmed-dip leg (ret5 < −2% AND ret10 > −2%) — ret5 −2.40% satisfies the first clause, but ret10 −3.36% is not > −2%, so the downtrend is confirmed. This is the regime the guard is designed to stay out of and let the short leg work.

The 2026-07-27 near-52w-low squeeze (+2.4% mean excess, 81% positive) was exactly the unconfirmed bounce the guard exists to catch — and it resolved as a failed bounce, not a reversal. It has now been punished for two straight sessions.

---

Directional Book (excess-scored)#

No new starters. One position, unchanged size.

HRI — the near-miss, and why it was dropped#

HRI was the only name in the OI_FADE top-15 to clear every mechanical gate, and it cleared them cleanly: oi_net_5d +6,673 (call +6,742 / put +69), persistence_ratio 0.568 (organic accretion, not a single-day block), oi_rel_build 1.62, liquidity PASS, forward earnings PASS (next print 2026-10-27), and catalyst_split = LIVE_BUILD — 41.4% of the 10-day build accrued on or after its 07-28 print, i.e. the call crowd kept adding through and after the report rather than resolving pre-print (the SSNC failure mode).

Fundamentals inverted the thesis:

• Q2 (07-28): adj EPS $1.43 vs $0.73 est (+94.7%), revenue $1.204B vs $1.132B (+6.4% beat, +20.2% YoY)
• FY26 guidance RAISED (rental rev $4.375–4.475B from $4.275–4.4B; EBITDA $2.05–2.125B from $2.0–2.1B)
• Insider MSPR +23.89 (buying)
• The ALLE post-print PT-raise cluster is present: JPM 07-13 $140→$165, Citi 07-14 $155→$175, Keybanc 07-29 (post-print) $165→$185 — a fresh cycle-high target after the print.

That is the precise pattern that vetoed ALLE on 2026-07-24 (beat + guide-raise + multiple post-print PT raises). Per that precedent a contradiction of this scale is DROPPED, not watch-listed. Independently: HRI is already −14.6% from its 07-28 intraday high ($167.50 → $143.02) in 1.5 sessions, so the entry would also have been shorting into a completed collapse with a bounce tail — the lane's validated edge is measured on the general population of heavy-call-build names, not on names that have already dumped 14%.

One thing the veto does not resolve: the OI signal itself was clean, and the equipment-rental complex carries capex-cycle exposure into a tape repricing capex lower. If the analyst tape turns, this name re-qualifies.

PLNT — held short, honoring the rule#

Reconciled against the 07-29 close plus tonight's OI panel.

• Entry $55.20 (2026-07-20) → close $56.19. Gross −1.79%.
• Short excess since entry = −3.50% (PLNT +1.79% vs SPY −1.70%). This is the scoring currency and it is materially worse than the gross number. held_book.py prints "gross −1.79%" (since entry) beside "excess −1.99%" (today only) — those are different horizons and must not be conflated.
• Exit rule NOT triggered: 07-29 print is call +341 / put +256 = net +85. Calls still net-opening.
• Signal quality is degrading, materially: net_5d +1,451 but persistence_ratio 0.81 (up from 0.668) — 81% of the entire 5-day build is the single 07-28 print (+1,176), one point off the 0.85 single-block disqualifier. oi_rel_build has collapsed to 0.025. Screened fresh tonight, this name would likely not clear Phase B's own organic-build bar.
• Stock momentum runs against the short: ret5 +3.96%, ret10 +7.56%, RSI14 62.1.
• Fundamentals CAUTION (unchanged from 07-28, no movement). Named soft-invalidation ruled NOT FIRED — only 2 raises exist (Canaccord $80→$82, JPMorgan $60→$62) and both are same-day as entry (07-20), alongside a Deutsche Bank cut ($61→$55, also 07-20). Zero incremental analyst actions in the 7 sessions since. Counting ambiguity flagged: one further raise makes a literal count of 3.
• Squeeze axis benign: short float 11.79%, days-to-cover 3.67. Securities class-action filings 07-21 (short-supportive).

Decision: HOLD at half — honoring the pre-registered rule, not overriding it. The rule's early-exit valve is a genuinely negative call_net print, and tonight's decelerated but positive +85 does not cross it. Invariant #5 says a pre-registered rule is not bent mid-hold on discretion, and nothing tonight clears a bar — the decay is a continuation of the trend the rule was already built to catch. Downside is time-boxed either way: 3 sessions (07-30, 07-31, 08-03).

08-03 is BINDING, not advisory. earnings_gate.py puts PLNT's next print at 2026-08-06, three days after the exit. Exiting on schedule clears the print; any drift past 08-03 walks a losing short into earnings it was never sized for. This is the ORCL/PRMB horizon-drift lesson with a catalyst attached.

Stop tightened: $59.50 with confirm cut from 2-day to 1-day. The level is unchanged; only the confirm window moves. Reason is arithmetic, not sentiment — with three sessions left, a 2-day confirm can only complete by the 07-31 close, which exits 08-03, the h10 date the rule already delivers. As configured the stop was decorative: it could not get you out one session earlier than the rule. At 1-day confirm, a close above $59.50 on 07-30 exits 07-31 — one session of real protection. Adjusting a discretionary overlay added 07-28 is not a pre-registered-rule change, so invariant #5 is untouched; pulling the level in (e.g. to $58.50) was considered and rejected, because substituting a discretionary exit for a live rule with 3 sessions left is exactly what #5 guards against.

Vigilance instruction for the remaining sessions (not a rule change): a flat or negative print on 07-30 or 07-31 is a same-session exit — do not wait for confirmation math.

---

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

0DTE-VRP premium-selling: STAND ASIDE, size_scalar 0.0 (SPY and QQQ). zerodte_setup.py returns sell_premium=false for both on its VIX-spike gate. This overrides a trailing-60d net-of-cost positive mean (SPY +0.142%/day, QQQ +0.269%/day after a 0.10% cost charge) for good reason: the validation sample contains no vol-shock day, so tonight is plausibly the unsampled left tail itself. Reinforcing this — the SPY VRP print of +5.11% (iv30d 17.3% vs realized 12.2%, regime PREMIUM_SELLING) rests on a backward-looking realized-vol denominator that has not yet absorbed today's spike; it will compress or flip on the next print. Selling premium into FULLY_NEGATIVE gamma on a stale VRP read is the textbook worst short-vol setup.

Earnings IV-crush SELL-VOL (advisory). All candidates show front-end backwardation, confirming imminent-event pricing rather than stale IV. Quoted net of a ~0.4–0.6%-of-notional single-name round-trip cost.

Structures: short iron fly / wide strangle at 1.1–1.3× the implied move (not tight-to-the-move — a pre-print gap on a negative-gamma tape blows through narrow wings). Defined-risk, delta-neutral throughout.

Caveat on MSFT/META: uw insights earnings-play did not surface either name despite their filters clearing elsewhere (a coverage gap in that tool this session, not evidence of no event). Their implied moves are back-calculated from the AMZN/AAPL implied-move-to-front-IV ratio (≈0.65× IV×√T). Directionally right, but confirm against a live ATM straddle before execution — do not size mechanically off those two figures.

---

Watch / Stood-down#

• MOM_SHORT — 11 names, watch-only, zero sizing. The PROVISIONAL cap (2026-07-18 audit, pending durable-N ≥ 30 forward revalidation) binds regardless of setup quality. Prior is −0.03% mean — the negative mean is the binding read, not the +20.6pp hit-base. Notables: FRVO, MIR, INIO, SKHY, CNM, KLAC (high IV-rank caution), TSLA (barely inside the screen, 1.26% above its 52w low, RSI 25.4 — ambiguous between continuation and mean-reversion). 4 of the 11 were ETFs and should never have been listed — see Data Integrity below.
• MOM_LONG — watch-only by the rubric, not by discretion. Validated prior +0.18% mean / −1.06% median, tail-driven, basket-only. Phase-C tiering puts HIGH at ≥ +1% and MEDIUM at +0.3–1%, so +0.18% cannot be sized above watch. This holds despite genuinely supportive cohort data (below) — the 1-day cohort excess is not the h10 basket statistic, and three good days in a flight-to-quality tape do not establish h10 edge. Separately, the lane's proposed basket included ROST, which carries a standing fundamentals VETO/DROP (2026-07-20) — the lane did not run prior_verdicts.py against its own basket.
• S2 (dp-reversion) — nothing advisable. 12 names cleared all gates (9 failed liquidity, 34 blocked on the trading-day earnings gate, 4 carried blocking prior verdicts). A +0.06% prior cannot carry a long-tilted mean-reversion trade through PCE + AMZN/AAPL into amplifying gamma. Three survivors (VC, TXNM, REZI) were >90% one-sided SELL-side — informed liquidation, not a fadeable overshoot.
• S4 (pcr-fade) — advisory only, nothing sized. 12 gate-clean names, earnings gate verified clean 12/12 on trading-day windows. The lane's own mechanism caution is the binding read: a high PCR on a −1.54% day with VIX spiking into negative gamma is rational hedging demand, not a fadeable sentiment extreme. 3 of 12 (VRSN, CARR, VLY) show puts net SOLD — the premium-sign contradiction that held HOG back on 07-28.
• HOG — WATCH, unchanged. Re-qualified on tonight's data (vol_PCR 5.52, call_vol 226 clears the 75 floor, OI-PCR 1.25 = 86th pctl, earnings 2026-11-03 clear of h10). Still no fundamentals-gate verdict on record; net_put_premium is now positive (+$15.5K) but on a thin book. Advisory-lane cap binds either way.
• ivrank_chg_5d h3 advisory tilt: BANC, BBY, ANF, BURL, SHAZ, CCXI, ULTA, NVDA. BURL also appears in S4 — noted as evidence-type diversification, explicitly NOT summed (invariant #2). This list was not ETF-filtered — BKLN and JEPQ are ETFs and are struck.

Corrected 52w-range cohort data (I recomputed this; the momentum lane's table was wrong)#

Same-day excess vs SPY, valid pct_52w_range ∈ [0,1] only, ETFs excluded, price ≥ $5:

The normal factor sign (higher range percentile → better forward return, rank-IC t=+6.8) has now held three consecutive sessions and is widening. The 07-27 squeeze cohort is being punished hard. The momentum lane reported near-low as −0.69% on 07-27 when it was actually +2.387% — it inverted the single most important recent data point, so its own regime-fit reasoning was treated as unreliable and this recompute was used instead.

---

Risk#

• Correlation clusters: none live. PLNT is the sole position; HRI is dropped, so no multi-name cluster exists.
• Event calendar inside PLNT's 3 remaining sessions: FOMC landed 07-29 (priced). PCE 07-30 8:30am and AMZN + AAPL 07-30 AMC both fall inside the window on a vol-amplifying, fully-negative-gamma tape — market-wide gap/whipsaw risk rather than a name-specific catalyst. NFP 08-07 falls after the binding exit and is irrelevant to this hold. PLNT's own 08-06 print is the reason 08-03 admits no drift.
• Tail caps applied: PLNT held at half (no auto-full, invariant #4) with CAUTION already −1 tier; tail FLAG logged for persistence 0.81 / rel_build 0.025. Vol book sized for the unsampled left tail — 0DTE zeroed, META halved, AAPL/MPWR/COHU excluded. MOM_SHORT capped to zero by the standing PROVISIONAL cap.
• Hedge note: no hedge warranted. The book is one half-size short with a time-boxed exit inside three sessions and nothing correlated to stack against it. The live tail is idiosyncratic (a squeeze into a decaying signal), not systematic; short float 11.79% / DTC 3.67 is not a forcing function.

Data integrity — three findings from tonight (proposals for /calibration-audit, NOT hand-patched)#

1. ETFs are leaking into lane universes, and the fix is already in the panel. The Stock Screener export carries an issue_type column that labels ETFs cleanly. Tonight: the momentum lane's ETP filter claimed to gate on it and still passed AAPD, ASTX, RAM, IGIB (4 of its 11 MOM_SHORT names — all issue_type=ETF), mislabelling two of them as operating companies ("biotech", "Aerospace micro-cap"). S4 passed EWH (iShares MSCI Hong Kong) through a filter that caught AIQ/COMP/IYR/KIE/KRE/LQD/MDY. The ivrank_chg_5d list was not filtered at all (BKLN, JEPQ). This is the same universe pollution that inflated the OI_FADE prior ~2× (leveraged/thematic ETPs). Recommend a shared issue_type IN ('Common Stock','ADR') helper that every lane calls, rather than per-lane re-implementations. (SKHY was checked and is issue_type=ADR — legitimately in-universe, not a leak.)
2. pct_52w_range is out of range for ~3% of names, concentrated exactly at the decile extremes the momentum lane trades. On 07-29, of 1,865 names with a value: 35 have pct > 1 (close above the recorded w52h) and 28 have pct < 0 (close below w52l). Those land in the traded deciles at 35/233 (15%) of near-high and 28/171 (16%) of near-low. Mechanically this is UW's w52h/w52l lagging names that just made a new extreme, so cohort membership is broadly direction-consistent — but the values are unusable as a continuous factor and produce the impossible figures the lane reported ("−16.81%", "+125.8%"). A harder case exists: KLAC shows w52h 2431.29 with w52l 186.75 and close 170.19 — an internally inconsistent triple, not mere staleness. Recommend clamping to [0,1] against a chart-API-derived 52w range and quarantining inconsistent triples.
3. Truth-set correlation coverage gap, now recurring. The pairwise HRI/PLNT correlation recompute against data/prices.parquet returned n=0 overlapping rows — the same class of gap flagged for HOG on 07-28. Moot tonight (HRI is out on fundamentals, PLNT is a sole position), but gate 3 will fail silently the first night two names both clear gate 4 and need a live correlation number. Worth fixing before it is load-bearing.

Regression gate: NOT RUN AND NOT REQUIRED — no lane logic or threshold was changed tonight. All three findings above are logged proposals. Any of them that becomes a gate must first clear python3 scripts/retro_harness.py --all against the 4ec89dc baseline (MOM_LONG +0.0018 · MOM_SHORT −0.0003 · OI_FADE +0.0070 · S2 +0.0006 · S4 +0.0044). The PLNT stop-confirm tightening is a discretionary overlay on a single position, not a lane threshold, and does not trip this requirement.