Market scan
Jul 28, 2026
Tier mix
- MEDIUM1
- ADVISORY1
- DROP1
Tier mix — data table
| Tier | Calls |
|---|---|
| MEDIUM | 1 |
| ADVISORY | 1 |
| DROP | 1 |
Directional book
Sizing is removed from every call by policy — see the disclaimer.
PLNT
MEDIUMShortlaneOI_FADE10dOI_FADE short nearing its own h10 horizon (4 sessions left), currently losing on a short-excess basis. The mechanical exit rule (earlier of h10 or a negative call_net print) is explicit and has NOT been triggered -- tonight's record call-OI build is a re-acceleration of the very signal the lane trades, not a reason to abandon the rule early. Evidence outside the validated spec (DP block-print distribution, negative-premium call writing) corroborates rather than contradicts the short. Fundamentals CAUTION and an unexplained pop are real reasons for vigilance, not for overriding a pre-registered rule four sessions from its own resolution -- especially since the named soft-invalidation (analyst upgrade/PT raise) still has not fired. HOLD at current half size, add a discretionary risk stop ($59.50, 2-day confirm) to bound the FOMC/Mag7 event-window tail risk that a pure buy-and-hold-to-horizon posture would otherwise carry uncapped.
- Structure
- short equity, half-cap, unchanged size. Discretionary risk stop ADDED tonight (new, not part of the original pre-registered rule): close above $59.50 with 2-day confirm -- bounds tail risk through the FOMC/Mag7 event window without pre-empting the pre-registered horizon/call_net exit rule. This is a supplementary risk control, not a replacement for the exit rule.
- Entry / trigger
- Entered 2026-07-20 @ $55.20. Close 2026-07-28 $55.94, -1.51% short-excess (losing). 5d net call-minus-put OI trajectory (most recent last): +395/-4/+180/+14/+1,176 (tonight, call +1,197/put +21). persistence_ratio 0.668 (< 0.85 disqualifier). Build re-accelerated hard tonight after appearing dead on 07-27.
- Invalidation
- HOLD, not CUT or TRIM. Exit at the EARLIER of: (a) 2026-08-03 close [h10, 4 trading days remaining: 07-29/07-30/07-31/08-03], or (b) the first genuinely negative call_net print (neither triggered tonight). Supplementary discretionary stop: close > $59.50, 2-day confirm (added tonight for FOMC/Mag7 event-window risk, does not override (a)/(b)). Would flip to CUT before the horizon if: a 2nd/3rd analyst PT raise or fresh Buy upgrade lands (the still-unfired soft-invalidation), OR dark-pool buy_ratio flips decisively positive (>0.5) with volume confirming accumulation (reversing tonight's 0.00 read), OR the discretionary stop is hit.
- Losing position (-1.51% short-excess) with only 4 sessions to its pre-registered horizon and a dense Tier-1 macro/mega-cap event stack (FOMC + 5 Mag7 prints) running through the entire remaining window on a vol-amplifying negative-gamma tape
- Tonight's call-OI build re-accelerated to the largest single-day print of the window (+1,176 net) -- directly contradicts the 07-27 'build is dead, low remaining edge' read that supported holding passively; the pre-registered cut condition (negative call_net) is further from firing tonight than it was two sessions ago
- Fundamentals gate downgraded CONFIRM -> CAUTION tonight; a genuinely mixed fundamental picture (4/4 EPS beat streak + insider buying vs. guidance-cut narrative + 2.61x D/E + fraud litigation) that could break either way
- Today's +3.17% is unexplained (no dated name-specific catalyst found) and lands on the same session as the record call build -- plausibly early positioning into the 08-06 print, which falls after this position's own exit, but this is inference, not evidence
- The two new corroborating signals found tonight (DP block-print distribution, negative net call premium on the build) are NOT part of the validated OI_FADE spec (which trades raw oi_net_5d contract counts, not premium-signed or DP one-sidedness) -- they support the HOLD decision as evidence quality, not as an additional validated gate; do not hand-patch the lane spec off this reasoning (invariant #5). Log as a /calibration-audit proposal (premium-sign + DP corroboration overlay for OI_FADE) instead.
regime crashN/A
N/A -- OI_FADE is not S1/relative-weakness-gated, s1_standdown irrelevant to this lane. General out-of-strong-regime half-cap ceiling already respected (position at half, not above).
liquidityPASS
PASS -- scripts/liquidity.py: close $55.94, 20d $-ADV $94.4M, clears the $50M floor with room; confirms no repeat of the 07-24 $250M hand-multiplication error.
clusterPASS
PASS -- only other tonight's-book name is HOG, no shared theme (fitness services vs. motorcycles/recreational). See correlation-cluster note in watchlist_write_back re: HOG's absence from the truth-set panel for a quantitative recompute.
fundamentalsNOTE
CAUTION (-1 tier, NOT veto) -- downgraded from CONFIRM (07-27) to CAUTION tonight. Zero post-07-20 PT raises or fresh Buy upgrades (the named soft-invalidation trigger has NOT fired); Deutsche Bank CUT $61->$55 (07-22); Wells Fargo flat $65 (07-20); William Blair flat Hold (07-27) -- marginal analyst tape tilts bearish, consistent with the short, not against it. Contradicting: clean 4/4 EPS beat streak, net insider buying, rev +14.4%/EPS +31.4% YoY. Supporting: guidance-cut/membership-execution deterioration, securities-fraud class actions, D/E 2.61 securitized whole-business debt.
event riskNOTE
PLNT's own ER (2026-08-06 pre-market) falls AFTER the position's own h10 exit (2026-08-03) -- the print is correctly outside the remaining window, unchanged from 07-27's reasoning. FOMC (07-29) + MSFT/META/ARM/QCOM (07-29 AH) + AAPL/AMZN (07-30 AH) fall INSIDE the remaining 4-session hold window as broad market-wide vol risk (not company-specific catalysts) -- noted, does not independently veto a held short with an already-defined exit rule.
HOG
ADVISORYLonglaneS4_pcr_fade5dS4 sentiment-contrarian long candidate: a 4-session post-earnings capitulation (07-21 high $28.45 -> $24.79, -12.9%) followed by today's reversal bar (open 24.53, low 24.45, close 25.29, +2.02%/+1.78% excess, closing near the high) on an elevated (84.1st pctl) but not extreme OI-PCR, clearing the new call-volume denominator floor that killed PFGC's divide-by-near-zero artifact. Advisory-only lane (+0.44% mean prior, never sized beyond watch) combined with a contradicted premium-signed mechanism, a missing fundamentals verdict, and a dense Tier-1 event stack inside the trade horizon -- orchestrator's WATCH read is CONFIRMED, not overturned. Log the premium-sign contradiction as a /calibration-audit evidence-quality proposal (a denominator-style hygiene check analogous to the PFGC fix), not as a new hard veto -- invariant #5 (pre-registered ≠ scored).
- Structure
- WATCH only -- no position taken.
- Entry / trigger
- Would-be long entry near today's close $25.29 off the reversal-bar setup, IF sized. Not triggered tonight.
- Invalidation
- N/A -- no position. Would reconsider for starter sizing (never above starter -- S4 is capped, see CLAUDE.md lane table) post-FOMC (earliest 07-30) if: (a) fundamentals-gate returns CONFIRM or CAUTION (not VETO) once spawned by the orchestrator, (b) premium-signed options flow flips to corroborate the contract-count PCR read (net_put_premium turns positive, i.e. puts actually being bought), and (c) the FOMC/Mag7 event stack clears without a regime break.
- MECHANISM CONTRADICTED: net_put_premium -$11,791 (puts net SOLD) vs net_call_premium +$4,669 (calls net bought); bull_bear_skew +0.249 bullish -- on a premium-weighted (dollars-at-risk) basis, the put-heavy crowd S4's contract-count spec (vol_PCR 5.37, put_vol 1,649/call_vol 307) claims to fade does not exist
- Book is ~$66K total premium across 1,956 contracts (~$0.34/contract) -- too thin to confidently call a crowd in either direction
- OI-PCR 84.1st percentile (1.183) is elevated but NOT extreme (p90=1.446, p95=2.203); the same test would have similarly ranked PFGC's already-rejected 1.26, so this metric alone does not separate a real signal from noise -- the denominator floor (call_vol >=75, HOG clears at 307) is what separates them here, not the OI-PCR level
- Consumer-discretionary, rate-sensitive name entering an h5-10 window that runs through FOMC and five Mag7 prints on a vol-amplifying negative-gamma tape
- No fundamentals-gate verdict on record -- unresolved risk-filter gap
regime crashN/A
N/A -- S4 is not S1/relative-weakness-gated. General out-of-strong-regime half-cap ceiling would apply if sized, moot at watch.
liquidityPASS
PASS -- scripts/liquidity.py: close $25.29, 20d $-ADV $68.9M, clears the $50M floor.
clusterPASS
PASS -- no shared theme vs. PLNT (the only other name in tonight's book); see truth-set coverage gap note in watchlist_write_back.
fundamentalsNOTE
MISSING -- no fundamentals-gate verdict on record for HOG; analyses/scan/2026-07-28/ was empty at the start of Phase D. Flagging back to the orchestrator per gate rule 4: hold at watch until a verdict arrives, before any future starter-scale sizing. This is not independently determinative tonight since the advisory-lane cap already limits HOG to watch, but must be resolved before HOG can ever be sized.
event riskNOTE
-1 tier -- Tier-1 macro/mega-cap event stack (FOMC 07-29 + MSFT/META/ARM/QCOM 07-29 AH + AAPL/AMZN 07-30 AH) sits inside the h5-10 trade horizon on a vol-amplifying negative-gamma tape. HOG's own earnings (11-03) clear the h-window veto cleanly (see tail_cap).
Vol book
Non-directional, delta-neutral, net-of-cost — advisory only.
No vol-book entries in this scan.
Watch / stood-down
| Ticker | Tier | Lane | Direction | Invalidation |
|---|---|---|---|---|
| ORCL | DROP | MOM_SHORT | Short | N/A -- resolved via the pre-registered h10 horizon-expiry rule, not a stop-out. Note: ORCL made a new 52w intraday low ($114.50) today and reversed to close $119.96. On a SHORT, the $118.18 open fill therefore captured MORE than a close fill would have: +8.82% short-excess at the open vs +7.69% at the close (+10.16% vs +8.80% gross), i.e. closing at the open beat closing at the close by +1.13pp of excess. [CORRECTED by orchestrator: the sizer's original text asserted the opposite sign; watchlist_write_back had it right.] |
Lane status
No lane produced a conceptual fire without a sized call tonight.
Watchlist write-back
Carried forward to the next session's watchlist — not calls.
Lane glossary (3)
- MOM_SHORT
- Momentum, short leg — near-52w-low relative weakness. Regime-gated against momentum-crash/squeeze rebounds (h10).
- OI_FADE
- OI-flow fade — fades persistent multi-day net call-OI building; heavy call-OI build precedes underperformance. The most robust lane measured (short, h10).
- S4_pcr_fade
- Sentiment contrarian — fades a put-heavy (high put/call ratio) crowd long; the only sentiment leg that measured positive. Advisory size only (h5-10).
Regime & Verdict#
- Regime: CHOP · ret5 −0.990% · ret10 −1.460% · dd15 −1.820% (
scripts/regime_check.py, the same function the harness gates on) - vol-state: FULLY_NEGATIVE dealer gamma on both indices — SPY total GEX −$947.1M, QQQ −$1,076.6M → vol-amplifying, realized moves should run larger than usual in both directions. VRP FAIR both (SPY +2.77%, QQQ +0.55%) — no premium edge either side. VIX 18.21, down from 18.67, i.e. not bid into peak event risk. GEX/DEX fed the vol-state only; no directional read taken.
- breadth: price/flow divergence. Price breadth green (S&P500 71.0%, broad mcap>$1B 61.9%) but flow breadth only 36.9% bullish tickers.
uw risk market-regime= TRANSITIONAL/CHOPPY, "reduce size". SPY below its 20sma (746.65) and 50sma (744.86). Sector flow into Tech and Comm Services — the MSFT/META/AAPL/AMZN complex — and out of cyclicals: pre-earnings positioning, not broad risk-on. directional_tradable= TRUE (all four lane cohorts non-empty pre-gate) ·s1_standdown= FALSE- Bottom line: ZERO new starters. One held short maintained (PLNT, HOLD to 08-03), one close executed today (ORCL, +10.0% blended short-excess). One watch-only S4 name (HOG).
Preflight was not clean on first run — the truth-set parquets ended 07-27, one session behind the trade date. All three were rebuilt (build_prices → build_returns → build_features) and preflight re-run clear before any lane read them. This is the failure mode that produced an inverted regime on 07-24.
The AI-capex repricing broadened into semicap + memory — and the index is hiding it#
This is the most important thing on the tape tonight, and no directional lane can see it.
| Ticker | Close | ret5 | ret10 | pct_52w_range |
|---|---|---|---|---|
| SKHY (SK hynix ADR) | 130.17 | −24.3% | −32.9% | 2.8% — new 52w low today |
| LRCX | 269.61 | −16.3% | −22.1% | 51.4% |
| MRVL | 174.47 | −16.1% | −21.6% | — |
| AMAT | 476.46 | −15.6% | −20.0% | 55.0% |
| WDC | 463.51 | −15.5% | −17.7% | — |
| KLAC | 190.80 | −12.3% | −17.2% | — |
| AMD | 454.62 | −16.5% | −17.1% | — |
| MU | 820.53 | −15.5% | −16.5% | 62.3% |
| STX | 747.30 | −16.2% | −14.9% | — |
| ARM (reports Wed AH) | 244.74 | −15.5% | −13.0% | — |
| QCOM (reports Wed AH) | 162.88 | −6.1% | −8.6% | — |
| NVDA | 197.01 | −5.0% | −7.0% | — |
| TSM | 392.31 | −7.6% | −6.7% | — |
| ORCL (closed today) | 119.96 | −5.6% | −6.2% | 2.4% |
| AVGO | 380.91 | −1.5% | −2.1% | — |
| SPY | 740.86 | −1.0% | −1.5% | 86.2% |
Verified in data/prices.parquet and re-verified against the Yahoo chart API.
- The index is masking a violent rotation. SPY −1.5% over 10 days while semicap equipment and memory are −15% to −22% and SK hynix is −33%. Same AI-capex repricing logged in the 2026-W30 weekly review, now accelerated and broadened from the hyperscaler/software leg (ORCL) into the equipment and memory supply chain.
- The 52-week-range factor is structurally blind to it. LRCX/AMAT/MU are 35–38% off their highs yet still sit at 51–62% of their 52-week range, because they had roughly tripled into June. They are not MOM_SHORT candidates and will not be at these levels — they are control-cohort names. This is precisely why tonight's 10–90% control bucket printed −0.90% mean excess: the damage is in the middle of the distribution, where no directional lane looks. Observation for the journal, not a proposed signal.
Why it matters in the next 48 hours: MSFT and META — the two largest capex guiders — report Wed AH; ARM and QCOM report Wed AH into a −13.0% and −8.6% ten-day tape; AAPL/AMZN Thu AH. All stacked on FOMC Wed 2pm into fully-negative dealer gamma. Wednesday and Thursday's capex commentary is the direct catalyst that either extends this repricing or reverses it violently. That is the principal reason nothing new was sized tonight.
The 07-27 cohort squeeze REVERSED — the guard was correctly silent#
Last night flagged a violent one-day squeeze in the pct_52w_range 0–10% cohort (the MOM_SHORT target) that the index-level s1_standdown guard is structurally blind to. It did not persist. Two independent recomputes (orchestrator + regime-classifier, differing universe hygiene, both spot-checked against the Yahoo chart API):
Cohort (pct_52w_range) |
07-27 (squeeze day) | 07-28 (today) |
|---|---|---|
| 0–10% (MOM_SHORT target) | +2.35% to +3.68% mean · 80–89% positive | −0.40% to −1.15% mean · 33–49% positive |
| 10–90% (control) | +0.66% mean | −0.78% to −0.90% mean |
| 90–100% (MOM_LONG target) | +1.22% mean | +0.59% to +1.48% mean · 64–79% positive |
The 52-week-range factor's normal sign reasserted today (near-high beats near-low). The inversion was a single session that round-tripped. Caveat: some of the 0–10% dispersion is earnings-idiosyncratic given the print calendar (BKNG +6.7%, POWL −8.3%, FRMI −13.1%), so this is one day of reversal against one day of squeeze — direction unambiguous, n thin. The proposed cohort-arm to s1_standdown remains a /calibration-audit proposal, not a live patch (a gate change requires the regression gate).
Corrections to the 07-27 record#
- The 07-27 flag listed SKHY among "all ETPs" in the momentum lane's failed exclusion. That was wrong. SKHY is SK hynix Inc. ADR, $1.11T market cap,
issue_type = ADR. The ETP is SKUU (GraniteShares 2x Long SK Hynix Daily ETF). TIP, AAPD and SKUU were correctly identified; SKHY was not. Tonight's momentum lane classified it correctly and its ETP filter verified as biting (1,874 → 1,451 names, 423 excluded, zero ETPs in output). pct_52w_rangeis stored on a 0–1 scale and legitimately exceeds 1.0 for names trading above their trailing 52-week high. The 07-27 "1.1341 mislabelled ratio" flag was most likely a misread of a valid value.
Directional Book (excess-scored)#
No new positions. One held short.
| Ticker | Lane | Dir | Horizon | validated_excess | regime_fit | invalidation | gates |
|---|---|---|---|---|---|---|---|
| PLNT | OI_FADE | short | h10 → 2026-08-03 | +0.70% mean / +1.21% median (n=906) | 0.6 | Exit at the earlier of (a) 2026-08-03 close (h10, direct trading-day count) or (b) first genuinely negative call_net print. Plus a new discretionary stop: close > $59.50, 2-day confirm |
liquidity PASS ($55.94, $94.4M ADV) · cluster PASS (sole position) · fundamentals CAUTION · ER 08-06 falls after the exit |
New tonight — a discretionary stop was added at $59.50 (2-day confirm), ~6.4% above spot. This is a supplement to the pre-registered exit rule, not a replacement, and it is explicitly discretionary: it bounds event-window tail risk across FOMC + five Mag7 prints rather than expressing any measured edge. Flagged plainly because the repo's discipline is pre-registered exits, and a hand-added stop is the kind of thing that should never pass unremarked. Next run must check whether it was hit before re-deriving any stop, and must not treat it as vestigial if PLNT has moved meaningfully in the position's favor.
PLNT — HOLD the remaining 4 sessions#
Marked at $55.94 vs $55.20 entry (07-20): −1.51% short-excess. The position is losing, and the honest read is that the evidence is genuinely mixed. It holds anyway, because the pre-registered exit rule governs and neither leg has fired:
Cut condition NOT met — the build re-accelerated. Tonight's OI print: call +1,197 / put +21 = net +1,176, the largest single-day call build of the entire 5-day window (trajectory +395 / −4 / +180 / +14 / +1,176). persistence_ratio rose to 0.668, still under the 0.85 single-block disqualifier. The only near-zero print (−4 on 07-23) was previously established as a put-driven wash, not a call unwind.
The pre-registered analyst soft-invalidation has NOT fired. This clause had been carried unresolved for six sessions ("PLNT is one confirmed analyst PT raise away… if landed, CUT"). Full record since the 07-20 Canaccord $80→$82 move: Deutsche Bank CUT $61→$55 (07-22); Wells Fargo flat $65 reiteration (07-20); William Blair flat Hold reiteration (07-27). Zero additional raises, zero upgrades to Buy. The marginal analyst tape tilts bearish. Clause resolved, not triggered.
Corroborating the short: PLNT's dark-pool block-tier prints are overwhelmingly sell-side — 07-27 buy_ratio 0.16 ($29.4M, 9 trades), 07-28 buy_ratio 0.00 ($10.4M, 187,690 shares, zero buy volume) — on ~1.35× normal share volume. That is institutional distribution into a +3.17% up day. Separately, the +1,197 call-OI build came with negative net call premium (−$45,693): the calls are being written, not bought. Caveat: DP one-sidedness is the S2 factor and measures only +0.06% excess standalone — this is corroboration for an existing position, not independent evidence.
Cutting against the short: a clean 4/4 EPS beat streak, net (noisy) insider buying, rev +14.4% / EPS +31.4% YoY. And today's +3.17% is unexplained — no dated name-specific catalyst was found — landing on the same session as the record call build. That pattern fits early positioning into the 08-06 print, which falls after our 08-03 exit. Inference, not evidence, but it is the main risk carried into the final four sessions. Fundamentals accordingly downgraded CONFIRM → CAUTION.
What would change this: a genuinely negative call_net print (cover immediately), or a fresh Buy upgrade / 2nd–3rd PT raise (cover). Otherwise the position closes at the 2026-08-03 close regardless of mark. Holding past h10 is unvalidated drift — the lesson ORCL and PRMB both taught this month.
Exit-date note: hz_end('2026-07-20', 10) returns 08-04 because it rounds outward — correct fail-closed behavior for earnings-gate windows, wrong for a precise exit. Direct trading-day counting gives 08-03 (6 observed sessions + 4 projected, no US market holiday intervenes). 08-03 governs.
Vol Book (non-directional) — advisory throughout#
- 0DTE-VRP, Wed 07-29 (FOMC): STAND ASIDE.
zerodte_setup.pyreturnsGO_PREMIUM_SELL_INTRADAYfor both SPY and QQQ — overridden. The validated stack was never conditioned on a scheduled macro binary. Independently corroborated by the raw data: SPY front-end IV ratio 1.33× and QQQ 1.42× are both in backwardation, which lights one of the model's own two hard stand-aside gates. Two independent reasons to sit out, not one. - 0DTE 07-30+: cannot be scored until Wednesday's post-FOMC export. FOMC resolves before Thursday's cash session, so Thursday is not itself sitting on a live binary — re-score fresh on its own merits. The real hazard is Friday 07-31's open, which gaps in MSFT/META/ARM (Wed AH) and AAPL/AMZN (Thu AH) together.
- Earnings IV-crush: only PLNT (08-06), MSFT and FSLR clear all three legs (IV-rank ≥80 + genuine backwardation + positive VRP). META, AAPL and AMZN fail on VRP (−0.006 / −0.040 / +0.090 with iv-pctile 77) — gross option richness that does not survive netting against realized vol, exactly the Vilkov failure mode. Do not size those three. ARM has the fattest implied move (11.0%) and steepest backwardation (1.93×) but its IV-rank disagrees across sources (82.7 vs 59.5) — watch-only.
- PLNT vol structure stands (implied move 10.9% / ≈$6.10, VRP +0.265, IV-rank 87.5, backwardation) and does not overlap the directional short, which exits 08-03 before the 08-06 print. If taken it must be booked and P&L-tracked as its own structure, never as a modification of the short.
- Nothing here has cleared the standing promotion bar (vol-shock day sampled + tail-aware net expectancy). Thin book or none is the correct posture into this event stack.
Watch / Stood-down#
- HOG (S4 sentiment-contrarian, long, h5–10) — WATCH, not sized. The first S4 name this month to survive scrutiny: vol_PCR 5.37 (95th pctl 3.76) on a real denominator of 307 calls (clears the new ≥75 floor that killed PFGC's 16-call artifact), OI-PCR 1.183 = 84.1st percentile, liquidity PASS ($68.9M ADV), earnings clear (next print 11-03), no prior verdict. Tape is a textbook setup: four-session post-earnings capitulation from $28.45 → $24.79 (−12.9%) despite a raised 07-23 outlook, then today a reversal bar (open 24.53, low 24.45, close 25.29, +1.78% excess, closing near the high).
Held at watch because the mechanism is contradicted.
net_put_premium= −$11,791 (puts net SOLD) andnet_call_premium= +$4,669 (calls net bought) — netprem +$16,460, flow reads bullish. S4 fades a crowd that bought puts; on a premium-weighted basis that crowd does not exist here. The whole options book is$66K of premium across 1,956 contracts ($0.34/contract) — too thin to call a crowd. Against a +0.44% advisory-only prior, over a horizon running through FOMC + five Mag7 prints, on a rate/consumer-sensitive name in a vol-amplifying negative-gamma tape, that is not enough. Re-look post-FOMC. Note: premium-signed PCR is not part of the validated S4 spec. It was weighed as evidence quality, not applied as a new gate — doing so would be an unvalidated patch (invariant #5). Logged as a/calibration-auditproposal. Nofundamentals-gatewas run on HOG. That is correct per gate 4 — the gate is spawned only on names about to be sized, and HOG is not. It must be run before HOG can ever move past watch. - SKHY (SK hynix ADR) — MOM_SHORT cohort, watch-only capped. $130.17, new 52-week low today, ret5 −24.3% / ret10 −32.9%, $6.55B ADV, no scheduled earnings in h10. Four consecutive down days closing near the low each session; notably it fell 7.5% on 07-27 while the near-52w-low cohort squeezed +2.4–3.7%, i.e. genuine distress rather than cohort noise. Not sized: MOM_SHORT's prior is −0.03% mean and shorting after −33% in 10 sessions is the momentum-crash tail the lane's negative mean describes.
- FRVO (Fervo Energy) — MOM_SHORT cohort, watch-only capped. $21.12, ret5 −21.7%, pct_52w_range 1.6%, $70.3M ADV, ER 09-28. Carries a prior CAUTION (07-07) and sits outside the 783-name truth-set panel (data gap, unverified).
- MOM_LONG — 298 names pass screening; basket/tail-only by construction (+0.18% mean, median −1.06%), never sized per-name. No basket initiated into the event stack.
- S2 liquidity-reversion — 24 names cleared all gates, nothing advisable. Prior is +0.06% (≈zero) and a 3–5 day horizon opened today runs straight through FOMC and every mega-cap print.
- OI_FADE near-misses — HRI (rel_build 1.70, persistence 0.57, catalyst LIVE), TPG (1.49, 0.43), AME (0.53, 0.41) all cut solely on earnings inside h10. The dense print calendar is what emptied this lane, not signal quality.
Risk#
- Event stack: FOMC Wed 07-29 2pm ET (no SEP) · MSFT/META/ARM/QCOM Wed AH · AAPL/AMZN Thu AH · NFP Fri 08-07 · CPI Wed 08-12. Five of the Mag7 report inside 48 hours, stacked on FOMC, into fully-negative dealer gamma on both indices.
- Correlation clusters: none active — PLNT is the sole live position, so the cluster gate is trivially satisfied.
- Tail caps applied: MOM_SHORT watch-only cap held (SKHY, FRVO unsized); S4 advisory cap held (HOG unsized); vol book capped by the unsampled left tail into a mega-cap AI-capex print week.
- Held-book reconciliation: swept every July envelope for names with a real size (29 found) and cross-checked against the audit resolvers. All 29 appear in
analyses/audit/2026-07-25/resolved_calls.json. The PRMB failure was therefore specifically a live held-book management gap (it drifted 4 sessions past its h10 unmanaged), not a calibration-grading gap — the audit catches grading; the nightly reconciliation is what missed the exit. Tonight's book is fully reconciled: PLNT live, ORCL closed. - Regression gate: not required — no lane logic or threshold was changed tonight;
git statusclean. Any proposed fix (cohort arm ons1_standdown, premium-signed PCR corroboration for S4) must clearpython3 scripts/retro_harness.py --allagainst the4ec89dcbaseline before shipping.
Resolved for /calibration-audit#
ORCL (MOM_SHORT short) — CLOSED 2026-07-28 at the open, $118.18. Entry 07-13 $131.54.
| Leg | Exit | ORCL | SPY | Short excess |
|---|---|---|---|---|
| First half | 07-24 @ $114.99 | −12.58% | −1.37% | +11.21% |
| Second half | 07-28 open @ $118.18 | −10.16% | −1.33% | +8.82% |
| Blended | ≈+10.02% |
Closed on h10 horizon expiry, not a stop — the $123.50 stop never triggered. Worth recording: ORCL made a new 52-week low intraday today ($114.50) and then reversed to close $119.96. On a short, the $118.18 open fill therefore captured more than a close fill would have — +8.82% vs +7.69% short-excess (+1.13pp), or +10.16% vs +8.80% gross. (The sizer's envelope originally asserted the opposite sign in calls[ORCL].invalidation; corrected in place, and its watchlist_write_back had it right.) A single +10% short-excess win against a lane whose measured mean is −0.03% is a tail draw, not a validation of MOM_SHORT — it should be logged as one observation in a fat-tailed distribution and must not be used to argue for lifting the watch-only cap.