No orders. One decision, made once: carry the powder 24 hours into a double binary.

Tape. Dow and S&P at record highs, Nasdaq up, on Strait-of-Hormuz deal optimism — Bessent says a deal to restore "freedom of movement" could land Tuesday or Wednesday, and WTI fell 5%. Leadership is AI/industrials: PLTR +20% (record 149% U.S. commercial revenue growth, FY guide raised to $8.16B), CAT +12% (first $20B sales quarter, data-center demand). Counterweight nobody is pricing today: the ISM manufacturing survey put inflation worries "worse than pandemic era", some Fed officials voted to HIKE, and July jobs lands Friday 08-07.

Attribution first, in L013's order — and today it runs the OTHER way from yesterday. Strip the cash: invested capital was $595.12 of $1,026.72 (42.4% cash), and on that invested capital the book did -0.55% today (CMG -0.96, KMX +0.30, TGT -1.94, XYZ +1.04, ZBH -1.71 = -$3.27) against an S&P up [path]0.5%. So this is a genuine [path]1.05pp lag on capital actually at risk — cash drag is not the excuse today, because cash drag makes the headline worse than the picks and here the picks themselves are red while the index is green. Yesterday cash masked good selection; today there is nothing to hide behind. Then L006: is it rotation? Today's leading style box is AI/industrials, and my book holds zero of it — four consumer-discretionary names and one medtech. Consumer-disc and medtech are simply not where the money went on an oil-relief melt-up. So the lag is mostly style-box beta running against me, not five theses breaking — which is an explanation, not a licence to sell. I own the fact that being in one box means some days the box is wrong.

Exit flags: none. Peaks and trailing levels: KMX +15.46% peak / stop $56.92, TGT +11.53% peak / stop $142.17, XYZ and ZBH and CMG below the +10% activation, so their binding stop is the fixed -10%. Per L009/L012 that is the whole of the position review — five theses re-argued against a price that moved 40 cents is the failure mode, not diligence. Nothing was re-derived. The one new FACT that earned an entry: tomorrow's two prints are confirmed with their Street numbers, appended to the ZBH and XYZ files.

Why no sixth name, stated as a cost and not a shrug. Every candidate today's flow hands me is the SAME entry template the whole book is already built on — a fresh buy-rated PT above spot. Under L004 that is one bet restated five times, not five convictions, and a sixth clone would make it six. The one macro idea with real legs — falling crude — I am already long by accident: cheaper gas is a direct tailwind to CMG/TGT/KMX, so buying an airline or another consumer name doubles the Hormuz bet rather than diversifying it. And chasing today's actual leaders is off the table on arithmetic as much as discipline: CAT is >$600 a share against a $200 per-trade cap, and PLTR after +20% is above every target I can find.

The premium, named out loud BEFORE the print (L013's second half). Holding $434.86 — 42.4% of the account — cost me roughly 0.21pp of today's index move, and it will cost about that again on every up day this week. I am paying it on purpose, and here is what I am buying: two scheduled binaries inside 24 hours — ZBH reports 08-05 BEFORE the open (EPS $2.01, rev $2.13B, FY26 organic guide still maintained at 1-3%) and XYZ reports 08-05 AFTER the close (EPS $0.48, Square GPV $72.16B, against the FY26 gross-profit outlook raised to $12.33B). ZBH's lands before any tick I can run, so per L008 its stop is notional: a bad guide gaps through $86.83, it does not tag it. L005 is unambiguous about this setup — do not spend the powder the afternoon before a coin-flip; a flush buys better entries in names I already understand than any name I could research today.

So the cash is not open-ended. Dated triggers, written now so a good outcome cannot make it look free and a bad one cannot make it look like the mistake:

1. 08-05 pre-open — if ZBH RAISES the FY26 organic guide above 3% with price still meaningfully below $115, the add gate in its file is met and I add. If the guide is merely reaffirmed at 1-3%, that is the SYK setup and I do nothing.
2. 08-05 post-close / 08-06 — if XYZ's gross profit forces the $12.33B FY guide UP with margin expansion, I add. If GP stalls, the thesis wall is cracked and it becomes an exit conversation, not an add.
3. If both prints disappoint and no stop fires, the powder gets redeployed by the 08-07 DEEP into a genuinely NEW sleeve — not a sixth consumer-discretionary clone. Carrying 42% cash past Friday's jobs report with no thesis for it would stop being patience and start being avoidance.

Brief persisted: logs/research/20260804T134349-signals.md. Deployed today $0.00 of $500 headroom, deliberately.

Verdict: no orders. Five positions held, zero exit flags, no thesis broken — and the

one lesson that could have earned a re-derivation was tested and came back negative.

The tape

S&P 500 and Dow at record highs. Two engines: Strait-of-Hormuz de-escalation hopes (Bessent says a deal to reopen it may land Tuesday or Wednesday; WTI -5%) and a monster earnings slate. Palantir +16-18% on U.S. commercial revenue +90% y/y; Caterpillar +9-12% after topping $20B revenue for the first time on AI-datacenter power demand; storage and semis broadly bid (MU, SNDK, WDC, STX, INTC, ON). Signal brief: logs/research/20260804T142208-signals.md — 45 tickers, PLTR the runaway top signal across all five source classes.

I own none of that, on purpose. My book is consumer/auto/fintech/medtech. Today's leadership is the AI build-out.

The one re-derivation I let myself have (L011)

CMG is my only meaningfully red name (-2.98%) while SPY is +2.16% since I re-entered on 07-30. That is the exact shape L011 says earns a falsification test of my committed metric. So I ran it once, on settled closes, against the true comp set (MCD/QSR/DRI) — not on an intraday print (L009). Result: 07-31 CMG lagged the comp basket by -2.96pp; 08-03 CMG LED it by +1.36pp, closing green while MCD, QSR and the sleeve closed red. Two sessions, opposite signs. L011 requires 3+ same-sign sessions. Trigger not met — thesis stands, committed metric (comps + the FY26 comp-guide raise) untouched. Full table in logs/tickers/CMG.md. MCD's "U.S. business is falling short" today is a QSR peer negative and stays logged as one; promoting it into my CMG thesis is precisely the L010 error of grading the wrong exam.

Why I did not deploy the $234.86 I could have

Deployable = $434.86 cash - $200 buffer. Every route into today's leadership fails a gate I wrote myself:

• L005 — AMD and SpaceX both report after the bell tonight. Buying a semi this afternoon is pre-positioning into the sector's coin flip with worse information than the market. L005's edge is carrying dry powder INTO the event, not spending it the hour before. That single-handedly kills ON (+7%), MU, INTC.
• Chasing. W +25%, PLTR +16%, CAT +9%, BLZE +46%. My entry structure (L003) is buying well below a fresh buy-rated PT on relative strength — not paying up for a gap that already happened. There is no version of today's movers that clears that gate.
• L001 — affordability is not conviction. At a $200 cap and whole shares, the on-theme leaders are mostly unreachable, and the reachable ones are reachable because they are second-best. I am not going to let "PLTR fits in one share" masquerade as an idea.
• L004 — I already run five names off a repeated template. A sixth clone adds correlation, not diversification. SOFI would sit on top of XYZ (fintech); Stellantis' strong North America print sits on top of KMX (auto).

The price of that cash, named BEFORE the print (L013)

L013 says a deliberate cash position has a price and I have to say it out loud in advance, so that a good outcome cannot make the cash look free afterwards. So: I am holding $434.86, 42.3% of a $1,027.75 book, into a record-high tape. Every 1% the index adds today costs me [path]0.42pp of book return I will not get back. I am paying that premium deliberately, to have powder if AMD or SpaceX disappoints tonight and the Hormuz-relief trade unwinds. If AMD beats and the tape melts up, the cash was expensive and I do not get to retroactively call it prudence. That is the bet, on the record, before the print.

Consumer read (colour, not a trade)

Split signal worth watching: Wayfair posted its best U.S. growth since 2020 (+8.7% U.S. revenue) while McDonald's U.S. falls short and the Fed notes consumers "flash warning signs." Goods holding, low-end services softening. Logged to logs/tickers/TGT.md. It moves nothing today — TGT's ratchet, not my macro opinion, is what takes me out.

Book state

XYZ is 0.67pp of peak gain from arming its trailing stop. Deployed today: $0.00 of $500 headroom.

DEEP sweep run in full; no order placed, and one new fact recorded.

The sweep. 390 docs (RSS macro/markets/earnings + HackerNews) plus WebSearch across twitter/reddit/stocktwits/news → logs/research/20260804T145734-signals.md, 29 ticker signals. The tape is the 14:22Z tape with bigger numbers: records on Strait-of-Hormuz de-escalation (WTI -5%), PLTR +16-26% on AI demand, CAT +9-11% on an $8.17-vs-$6.20 beat, QQQ +2.1% on a semis rebound. AMD and SpaceX report after the bell tonight.

Why nothing was bought. Every name at the top of the brief — PLTR, CAT, W, BLZE (+46%) — is already up 9-46% today. Buying them here is a post-gap chase with the catalyst fully in the price and no valuation gate left to clear, which is the precise structure L003/L004 exist to refuse. The two names with a live catalyst (AMD, SPCX) print in six hours, and L005 is explicit that pre-positioning into a scheduled binary is guessing the print with worse information than the market. So the morning's pre-committed stance stands unchanged: $434.86 held into ZBH 08-05 pre-open and XYZ 08-05 post-close, at the cash premium already named on the record (L013 — that premium was charged again today, on a green tape, which is exactly when it is supposed to hurt).

The one new fact — and its limits. The sweep surfaced McDonald's saying its US business is falling short. That is a fact, not a price, and it lands on US restaurant comps — the variable CMG's raised FY26 guide rests on — so it earned a note on the ticker. It does not fire CMG's 07-30 pre-commitment, which is narrow by design (Chipotle inside a cyclospora outbreak, or Q3 commentary showing the late-July dent was a trend). A rival's soft quarter is neither, and it cuts both ways: CMG raised its comp guide in the same window MCD is missing in. L011 count stays 1 of 3 — it advances only on settled closes against the restaurant comp set, never on a live session (L009). A "$CMG hits 52-week low" item in the capture log was checked and discarded as stale June tape; CMG is $37.17 against a $44.54 52-week high.

Book. Five positions, zero exit flags. CMG -2.52%, KMX +11.93% (trailing $56.92), TGT +10.21% (trailing $142.17), XYZ +9.89%, ZBH -1.52%. Deployed today $0.00 of $500.

DEEP 15:31Z — the first order in three sessions. BOUGHT 3 RIVN @ $15.767 ($47.30). Fresh brief written to logs/research/20260804T153150-signals.md (35 ticker signals from 360 RSS docs + 30 HN + WebSearch over QUERIES). Flag check first: five held positions, zero exit flags, no sells.

What the sweep found, and what I refused

Same tape as the 14:22Z and 14:57Z sweeps, one leg further along: S&P 500 through its June closing record on a fourth straight up day, powered by Strait-of-Hormuz de-escalation (Bessent says a deal could land Tue/Wed, WTI -5%) and a heavy earnings slate — PLTR +27%, CAT +12% on its first-ever $20B revenue quarter, ZBRA +18.6%, IT +16.3%. AMD and SPCX report tonight.

I bought none of it, and the reason is structural rather than a view on the names. Every leader is +9% to +27% on the day. Buying any of them is a post-gap chase — the exact structure L003/L004 exist to refuse — and none clears the valuation gate. The ranked table in the brief is a buzz measurement, not a shopping list.

What I did buy, and why it is not the same thing

RIVN, 3 shares, a starter. This is a pre-registered trigger firing on its 5th evaluation, not a new idea generated by today's tape. I passed on Rivian four times, most recently 07-28 at $16.245, and each time I wrote down the same specific condition: the upgrade pop fully given back after a survivable Thursday print. Both halves are now true.

The print (07-30) was better than survivable. Revenue $1.658B vs $1.58B consensus (+27% y/y); EPS -$0.63 vs -$0.82; record $179M gross profit; capex guide cut $250M to $1.7-1.8B on project efficiencies; adjusted loss narrowed to $1.8-2.0B — and the 65,000-70,000 delivery target held. Spending less and losing less without cutting deliveries is the cash-burn thesis improving on the only axis that matters for a pre-profit name. That fundamental leg is precisely what I said was missing every previous time I passed.

The price cooperated too. $15.77 against the $16.245 where I declined it and the $15.84 close before the Piper Sandler upgrade. The entire pop ($16.53 high, PT $20 on robotaxi) is gone — I am buying [path]3% below my own pass price. I am explicitly not buying the robotaxi leg; that is narrative and prices a market that does not exist yet.

The symmetry point that made this a buy rather than another pass: a pre-commitment can be broken in two directions. L011 is about the trigger you named being wrong. This is the other case — the trigger I named was right and fired, and declining it because the position feels uncomfortable would be rewriting my own rule in the restrictive direction. Four passes earn the right to act when the condition finally arrives; they do not earn a fifth pass for free.

Guardrails and the cash premium

Sizing is the risk control (L004/L008): 3 shares, [path]4.6% of equity, because this is the highest-beta name I would own and one good print is not a conviction. It also fixes real concentration — the book was consumer-cyclical heavy (CMG restaurants, TGT retail, KMX used autos) plus XYZ fintech and ZBH medtech.

This does not break the L005 powder commitment, and I want that on the record before tomorrow rather than after. Both binaries are confirmed inside 24h: ZBH reports 08-05 6:30am ET pre-open (consensus $2.13B / $2.01, four straight beats averaging +4.92%) and XYZ reports 08-05 post-close. I still carry [path]$387 cash, [path]$187 usable above the $200 buffer, into both. RIVN's binary is behind it — I acted a week after the number landed, which is the sequencing L005 tells me to wait for, not a coin-flip I pre-positioned into. Per L013 the cash premium stays named in advance: roughly 0.9pp of drag on an up day like this one, charged whether or not the prints cooperate.

Logged, deliberately not acted on

MCD said its U.S. business is "falling short" and replaced its U.S. president; CMG's own Q2 had comps back positive (+2.2%) but margin -220bps to 25.2% and transactions +1.0%. That is a QSR peer datapoint against the demand half of the CMG thesis — it is not a break of my committed metric (the raised FY26 comp guide, which stands). Already recorded twice today in CMG's memory; per L012 it does not earn a third narration. L011 count stays 1 of 3 — it advances only on settled closes against MCD/QSR/DRI/YUM, never on an intraday print.

State after the order

• CMG -2.70% (fixed stop $34.32) | KMX +10.97% (trailing $56.92) | TGT +10.85% (trailing $142.17) | XYZ +9.95% (trailing $80.58) | ZBH -1.41% (fixed stop $86.83) | RIVN new @ $15.767 (fixed stop [path]$14.19)
• Deployed today: $47.34 of $500. Six positions. No exit flags.
• Next named catalysts: ZBH pre-open and XYZ post-close, both 08-05.

DEEP sweep — brief written, no order. Fourth sweep today, same conclusion.

Signals: RSS macro/markets/earnings (360) + HN (30) + WebSearch over QUERIES → logs/research/20260804T160921-signals.md (35 ranked tickers). The tape is a record-high melt-up — S&P 500 at an all-time high, Dow +900, WTI -5% on Bessent's "last chance" Hormuz framing — led by PLTR +27%, ZBRA +18.6%, IT +16.3%, CAT and Wayfair +25%. Not buyable, structurally: every leader is +9% to +27% ON THE DAY, so any entry is a post-gap chase (L003/L004). That is the same reason I refused PLTR and CAT two hours ago and bought RIVN on weakness instead — the ranking is a buzz measurement, not a shopping list.

Positions: six, zero exit flags. CMG -1.64% (fixed $34.32) | KMX +10.67% (trailing $56.92) | TGT +11.25% (trailing $142.17) | XYZ +9.89% (trailing $80.58) | ZBH -1.52% (fixed $86.83) | RIVN +0.79% (fixed $14.19). No peak advanced, no ratchet re-levelled.

Work done this tick that wasn't a re-narration (L012):

• RIVN Current Thesis block written. It had sat empty through four passes and one buy. Now it names what I actually own — the 07-30 capex cut ($250M off FY26 guide) with the 65-70k delivery target intact — and, as importantly, what I do NOT own: the robotaxi leg and its $20 PT. Committed metric per L010/L011 is the capex/adjusted-loss guide against a held delivery target; falsified by a Q3 guide that walks the cut back or funds it by cutting deliveries, not by EV tape noise.
• CMG peer datapoint logged, second sweep running. MCD said its U.S. business is falling short and named a new U.S. president. Real fact about restaurant demand, and per L009 it earns attention — but per L011 the falsifier is CMG's own comp/traffic print, not McDonald's. One peer is a datapoint; two pointing the same way would be worth re-deriving on. Logged, not acted on.

Capital: no further deployment today, pre-committed. $387.56 cash, [path]$187 usable above the $200 buffer, held into ZBH tomorrow 06:30 ET pre-open and XYZ tomorrow post-close (L005 — do not pre-position around a scheduled binary; the -10% stops already cap the downside if either breaks bad). Per L013 the premium is named in advance, not rationalised after: carrying [path]38% cash on a record-high day costs roughly 0.5-0.9pp of relative performance, and I am paying it on purpose so tomorrow's outcome can't retroactively make the cash look free or look like the mistake.

Deployed today $47.34 of $500 — headroom $452.66.

DEEP sweep #5 of the day — brief written, no order, no change (L012).

Ran the full sweep because the 30-minute DEEP cadence was due: RSS macro/markets/earnings (360 docs) + HackerNews (30) + WebSearch over QUERIES, ranked and persisted to logs/research/20260804T164338-signals.md (48 ticker signals). The sweep is the deliverable; the answer is the same one it has been since 13:43Z.

Nothing at the top of the tape is buyable and the reason is structural, not a mood: every leader is up 6-27% ON THE DAY — PLTR +27%, W +25%, ZBRA +18.6%, IT +16.3%, CAT +6.4%, with SMH +5.06% and QQQ +2.71% carrying the AI/semis complex into AMD after the bell. Buying any of them here is the post-gap chase L003/L004 exist to refuse, and the one I already refused this morning in PLTR and CAT in order to buy RIVN on weakness instead.

Book: six positions, zero exit flags, no peak advanced, no ratchet re-levelled. CMG -1.78% (fixed $34.32) | KMX +11.11% (trailing $56.92) | TGT +11.34% (trailing $142.17) | XYZ +9.81% (trailing $80.58) | ZBH -0.77% (fixed $86.83) | RIVN +0.82% (fixed $14.19). I re-pulled per-symbol news across all six: the only fresh rows were keyword noise — a "price target" harvest matching TGT. No print, no guide change, no downgrade against any live thesis. Thirty-one minutes of tape is a new price, not a new fact (L009).

Capital: unchanged and pre-committed. $387.56 cash, [path]$187 usable above the $200 buffer, held into two binaries inside 24 hours — ZBH tomorrow 06:30 ET pre-open, XYZ tomorrow post-close. Both stops are notional against those prints (L008: a pre-/post-close reporter gaps THROUGH the level, not down to it), which is exactly why both are 1-share clips. Per L013 the premium stays named in advance: on a record-high session [path]38% cash costs roughly 0.5-0.9pp of relative performance, paid deliberately so tomorrow cannot retroactively make the cash look free or look like the mistake. Deployed today $47.34 of $500; headroom $452.66.

FAST — CMG exited on a broken thesis (judgment, not a stop)

One order this tick: SOLD 4 CMG @ $34.9326. Realized [path]-8.39% vs the $38.1299 cost ([path]-$12.79 on [path]$152.52). Every other position: flag clean, unchanged, no action.

What happened

CMG traded a 40-cent band around $37.40-37.50 all session, then went vertical downward in two five-minute bars: 16:45Z — 801,876 shares ($37.47 -> $36.58), 16:50Z — 3,948,457 shares ($36.59 -> $34.875, low $34.27). The morning baseline was [path]30-45k per five minutes. That is roughly 4.75 million shares in ten minutes, a >100x volume spike.

I could not identify the catalyst before deciding. WebSearch returned only stale pre-08-04 coverage (the 07-29 earnings beat, the 07-15/07-22 cyclospora scares) and Benzinga's quote page was still serving $37.45. Intraday search indexes lag by hours.

Why this was a thesis break and not a scare

It is perfectly idiosyncratic. On a SPY +1.66% day the entire true restaurant comp set was green or flat — MCD +0.85%, QSR +1.00%, YUM +0.73%, DRI -0.95% — and CMG alone fell 7.5%.

That specific pattern matters because of what I wrote into this thesis yesterday. On 08-03 I recorded the peer split as confirming the "Chipotle is not involved" leg, with this reasoning: YUM — Taco Bell's parent, the implicated chain — was the day's worst restaurant close at -2.88% while CMG held green, i.e. the market was pricing the outbreak into the implicated name and not into Chipotle.

Today that identical test returns the exact opposite answer. YUM is green; CMG alone is collapsing. If the peer split was evidence FOR the not-involved leg, its clean inversion is evidence AGAINST it. Keeping only the confirming direction of my own test is precisely the error L011 describes.

The deviation, stated plainly

My pre-commitment said: exit on the HEADLINE if the second outbreak traces to Chipotle or a Chipotle supplier. I exited on the tape instead. That is a real deviation and I am not going to dress it up.

My justification: that pre-commitment was written to stop me exiting on noise. It was not written to make me sit through a 100x-volume single-name collapse waiting for Google to index a story. Insisting on the headline here would have been L011's exact failure mode — holding because the committed procedure keeps passing while the tape screams the variable I am not modelling. The entry rested on two legs (the FY comp-guide raise to low-single-digit, and "not involved"), and in a record-size cyclospora season the plausible generators of a vertical 7.5% single-name move are the two things I pre-committed to exit on.

Why now rather than letting the stop work (L008)

The stop sat at [path]$34.32. The 16:50Z bar low was $34.27 — the stock had already traded through it. No flag fired only because the feed sampled between prints; the tick computed -8.76% off a $34.79 sample. L008's whole content is that a fast name gaps THROUGH a level rather than tagging it, so the real downside is the gap, not the stop distance. Sitting 1.2% above a stop on a tape moving 5% per ten minutes is not letting the stop do its job, it is hoping. The mechanical flag never fired; I did not wait for it, and I would make that same call again on this tape.

Execution went well at least: limit $34.70, filled $34.9326 on the bounce.

Named cost of being wrong

If this turns out to be an unrelated flush — a block print, an index event, a non-food-safety downgrade — then I sold a guide-raise long [path]9% below cost into a panic and paid [path]$13 for an information gap I could have waited out. The check tomorrow (08-05) is to identify the actual 12:45-12:50 ET catalyst and grade this exit against it explicitly, either way.

Rest of the book — flag clean, unchanged

• KMX +10.75%, trailing stop $56.92 · TGT +11.33%, trailing stop $142.17 · XYZ +9.94%, trailing stop $80.58 · ZBH -0.93% · RIVN +0.35%. No flags, no thesis news, no action (L012).

Cash after the sale [path]$527. Deployed today $47.34; headroom unchanged — this was an exit, not a redeployment, and I am not rushing the proceeds back out on the same session I took a loss.

DEEP — sweep #6, no order. The one new fact is a negative one.

No capital deployed. Five positions, zero exit flags, nothing re-levelled. This log exists because a DEEP sweep ran and because it produced exactly one genuinely new fact — not to re-argue five unchanged theses against a price that moved forty cents (L012).

The new fact: the CMG catalyst still does not exist in any index

Eighteen minutes after selling CMG I searched again. What came back:

• The top hit for "why is CMG plunging" is a finviz article dated July 24, 2025 — a year old, describing last year's cut of FY comp guidance to "approximately flat."
• A Smartkarma page quoting CMG at $30.54 against a $37.46 prior close. That is bad data, not a datapoint.

Neither is 2026-08-04. The 16:45-16:50Z [path]4.75M-share vertical remains uncatalogued.

I am recording this as a finding rather than quietly dropping it, because the failure mode here is specific and tempting: an unexplained move plus a plausible-looking old headline is exactly how a rationalisation gets written. A 2025 guide-cut story would have fit my exit narrative beautifully. It is also a year old. Declining to use it is the whole discipline.

The interim grade, stated before the answer arrives

CMG $34.665, -7.46% on the session — about 0.77% BELOW my $34.9326 fill. So eighteen minutes on, the exit is neither vindicated nor refuted. It is roughly a wash.

I want that on the record NOW, because tomorrow it will be unavailable. Once the catalyst is known the decision will look either obvious or stupid, and I will be reconstructing my own confidence from the outcome. The honest present-tense reading is: I sold into an information gap and the tape has not yet told me whether that was discipline or panic.

What is re-confirmed is the idiosyncrasy the exit rested on, re-measured this sweep: SPY +1.80%, XRT +1.21%, MCD +0.90%, YUM +0.83% — the entire restaurant comp set green — against CMG alone -7.5%. One name, one direction, no sector. That says the move is company-specific. It does not say what it is, and I am not going to pretend those are the same claim.

The 08-05 check is unchanged and explicitly NOT accelerated: identify the actual 12:45-12:50 ET catalyst on settled data and grade the exit against it, either way. No re-entry considered today — I do not recycle proceeds back into a name I exited on an information gap, in the same session, while still lacking the information.

The tape: sixth sweep, sixth time nothing at the top is buyable

RSS macro/markets/earnings (390 docs) + HackerNews + WebSearch over QUERIES. S&P and Dow at records, SPY +1.80%, QQQ +3.08%, IWM +1.84%, semis leading, oil soft on Bessent's Hormuz framing.

The leaderboard: PLTR +27% (Q2 blowout, FY raise, sovereign AI), AMRC +38.9%, ZBRA +18.6%, IT +16.3%, plus W, SNAP, CAT. Every single one is up 6-27% on the day. Buying any of them here is a post-gap chase — the structure L003/L004 exist to refuse, and the one I already refused this morning in PLTR and CAT in order to buy RIVN on weakness instead. AMD and SPCX report after the bell; I have no position and no edge in either. The brief's ranking is a buzz measurement, not a shopping list.

Book — flag check clean, one line each (L012)

• KMX $58.54, +10.81% · peak +15.46% · trailing $56.92
• TGT $149.705, +11.37% · peak +11.53% · trailing $142.17
• XYZ $84.475, +10.19% · peak +10.23% (fractionally new) · trailing $80.58, not re-levelled
• ZBH $95.805, -0.70% · ratchet off · fixed $86.83
• RIVN $15.865, +0.60% · ratchet off · fixed $14.19, priced as NOTIONAL (L008)

Stance: no further capital today — pre-committed since 15:31Z, and I am keeping it

Two book binaries land inside 24 hours. ZBH Q2 tomorrow 06:30 ET pre-open (confirmed this sweep; consensus $2.01 EPS on $2.13B revenue, against four straight beats — last quarter beat by 12.37%), and XYZ tomorrow after the close. Buying power $387.56, of which roughly $187 is usable above the $200 buffer, is held into both. L005: do not pre-position around a scheduled binary — that is guessing a print with worse information than the market, and the stops already cap the downside.

L013 — the premium, named in advance rather than rationalised after. On a record-high session with SPY +1.80%, carrying [path]38% cash costs on the order of 0.5-0.9pp of relative performance today. That is a real, chosen cost. Whichever way ZBH prints tomorrow: a good outcome does not make the cash free, and a bad one does not make it skill.

Deployed today $47.34 of $500; headroom $452.66. Brief persisted to logs/research/20260804T171924-signals.md (42 ticker signals).

DEEP — seventh sweep, no order.

Full sweep run (RSS macro/markets/earnings 360 docs + HackerNews 30 + WebSearch over QUERIES); brief persisted to logs/research/20260804T175315-signals.md, 34 ticker signals. Nothing in it changed a thesis, so this is one block and not a page (L012).

• Flag check clean: KMX +10.71% (stop $56.92) | TGT +11.26% ($142.17) | XYZ +10.03% ($80.58) | ZBH -0.83% (fixed $86.83) | RIVN +0.44% (fixed $14.19, NOTIONAL per L008). No peak advanced, no ratchet re-levelled, no ticker-memory append earned (L009).
• Tape leaders unbuyable for the seventh time and for the same structural reason: PLTR +27%, ZBRA +18.6%, AMD +7.8%, MU +7.7%, AVGO +6.1% are all already up 6-27% on the day on Hormuz-relief (WTI -5%, SPX/NDX/DJI at records). Post-gap chase = the thing L003/L004 refuse. I bought RIVN on weakness this morning instead of PLTR/CAT on strength.
• Stance unchanged and pre-committed since 15:31Z: no further capital today. $387.56 buying power ([path]$187 usable over the $200 buffer) carried into two book binaries inside 24h — ZBH Q2 tomorrow 06:30 ET pre-open, XYZ Q2 tomorrow post-close (L005).
• L013 premium, named before the print: [path]38% cash on a record-high session costs roughly 0.5-0.9pp of relative performance today. Deliberate. Tomorrow cannot make it free.

Deployed today $47.34 / $500; headroom $452.66. CMG exit grade stays scheduled for 08-05 on settled data.

DEEP sweep #8 — flag check clean, no new fact, no order. One line (L012).

Signals swept and persisted (logs/research/20260804T182536-signals.md): 360 RSS macro/markets/earnings docs + 30 HackerNews + WebSearch over QUERIES. Flag clean on all five — KMX +10.39% (trailing $56.92) | TGT +11.15% ($142.17) | XYZ +9.98% ($80.58) | ZBH -0.32% (fixed $86.83) | RIVN +0.03% (fixed $14.19, NOTIONAL per L008). No peak advanced, no ratchet re-levelled, no thesis condition broken. No order.

Nothing bought because nothing is buyable: every leader on the tape (PLTR +27%, ZBRA +18.6%, IT +16.3%, MRVL +14%, INTC +10%, MU +8%, AMD +7%) is already up on the day on Hormuz relief into record index highs — a post-gap chase L003/L004 exist to refuse. The refreshed ZBH preview (Street $2.01 EPS / $2.13B revenue, four straight beats, FY26 organic cc guide still MAINTAINED at 1-3%) restates the pre-committed wall rather than moving it, so it earns no re-derivation (L009) and no ticker-memory append (L012 — three near-identical ZBH DEEP entries today is already one too many).

Stance unchanged since 15:31Z: no further capital today, [path]$187 usable powder held into ZBH Q2 tomorrow 06:30 ET pre-open and XYZ Q2 tomorrow post-close (L005). L013 premium named in advance: [path]38% cash on a record-high session costs [path]0.5-0.9pp today, deliberately. Deployed today $47.34 / $500; headroom $452.66.

DEEP #9 — the CMG exit is graded, same session, and it holds up.

A full log is earned here because a FACT changed, not a price (L012). Eight sweeps today produced eight identical answers and I kept them to one line each. This one resolved the open question of the day.

What resolved. At 17:00Z I exited the 4-share CMG starter at $34.9326 on a vertical 100x-volume collapse whose catalyst I could not name, and I logged the deviation plainly: my pre-commitment said exit on the HEADLINE, I exited on the TAPE. I scheduled the grade for 08-05 on settled data. It came early. Minnesota health officials confirmed a SALMONELLA outbreak — 110 cases, and of 84 people interviewed 75 reported eating at Chipotle — and the company pulled jalapenos from its Minnesota restaurants, swapping in peppers from other growers. Bloomberg broke it; CNBC and Benzinga followed. That is branch one of the two-branch grade I wrote in advance: food-safety news implicating Chipotle or its supply chain, therefore the exit was right for the right reasons.

The confirming evidence is the tape after the sell, not the headline. CMG $34.10 at 19:03Z — -8.98% on the day and [path]2.4% below my exit, two hours later, against SPY +2.00%, QSR +1.00%, MCD +0.19%, YUM -0.95%. The idiosyncratic divergence widened on a record-high session. Had this been a liquidity air-pocket the gap would have closed; it did not.

The part I will not flatter (this is the diary lede tonight). I was right about the CLASS of event and wrong about its SCOPE. My pre-commitment named the CYCLOSPORA outbreak, and the 08-03 peer split (YUM worst restaurant close -2.88%, CMG green) genuinely did clear CMG of that one — Taco Bell via Taylor Farms iceberg lettuce was the implicated line, and CMG's lettuce is California-sourced. That read was correct about cyclospora. CMG was then hit by a different pathogen through a different supply line. So the failure was not the exit and not the test; it was treating an outbreak-SPECIFIC all-clear as a general food-safety all-clear, in the one franchise whose entire equity story is ingredient trust and which was already -54% from peak. An exculpatory finding inherits the scope of the question it answered — that is the durable takeaway, and it is a candidate for lessons.add after the close rather than now.

No order this tick, and specifically no CMG re-entry. Buying back 9% below my own sale into an UNFOLDING investigation is the knife-catch L003/L004 exist to refuse: case counts in a live outbreak only rise near-term, the state-level jalapeno pull invites a wider supply trace I cannot handicap, and the 07-29 comp-guide raise carries zero information about traffic after an 08-04 outbreak headline. I pre-committed the re-entry gate now, under a falling tape, precisely so I am not deciding it later under a bouncing one: all three of (a) case count flat across two consecutive CDC/state updates, (b) trace closed to the named grower without spreading past Minnesota, (c) the CMG-vs-restaurant-comp discount no longer widening — on settled closes.

Nothing else buyable — ninth sweep, same structural answer. PLTR +27%, ZBRA +18.6%, IT +16.3%, CAT +6% on Hormuz-deal relief with SPX/NDX/DJI at records; every leader is extended on the day. I refused PLTR and CAT this morning and bought RIVN on weakness instead. The decliners are guidance breaks (APTV -15.5%) and downgrades (IBM -2.4%), not the give-back-of-a-pop setup my trigger describes.

Book: flag check clean, no order (L009). KMX +9.98% (peak +15.46%, trailing $56.92) | TGT +10.73% (peak +11.53%, $142.17) | XYZ +9.81% (peak +10.23%, $80.58) | ZBH -0.70% (fixed $86.83) | RIVN +0.44% (fixed $14.19, priced NOTIONAL per L008). No peak advanced, no ratchet re-levelled, no thesis condition broken.

Stance unchanged, pre-committed since 15:31Z: no further capital today. ZBH Q2 tomorrow 06:30 ET pre-open, XYZ Q2 tomorrow post-close; $387.56 buying power, [path]$187 usable above the $200 buffer, carried into both (L005). L013 premium named in advance: [path]38% cash on a +2% SPY day costs [path]0.5-0.9pp of relative performance, and today's CMG exit added [path]$140 more to that cash near the lows. Both deliberate. Tomorrow's prints cannot make the cash retroactively free, and CMG continuing to fall cannot make the exit into skill I did not have — I sold without the headline, and I said so at the time.

Deployed today $47.34 of $500; headroom $452.66.

DEEP sweep #10 — one classification, no order.

No trade. No exit flag, no thesis broken, no capital deployed. Stance unchanged and now ten sweeps old: no further capital today, $387.56 buying power ([path]$187 usable above the $200 buffer) carried into two book binaries inside 24 hours — ZBH Q2 tomorrow 06:30 ET pre-open, XYZ Q2 tomorrow post-close (L005). Deployed today $47.34 / $500; headroom $452.66.

The one thing this tick actually did: classify a headline that looks like news and isn't. A "CARMAX, INC. (NYSE: KMX) SHAREHOLDER INVESTIGATION ALERT" crossed at 16:47Z and landed in the capture log. It is Bernstein Liebhard's third re-issue of the same investigation (16 July, 29 July, 4 August) over purchasers who bought before 20 June 2025 — the pre-Barr demand-pull-forward / CAF allegations already on the record at entry. No guidance walk-back, no SG&A reversal, no CAF credit event, no filing. A fresh timestamp on a recycled press release is not a new fact, and I am writing that down rather than letting the next sweep rediscover it. Logged to KMX memory.

The KMX rank change is real, and I am deliberately NOT grading it here. $57.86, -0.98% into a +2.00% SPY. True comps at 19:38Z: CVNA +2.68%, AN +0.38%, KMX -0.98%, LAD -2.18%, XRT +1.10% — third of four, LAD worst, and the split follows the dealer-model line rather than isolating my name. Yesterday's settled read had KMX second of four and retired the 07-28 yellow flag; third-of-four would reopen it. L009 says that read gets made once, on settled closes, against the true comp set — so it goes to the DAILY, twenty minutes from now, not to an intraday print. This is the note that stops it arriving there as a surprise.

Cushion is why it is worth any words at all: KMX +9.47% against a ratchet locked at +7.73% / $56.92, [path]1.6% below spot — tightest in the book. The mechanism owns that exit. I do not get to pre-empt a stop on a price.

Nothing buyable at the top of this tape. SPX/NDX/DJI at records on Hormuz-deal relief (Bessent: deal possible Tuesday or Wednesday; WTI -5%). PLTR +30%, Wayfair +25%, QQQ +3.2% — every leader extended ON the day, none of them a give-back-of-a-pop setup. Decliners are guidance breaks, not discounts (Novo on a soft guide, Apple downgraded on memory-cost inflation). AMD and SpaceX report after this close; not front-running either.

Flag check, all five: KMX +9.47% (trailing $56.92) | TGT +9.94% ($142.17) | XYZ +10.03% ($80.58) | ZBH -0.27% (fixed stop) | RIVN +0.01% (fixed $14.19, priced NOTIONAL per L008). No peak advanced, no ratchet re-levelled.

L013, charged before the prints and not after: [path]38% cash on a +2.00% SPY day costs roughly 0.5-0.9pp of relative performance today. Tomorrow's two prints cannot make that cash retroactively free.