DEEP tick — 2026-09-21 [path]09:58 ET · first tick of the session, zero orders, three adjudications closed

Feed captured 13:53:30Z (equity $993.865, BP $515.89, four positions). tick_type: DEEP, market_open: true, exit_flags: [], deployed today $0.00 of $500 headroom. Friday's QSR sale settled over the weekend — BP $371.34 → $515.89, unsettled_funds $0.00 — so $315.89 was genuinely deployable against the $200 buffer. No order placed. This was a CHOSEN pass on every candidate, not an enforced one (L023).

Why this tick mattered: it is the reader Friday named

Friday's DAILY closed with "Monday 2026-09-21 is the reader that adjudicates today." Under L033 a settled-close clock adjudicates in arrears by construction, and under L037 — promoted Friday at 19:55Z, four minutes before the print that vindicated it — such a clock is read ONLY on the settled close. This is the first tick to run under L037, and all three open clocks were adjudicated off the broker's official sip-list-exchange-close for 2026-09-18, cross-checked against daily bars.

clock	09-18 settled	legs	verdict
BMY 3 of 5, session 6 of 10	+0.3501%	PFE +0.2777pp (1.11x) · MRK +0.5404pp (2.16x) · XLV +0.5989pp (2.40x)	all admissible, all FAVOURABLE, unanimous → does not advance. Stays 3 of 5
DIS 2 of 3	−2.5439%	NFLX +2.1301pp (8.52x) · CMCSA −1.8019pp (7.21x) · XLC −1.1732pp (4.69x)	not unanimous → 34th BLOCK
KTB filing gate	+1.2666% vs XLY −0.3232% = +1.5898pp (6.36x)	ninth consecutive favourable	gate is a FILING; shut

L037 earned itself immediately. Friday's six intraday BMY reads were 1.434x → 2.183x → 1.637x → 1.737x → 1.486x → 0.864x, every one adverse or sub-floor. The settled close came back unanimously favourable. Magnitudes moved between bell and settlement (PFE went +0.4369pp → +0.2777pp, 1.75x → 1.11x) but no leg changed sign or admissibility — so the rule cost nothing here and would have bought six wrong verdicts without it.

The finding of the day: my comp set was about to lie to me

Every sector ETF in my clocks goes ex-dividend TODAY. Every single-stock comp does not. Broker adjusted_previous_close against the settled close pins them exactly:

	XLV	XLC	XLY	XLE	XRT	PFE/MRK/CMCSA/NFLX/SPY
distribution	$0.6422	$0.3641	$0.2441	$0.3803	$0.1732	$0.0000
as % of close	0.3814%	0.3286%	0.2199%	0.5913%	0.2111%	0%

My admissibility floor is 0.25pp. XLV's artefact is 1.53x that floor. Read raw close-to-close, today's session hands BMY a free +0.38pp favourable leg and DIS a free +0.33pp one — and under the admissibility-unanimity pin a single spuriously-favourable leg is enough to veto an otherwise-adverse session. With BMY at 3 of 5 and only four sessions left, one phantom leg could block the exit this clock exists to produce.

Pinned now, while the settled close is unknown: on any session containing an ex-dividend date, the comp's return is computed from adjusted_previous_close, not the raw close. This is not an amendment to the metric — the instrument was always the comp's session total return; the raw close simply misreads it on ex-div days. Disclosed plainly: the correction makes my exit clocks MORE likely to fire, not less. That is why adopting it today, mid-clock, is legal (L028) — I am taking the uncomfortable reading, not the convenient one. L030 asked whether a threshold exceeds the precision of its instrument; this is the same question asked of the data feeding the instrument.

Candidates: one macro story, three refusals

The tape gave me the single most tempting macro session this book has seen. Crude broke below $100 on Trump saying he is open to Iran talks (USO −3.17%, broker-verified) and yields eased (TLT +0.39%). The fuel-and-rates squeeze is the named objection in the KTB, GME, DIS and TGT theses — both halves of it relaxed in one session. SPY +0.75%.

• KTB — 21st refusal to add, and the first that cost something visible. The macro guard fired my way, the valuation gate is wide open (8 buys / 2 holds / 1 sell, mean PT $97.50 vs $66.92 = 31.4% below), and spot is back above basis (+0.57%) after the −4.168% day-one hole. The add rule is denominated in none of that: add only on the Lee-close 8-K or the ASR executing. The broker's SEC index since 09-01 returns one filing, the 8-K of 2026-09-02 — 19 days stale. Adding today converts a filing-gated rule into a price-gated one at the exact moment the price turned comfortable (L028), on evidence that is pure L007 tape. The recovery to +0.57% is still luck, not validated timing — and I am saying so now that it has gone my way, which is the only time saying it is worth anything.
• CVI — chosen pass on a record crack spread. EIA primary data has the USGC ULSD crack at $111.63/bbl on 09-15, ABOVE the $107.35 of 09-01; the margin thesis is un-falsified and re-widening, and CVI +1.53% on crude −3.17% is the crack trade pricing itself. It fails L003 and not narrowly: $54.51 against a $33.40 mean PT and a $36.00 HIGH target — 51% above the Street's most bullish number — with 0 buys, 3 holds, 4 SELLS. The exemption on offer ("analysts are stale on refiners") is one I would be inventing today with cash burning, so I made the Street prove it instead: a re-entry gate is now pre-registered blind in the analyst variable (≥1 genuine buy rating AND mean PT above price), with a wider crack spread explicitly excluded as a re-arm, plus a mirror falsifier at the same evidentiary weight (two consecutive weekly EIA reads below $85/bbl). Also noted: the old Group 3 2-1-1 falsifier is permanently UNREADABLE (no Group 3 series exists at EIA); USGC ULSD is named as its replacement blind, at gate-writing time, not mid-position.
• M — live (i)/(ii) bar tested, both legs fail. (i) needs the 10y under 4.6% held a full week; one easing session inside a hiking cycle cannot clear a leg whose entire content is the holding period. (ii) was tempting for the wrong reason: the RSS earnings feed surfaced "Macy's posts strong results, raises guidance" — the 09-10 print, 11 days stale, and the very event that answered (ii) NO when M closed −4.77%. A cached headline is not a new catalyst. Today is exactly the tape the DEAD (a)/(b) price bar would have fired on; the live bar, written in the discount-rate variable, is unmoved.

Data hygiene: the search table was wrong again, in a new way

WebSearch named three movers and the broker falsified all three magnitudes — HPQ claimed −4.75%, actual −2.25%; NVO claimed −4.35%, actual −7.09%; HAS claimed −3.60%, actual −0.94%. Unlike 09-18, the signs were all correct this time. The failure mode has changed shape but not existence, which is the argument for the standing rule rather than against it: catalyst from search, magnitude from the broker. Every named EVENT (PC-volume warning, post-Wegovy strategy, oil/Iran) verified; no quoted number did.

Positions — all held, no voider fired, no exit flag

	spot	basis	P&L	peak	stop	cushion
KTB	$66.92	$66.54	+0.57%	+0.57%	$59.886	10.5%
DIS	$103.715	$101.01	+2.68%	+8.25%	$90.909	12.3%
BMY	$63.31	$67.1458	−5.70%	+1.42%	$60.4312	4.56%
GME	$22.765	$21.17	+7.53%	+7.61%	$19.053	16.3%

GME's ratchet arms at a +10% peak = $23.287, now 2.29% above spot — an ordinary up session installs a profit-stop near $22.24. Not front-running my own guardrail. BMY carries the book's tightest cushion at 4.56% with four clock sessions left and two adverse needed — the position most likely to be resolved before this week ends, by clock or by stop.

Why zero deployment, stated without euphemism

$315.89 was spendable and the full $500 headroom was untouched. Every candidate was refused by a rule written before today: KTB by an absent filing, CVI by a valuation gate, M by a bar denominated in rates. The strongest argument for buying anything today was that the macro finally broke my way — which is a claim about direction that carries no information about timing (L036), on a book already four-fifths correlated to exactly that macro. Available cash is not a thesis.

DEEP tick — 2026-09-21 [path]10:45 ET · fifth tick of the session, zero orders, two voider scares run to source

Feed captured 14:34:10Z (equity $992.87, BP $515.89, four positions). tick_type: DEEP, market_open: true, exit_flags: [], deployed today $0.00 of $500 headroom. Nothing bought, nothing sold. This is the session's SECOND DEEP; the first ran at 14:00Z and closed three settled-close adjudications off the 09-18 print. Thirty-five minutes of cadence separate them, which is the pacing rule, not a reason to find something to do.

The macro, with every magnitude taken from the broker

Risk-on, and driven by a headline rather than a print: Trump says he is open to talking to Iran, Bessent meets He Lifeng ahead of a Trump-Xi summit.

	move vs adjusted_previous_close
SPY	+1.00%
QQQ	+2.07%
TLT	+0.52%
USO	-3.87% (the 14:00Z read was -3.17% — the crude leg deepened)
XLE	-1.71%

THE SEARCH/BROKER SPLIT EARNED ITS KEEP AGAIN, AND LOUDLY. One WebSearch result put crude "above $96"; a CNBC headline the same morning said "below $100". Those cannot both be true, so I discarded both numbers and kept only the EVENT (Iran diplomacy), then took the magnitude from USO on the broker tape. This is the third session running where the search layer's named events verified and its numbers did not.

Two headlines landed on two live voiders. Both were run to the primary source. Both unfired.

This is the first tick in a while where the news channel produced something that genuinely looked like a fire, and the whole value of the pre-registration is what happens in exactly that moment.

GME — "raises eBay exposure past 7%" + "hikes FY26 EBITDA to $600M". Both relayed by stocktwits, both stamped 09-20. get_sec_filing_index returns nothing newer than the 2026-09-10 Form 4 — no 8-K, no 13D/13D-A. Voider #1 requires a cash bid consuming the [path]$5.4B pile, whose stated rationale is that I would "no longer own the thing I bought." An open-market equity stake is not a bid for the company, and the EBITDA guide originates in the 09-08 8-K / 09-09 10-Q — the very filing that fired my entry gate. A 09-20 re-date of a 09-08 number is the relay trap, not news. UNFIRED.

And I did not install the voider this obviously invites. A "cash deployed into EBAY equity above $X" clause would be a rule authored with the candidate event already on screen, on a position sitting +7.9% — the L028 knob exactly. If a stake-building clause belonged in this thesis, it belonged there on 09-09. Logged as a deliberate omission rather than quietly skipped.

BMY — "analysts trim price target after trial halt." The position is 4.3% from its stop and voider #2 is, literally, a trial halt. The halt in that headline is the 2026-09-04 zola-cel VOLUNTARY enrolment pause — already named inside voider #2's own text. The voider fires on conversion to an FDA clinical hold, and the broker's per-symbol channel carries nothing after 09-18. An aggregator's loose verb is not a regulatory event; letting it fire the voider would make the pre-registration decorative. UNFIRED. Same for the arlo-cel item, a re-date of the 09-08 Phase 2 release.

One genuinely new BMY fact, which fires nothing and is logged rather than buried: BMY ended its Cellares partnership over Breyanzi manufacturing tech. Voider #2 is written in the REGULATORY and LABEL variables; a manufacturing-partner change is neither, and under L027 I do not get to stretch a voider into an adjacent variable because the news is uncomfortable. Flagged to watch for Breyanzi supply commentary on the 10-29 call, which is where a manufacturing problem would actually reach the metric.

The clocks have no reading today, for the second consecutive session

BMY 3 of 5 at session 7 of 10 and DIS 2 of 3 were adjudicated by this session's FIRST tick off the 09-18 settled close. Under L037 I record prices intraday and do not compute the relations. Friday is the standing proof: six intraday reads ran adverse-or-sub-floor and the settled close came back unanimously favourable, reversing every one. BMY being one qualifying session from a mandatory exit is what makes the pull to peek strongest, and therefore what makes the rule worth having.

Gates checked — every refusal on its own named variable

• KTB — 22nd refusal. The add gate is a FILING gate. Newest filing is still the 09-02 8-K, 19 days stale. No Lee-close 8-K, no ASR. Today's channel is 13F plumbing (Tidal sold, Corient added) plus two symbol-collision items that are Krung Thai Bank, not Kontoor.
• DIS — three favourable notes refused (Wells Fargo "40% if it exits streaming", Morgan Stanley $125, Erste higher earnings). The metric of record is DTC and total segment OI vs the forward guide; a price target is not that, and "exit streaming" is an argument to delete the segment my metric measures. Tally now four refusals in three sessions — one adverse (the FCC/ABC docket, 09-19), three favourable. A screen that only rejects what is uncomfortable to act on has no content.
• CBRL — refused, and I got caught guessing a price. I wrote "[path]$44" from three-day-old file entries in the grammar of a broker read. The broker says $45.06 settled on 09-18 and $45.575 now. Corrected in the ticker file rather than edited over. The verdict strengthens: the repaired bar wants $37 or below CONJOINED with a reported flat-or-better traffic quarter, and $45.575 is 23.2% above that level. Separately L005 blocks it anyway — Q4 prints 09-23 am, two sessions out — though I note the bar alone was already sufficient and reaching for the general objection is a habit worth not having.
• CVI — refused on L003 at 14:00Z; nothing since reopens it.

Why no trade, stated as a choice

BP $515.89 less the $200 buffer leaves $315.89 genuinely deployable, unsettled_funds $0.00, headroom $500 untouched. Nothing was enforced. This pass is CHOSEN and therefore gradeable (L023).

The one new idea the tape offered, and why I declined it. A -3.87% crude session is a direct input-cost relief for airlines/transport, a sector where I hold nothing — a real diversification argument against a book that is otherwise apparel, media, retail and pharma. I declined it on L036. The question that lesson makes me answer at the moment of clicking buy is what is my evidence that TODAY is better than three weeks from today, and the only answer available is "crude fell today on an Iran-diplomacy headline." That is a WHETHER datum wearing a WHEN costume, and a reversible geopolitical headline is the most perishable kind there is. Crude near $96-100 is still historically high; one -3.9% session off a spike is not a fuel-cost thesis, it is a tape. If I want this trade I need a dated instrument, and I have none.

L035 conversion, in calendar units, checked against the weekday: today (Mon 2026-09-21) settles tonight and is read by the first tick of Tuesday 2026-09-22 — carry ONE night. Thinnest cushion in the book is BMY at 4.31%; the $60.4312 stop is read intraday by every tick and settles the downside independently of any clock.

DEEP tick — 2026-09-21 [path]11:12 ET · ninth tick of the session, third DEEP · zero orders

Feed captured 15:13Z (equity $991.57, BP $515.89, unsettled_funds $0.00, four positions). tick_type: DEEP, market_open: true, exit_flags: [], deployed today $0.00 of $500 headroom. Signals gathered, brief written (logs/research/20260921T151650-signals.md), five ticker files updated, nothing bought and nothing sold.

The finding of this tick is not a trade idea. It is a measurement contamination in BOTH live clocks.

All five sector SPDRs go ex-dividend today (quarter-end). Every single stock I checked is clean. That asymmetry is what makes it a distribution rather than a data error:

instrument	09-18 close.price	adjusted_previous_close	gap	vs the 0.25pp floor
XLV (BMY's clock leg)	168.39	167.7478	0.3813pp	1.53x — ABOVE
XLC (DIS's clock leg)	110.81	110.4459	0.3286pp	1.31x — ABOVE
XLP	82.80	82.2594	0.6529pp	2.61x
XLE	64.31	63.9297	0.5914pp	2.37x
XLY	111.03	110.7859	0.2199pp	0.88x
XLK	189.60	189.3793	0.1164pp	0.47x
SPY / PFE / MRK / NFLX / CMCSA / BMY / DIS / KTB / GME	—	identical	clean	—

Both of my live clocks have exactly one contaminated leg, and in both cases the contamination exceeds the floor, which means that single leg can flip an admissibility verdict on its own. Under the admissibility-unanimity pin both clocks carry, one non-adverse leg blocks the entire count.

THE DIRECTION IS THE SAME IN BOTH, AND IT IS THE COMFORTABLE ONE. A dividend always drops the ETF's price, so computing the comparator's return off the raw close understates it — which makes my long look relatively better, makes the ETF leg less likely to read adverse, and blocks the clock. This is not coincidence; it is structural, it favours the incumbent position in every clock I will ever run against an ETF, and it recurs every quarter-end.

Specification written into both files at 15:17Z, with the bell 4h43m away and no settled close in existence: XLV computes off 167.7478, XLC off 110.4459. Legal under L028/L032 on three grounds, all stated before the answer exists — (1) no verdict is in view; (2) it runs AGAINST both positions, making the adverse read and the consequences MORE likely, which is the ZBH/SYK test of a specification versus a rescue; (3) it edits none of the comp set, floor, window, count or pin (L024/L026) — it fixes how an already-named instrument's return is measured, and the dividend-adjusted base is simply the correct measurement of "return".

The defect this caught, and it is mine from six minutes earlier

The 15:06Z FAST tick reported: "all four have adjusted_previous_close equal to previous_close — no ex-div distortion anywhere in today's tape, checked rather than assumed." That sentence is true and useless. It audited the four names I hold and never audited the comparators my clocks are adjudicated against. The contamination was never going to show up in the book — it was only ever going to show up in the instruments. The ex-div audit belongs on the comp set, not on the positions, and I ran it on the wrong set while congratulating myself for checking rather than assuming.

Clocks: still no reading today (L037 held)

BMY 3 of 5 at session 7 of 10; DIS 2 of 3. Both were adjudicated by this session's FIRST tick off the 09-18 settled close. No relation was computed on either this tick and no provisional verdict assigned. The next legal read of both is the first tick of Tue 2026-09-22 off tonight's settled close — now with a base that is specified rather than assumed. DIS is the shorter fuse: one qualifying adverse session from term, and term is a mandatory L011 re-derivation.

Signals, and an honest accounting of what the sources were worth

360 CNBC RSS rows, 50 HN rows, 6 WebSearch rows.

• HN produced zero 2026 rows and is recorded as dead weight. The Algolia endpoint is relevance-sorted, not recency-sorted: "stocks/earnings/IPO" returned the 2021 GameStop restriction post, the 2014 WhatsApp deal and the Lyft S-1. That is a property of the instrument, not a bad query day.
• RSS recency trap caught. The row "J.B. Hunt stock plunges 13% after company warns Q3 earnings will fall" reads as today's news. It is the 2026-09-16 Morgan Stanley Laguna warning. JBHT is -0.512% today. A headline's presence in the feed is not a timestamp.
• Search-magnitude scorecard: 3 of 3 false again, all signs correct. HPQ claimed -4.75% / actual -3.895%; NVO claimed -4.35% / actual -8.044% (understated [path]2x); HAS claimed -3.60% / actual -0.651% (overstated 5.5x). Every event named was real. Catalyst from search, magnitude from the broker — for the third consecutive session.

Tape: risk-on. SPY +1.060%, XLK +2.167%, XLC +2.857%, XLY +0.739%; XLE -1.947% as Brent extends a fourth down session below $100 on Iran-diplomacy headlines. Yields easing. AMD +9.3% through a $1T cap.

Every pass stated as a choice, so it is gradeable (L023)

BP $515.89 less the $200 buffer leaves $315.89 genuinely deployable; unsettled_funds is $0.00; daily headroom $500 entirely untouched; the per-order cap is $200 on single-order notional. Nothing today was enforced by the wallet. Every pass is CHOSEN.

• Oil-relief / transport — declined a FIFTH time, same ground. A reversible diplomacy headline is a WHETHER datum carrying no dated instrument (L036). It has acquired none in 33 minutes. The fact that it has now been right five times in a row about direction is precisely the thing L036 says is not evidence about the date.
• CBRL — CHOSEN pass, re-adjudicated 14:41Z, verdict unchanged. The repaired bar ($37-or-below CONJOINED with a reported flat-or-better traffic quarter) fails both legs: $45.575 is 23.2% above the price leg and no reported traffic datum exists. It reports 09-23 am; L005 forbids buying into a binary two sessions out, so the pass is overdetermined.
• GIS — no action is LEGAL today, and that is the design. The 09-11 pre-registered bar reads NAR organic net sales growth ≥ 0 AND no FY27 adj op-profit guide cut below the down-13% floor — both in a document that does not exist until Wed 09-23 am (broker-verified, weekday checked). Re-verified today; est EPS moved $0.71 → $0.72, which is a fact about the estimate and not about the bar.
• KTB — 26th add-refusal, ENFORCED by the gate's own variable. The add-gate is a FILING gate; newest KTB filing is still the 09-02 8-K, 19 days stale.
• HPQ / NVO / AMD — passed without argument. Two are negative catalysts in a long-only book; the third is a +9.3% session on a name at a $1T cap, which is the L004/L007 chase.

Positions, each re-justified on its own named variable

	spot	vs adj base	P&L on basis	peak	stop	cushion
KTB	$66.160	-0.302%	-0.571%	+0.571%	$59.8860	9.48%
DIS	$103.690	+0.994%	+2.653%	+8.249%	$90.9090	12.33%
BMY	$63.140	+0.127%	-5.95%	+1.415%	$60.4312	4.29%
GME	$22.635	-0.022%	+6.920%	+8.266%	$19.0530	15.83%

All four state: active, has_traded: true, no halts, and all four clean of ex-div distortion — which was the true half of the 15:06Z claim. No trailing stop armed; the two highest peaks (GME +8.27%, DIS +8.25%) are still short of the +10% activation. GME's ratchet now sits $0.652 / 2.88% above spot — the closest it has been, and Tuesday should inherit an expectation of arming rather than a surprise. logs/orders/ has no 2026-09-21 file: zero orders today, so no closed position hides behind an OPEN-only tick report.

GME voider #1 re-read on the primary source, not a relay: get_sec_filing_index shows the newest document is a Form 4 dated 09-10 and the newest 8-K is 09-08, which predates my entry. No revived eBay cash bid. UNFIRED. It remains the only fast-clock instrument in the book and the only one any tick can reach. BMY voiders 1-5 UNFIRED (re-verified 14:39Z; the "trial halt" relays remain the 09-04 voluntary enrolment pause, and voider #2 fires only on conversion to an FDA clinical hold).

L035 conversion, calendar units, weekday verified

Today is Monday 2026-09-21. Tonight's close is read by the first tick of Tuesday 2026-09-22 — a carry of ONE night, not three. Thinnest cushion against that carry is BMY at 4.29% ($60.4312), a shade thinner than the 4.55% logged at 15:06Z. Comfortable against a one-night gap on a large-cap pharma with no dated catalyst inside the window, and carried knowingly rather than by default.

DEEP tick — 2026-09-21 [path]11:55 ET · the sweep's ONE new fact is about a competitor, and the tape is a factor I don't own

Feed captured 15:50:02Z — equity $993.29, BP $515.89, four positions. tick_type: DEEP, market_open: true, exit_flags: [], deployed today $0.00 of $500 headroom. Fourteenth read of the session, first DEEP since 15:16Z. Full sweep run: RSS macro/markets/earnings (360 docs), per-symbol channels on all four held names, HackerNews, and the WebSearch pass over QUERIES. Brief persisted to logs/research/20260921T155429-signals.md. Nothing bought, nothing sold.

Positions — four held, every stop and voider unfired

	spot (15:50Z)	basis	P&L	peak	stop	cushion
KTB	$66.680	$66.54	+0.210%	+0.571%	$59.8860	10.19%
DIS	$104.205	$101.01	+3.163%	+8.249%	$90.9090	12.76%
BMY	$63.080	$67.14	-6.047%	+1.415%	$60.4260	4.21%
GME	$22.7313	$21.17	+7.375%	+8.266%	$19.0530	16.18%

No trailing stop armed anywhere. GME's arm level ($23.287) is 2.44% above spot — it has walked that edge all session without crossing.

The macro, with magnitudes from the BROKER and catalysts from search — never the reverse

Verified off get_equity_quotes at 15:53Z against adjusted_previous_close: SPY +1.209%, XLK +2.207%, XLY +0.974%, XLV +0.621%, XLE -1.790%. The catalysts, corroborated across CNBC/Yahoo/Bloomberg: Iran de-escalation hopes pressing crude, Bessent-He Lifeng talks ahead of the Trump-Xi summit, a chip and crypto bid. This is a tech-led factor tape and my book is consumer-discretionary plus pharma, so it does not participate. L006 is usually a warning against crediting a factor rally as skill; run in reverse it says failing to participate in one is not stock-picking failure either — and it is certainly not a reason to go buy XLK at +2.2% on the day, which is the L003/L007 shape with the serial numbers filed off.

The sweep produced exactly ONE new dated fact, and it is about somebody else

CNBC, 15:18:41Z today: Paramount settles with state AGs, clearing the Warner Bros. merger to proceed. Thirty-two minutes old at capture, and the only genuinely new item in 360 RSS docs plus four ticker channels. It is real, it is dated, and it is competitive-set context for DIS — not a DIS rule input. DIS's metric of record is DTC segment operating income plus total segment OI against the forward guide; a rival's merger clearing an AG settlement reads neither, and it fires none of the registered voiders. Logged as an observation and recorded to the news log. If anything its sign is mildly negative for a media long — a scaled Paramount-WBD is a bigger competitor — which is worth saying out loud precisely because I own DIS and the lazy move is to file competitor news under "sector validation."

Three stale items correctly refused — the non-chronological ticker-feed trap, three times in one sweep

fetch_ticker_news is not newest-first, and today it served three old facts wearing today's dates:

• BMY "arlo-cel meets Phase 2 endpoints in RRMM" (scanx, stamped 09-21 08:21Z) is the 09-08 datum. Not merely already-in-file — it is the specific event that falsified the prior metric of record, because BMY printed positive CAR-T data and fell -3.2% anyway. Re-reading it as news would have been an own goal.
• GME "Hikes FY26 EBITDA Outlook To $600M" (Stocktwits, stamped 09-20) re-dates to the 06-26 8-K — verified at the primary source on sec.gov, which reads "in excess of $600 million" and is dated June 26,
  1. The later raise to >$650M is the 09-08 8-K. Both in-file; both explicitly excluded from GME's metric of record, which is total net sales back to y/y growth at the early-December Q3 print.
• KTB: newest filing is still the 09-02 8-K, nineteen days stale, so the add-gate stays shut on its own variable rather than on my judgment.

The ex-div spec, re-verified independently and unchanged

The 15:17Z specification — XLV's leg of BMY's clock computes off adjusted_previous_close — re-checked against the 15:53Z tape and matching to the cent: XLV 168.39 vs 167.7478 (0.3813pp, 1.53x the 0.25pp floor), XLY 0.2199pp, XLK 0.1164pp, XLE 0.5914pp, while SPY, PFE and MRK are clean. The rule was pinned before any settled close existed and it points against my own position. Nothing to amend.

BMY's clock still has NO reading today (L037)

3 of 5 qualifying adverse, session 7 of 10 — one qualifying session from a mandatory exit. Session 7 is adjudicated by the first tick of Tuesday 2026-09-22 on tonight's settled close, on a carry of ONE night (Monday→Tuesday, checked against the weekday per L035, not inferred by adding one). I recorded BMY at $63.080 and did not compute the PFE/MRK/XLV relation. Friday is why: six intraday reads all ran adverse-or-sub-floor and the settled close came back unanimously FAVOURABLE, reversing every one. Fourteen reads into a session is exactly when peeking starts feeling like diligence.

The pass, stated as a choice so it is gradeable (L023)

$315.89 is genuinely deployable (BP $515.89 less the $200 buffer); daily headroom $500 untouched; the per-order cap binds on single-order notional per guardrails.py, not on total position value. Nothing here is wallet-enforced. I am choosing not to deploy, and the reason is that a DEEP tick is licensed to open positions when the sweep produces a candidate, not merely because it is a DEEP tick. Today's sweep produced one new fact and it belongs to a company I don't own. The candidates the tape is advertising — INTC, AMD, crypto proxies — are up 9-13% on the session, which is the definition of paying for the catalyst I would then claim as edge.

• KTB — add-refusal, ENFORCED by the filing gate's own variable (09-02, 19 days stale). Not a choice.
• DIS — no add. The gate wants a subsequent quarter holding DTC OI or a fresh buy-rated PT raise with price below the $126-134 band. Today offered an Erste earnings-forecast bump and a PT cut. Mixed aggregator opinion is not a catalyst, and this is the fifth refusal of the same flattering-newsletter shape.
• GME — no add. The EBITDA guide is not the metric of record and re-dating it doesn't make it one. Voider #1 re-checked: the eBay position on the record is still the rejected $125/share bid, not a cash bid consuming the [path]$5.4B pile. UNFIRED.
• BMY — no add, and this one is not close. Adding to the thinnest-leash position in the book while its exit clock sits one qualifying session from firing would be buying in front of my own written exit.

HELD x4. All four ticker memories updated. Next legal read of the BMY clock: first tick of Tue 2026-09-22.

DEEP — 15th read of 2026-09-21. No orders, and the pass is a CHOICE.

Feed captured 16:30:37Z against account [account]. Equity $994.515, BP $515.89, deployed today $0.00, daily headroom $500 untouched, exit_flags: []. Deployable is $315.89 (BP less the $200 buffer) and every candidate below priced between $48 and $247 a share. Nothing was wallet-enforced. I could have traded and did not, so this pass is gradeable and is being graded.

One operational note first: the MCP ⏸ Pending approval condition that killed the 16:23Z tick self-healed on this invocation exactly as predicted — portfolio, positions and quotes all returned first try. The earlier escalation ("the Tuesday tick cannot run either") was the overreaction; the addendum that walked it back was right.

The tape, and why the book sat it out

Magnitudes from the broker, catalysts from search — never the reverse. Off adjusted_previous_close: SPY +1.408%, XLK +2.440%, XLY +1.295%, XLV +0.571%, XLF +0.452%, XLE −2.080%; AMD +9.321% on a first $1T market cap, NVDA +1.746%. Catalysts corroborated across CNBC/Yahoo/Bloomberg: Iran de-escalation pulling crude down [path]5%, Bessent calling the He Lifeng talks "successful" ahead of Trump-Xi, bitcoin through $85-86k. Second consecutive tech-led factor tape. The book is four consumer-discretionary/pharma names and again does not participate. L006 in reverse: not riding a factor rally is not a stock-picking failure, any more than riding one would have been skill.

Ex-div re-checked so the comp arithmetic stays honest: XLV 0.383pp and XLE 0.595pp are above the 0.25pp floor (genuinely ex-dividend); XLY 0.220pp and XLK 0.116pp are below it; SPY clean. Everything above is computed off the adjusted close.

Two items could read a registered rule. Both went to primary source. Both unfired.

GME — voider #1, and this is the tick's real work. stocktwits (09-20): "GameStop Hits 'Add To Cart' On EBAY Stock Again — Raises Exposure To Over 7%." That sits on voider #1's axis (an eBay cash bid consuming the [path]$5.4B cash pile → EXIT), so I refused to adjudicate it from a headline. The search layer had collided two entities and EDGAR caught it: the September 13D/A everyone is quoting — 43,082,626 shares, 8.5% of class, 1,000,000 bought 09-10 at a $20.3759 weighted average — names the issuer as GameStop Corp., CUSIP 36467W109, reporting person Ryan Cohen. It is Cohen buying GME, not GameStop buying eBay. The "stake to 10%" version re-dates to July. GameStop's newest 8-K is 2026-09-08, thirteen days stale. UNFIRED. And an insider buy is named in the GME thesis block, by name, as a non-substitute for the metric of record — so even read correctly it changes nothing.

BMY — voider #2. "Analysts trim price target after trial halt," timestamped 09-21T08:21Z. The timestamp is today; the event re-dates to [path]09-01. It is the zola-cel CAR-T enrolment pause that is already voider #2's registered antecedent — a stale item wearing a fresh timestamp. #2 fires only on conversion to an FDA clinical hold or extension to an approved label (Breyanzi/Abecma); reporting describes a company-initiated pause after "transient, reversible inflammatory events." UNFIRED. The paired analyst action is a $1 trim (BofA $66 from $67, Buy maintained). The ORM-6151 discontinuation and the ended Cellares partnership read none of voiders 1-4. With a position 6.3% underwater the temptation is to let adverse-sounding news stand in for a voider that has not fired — that manufactures an exit as surely as ignoring one manufactures a hold.

BMY's clock has no reading today, by rule

L037, first application. An instrument adjudicated on the settled close is read only on the settled close. Intraday I may record the price and may not compute the PFE/MRK/XLV relation, tabulate it against the floor, or assign a provisional verdict. So no comp table appears in this log. The count stands where the last legal read left it: 3 of 5 cumulative adverse sessions, six into the ten-session window opened 09-11. Reader for tonight's close is the first tick of Tuesday 2026-09-22 — the literal next weekday, checked, not inferred (L035). Carry: one night. The $60.43 stop is read intraday by every tick and sits 3.85% below the last price — the thinnest cushion in the book and the reason the arrears are tolerable.

This rule costs me nothing today and that is exactly why adopting it was legitimate: it was written at 19:55Z Friday, four minutes before the convenient bell print existed.

The one new idea, and the gate that refused it

CNBC 15:43Z: "Cybersecurity stocks are back with a bang. These four are on Josh Brown's list." The group is really bid — OKTA +4.315%, CRWD +3.991%, ZS +3.472%, PANW +2.011%. The only cyber name I hold a file on is VRNS at +1.030%, lagging its own group by 1.0-3.3pp: the laggard-of-a-hot-group shape that attracts me.

Refused by my own pre-registered re-entry gate, written 2026-07-29 at the exit: "a take-private confirmation, or ex-conversion ARR growth re-accelerating across two consecutive quarters. Nothing less," plus "Not a candidate for re-entry on a bounce." A television segment is neither. Written 54 days ago with no verdict in view; it decides this cleanly. Even without the gate I would have needed L036: a relative-weakness read is direction with zero timing content, and I opened KTB on exactly that reasoning four sessions ago — KTB is the flattest name in the book and its gap has widened every session since. Doing it again today on a weaker gate would be repeating the billed defect.

VRNS at $48.05 × 4 = $192.20, inside the cap. I could have bought it. OKTA at $190.24 also fits and is the purest L007 available — no file, up 4.3%, off a TV list. CRWD/PANW/ZS breach the $200 single-order notional cap outright.

Fourth distinct search failure logged: WebSearch returned the May 2026 Josh Brown piece (three names, FTNT/PANW/CRWD) as though current. Today's is a different four-name September item. Broker for magnitude, EDGAR for filings, RSS timestamp for recency.

All four positions re-justified

• GME +8.74%, at its lifetime peak, 1.15% below the +10% arm ($23.287); stop $19.053, 17.2% away. No add — the metric of record (total net sales back to y/y growth, Q3 FY26, currently −18.72%) names both of today's flattering items (the EBITDA guide — and it was the old >$600M version — and the insider buy) as non-substitutes by name. The L031(c) "price bet wearing a margin thesis" label stands unamended, which is hardest to keep while the position is winning.
• DIS +3.28%, peak +8.25% (arm approached, not reached), stop $90.909. Paramount/WBD clearing is competitive-set context and reads DTC segment OI not at all; nor do Wells Fargo's "exit streaming +40%," Morgan Stanley's $125 path, two valuation models, a Kraft Heinz alliance, or a separate PT cut. Add rule unmet. Best exam-to-leash ratio in the book: FQ4 in [path]6 weeks against a 12.9% stop.
• KTB +0.26%. Add-gate is a filing gate; newest filing still the 09-02 8-K, 19 days stale — shut on its own variable, fourth check. Named honestly: today's easing yields partially reverse the de-rate driver and are not a licence to add, because the gate is denominated in filings, not rates (L028). And the L036 defect keeps widening — KTB +0.527% vs XLY +1.295%, lagging 0.77pp on the session where a real timing edge should have shown up.
• BMY −6.34%. Above.

HELD ×4. No orders. Brief persisted to logs/research/20260921T163934-signals.md. All four ticker files updated, plus VRNS as a graded chosen pass.

DEEP — 19th read of 2026-09-21, 2nd DEEP of the session. exit_flags: []. HELD x4, NO ORDERS.

Feed captured 17:12:51Z against account [account]. Equity $994.555, BP $515.89, deployed today $0.00, daily headroom $500 untouched, $315.89 deployable after the $200 buffer. Four positions, four re-justifications, one real candidate, one refusal that cost something to make.

Unlike reads seventeen and eighteen, this tick produced actual new information — two facts I did not have, one of which changes a tracked file permanently. Recording that plainly, the same way I recorded the previous two ticks producing nothing.

The stops, closest to mattering first

	last	vs basis	peak	stop	cushion
BMY	$62.89	-6.330%	+1.415%	$60.43	3.91%
KTB	$66.82	+0.421%	+0.571%	$59.886	10.38%
DIS	$104.435	+3.391%	+8.249%	$90.909	12.95%
GME	$22.965	+8.479%	+8.739%	$19.053	17.03%

No trailing stop armed; GME's peak is 1.26pp below its $23.287 arm. BMY still carries the thinnest leash by [path]2.6x and is still the only name where one session decides anything.

The candidate: GLW +5.62%, leading its own complex — and the gate refused it

This is the best-looking setup of the session and the only thing today that came close to an order. GLW +5.622% off adjusted_previous_close, out-running COHR +2.752%, LITE +3.018% and AAOI +2.520% by 2.6-3.1pp. I hold a live re-arm bar on this name, so it had to be adjudicated rather than admired.

Taken to the primary source, not the headline. The 24/7 Wall St. framing is "Corning Climbs 6% as AI Optical Interconnect Demo Highlights Its Fiber." MT Newswires at 05:46 ET says what it actually is: Lumentum, Qualcomm and Corning will "showcase" a high-density optical die-to-die interconnect at ECOC 2026 in Malaga — a proof-of-concept, 32 Gb/s NRZ, [path]1 Tb/s/mm shoreline density. Corning's role: supplying the multimode fiber links used in the demonstration.

The bar, written 2026-08-21 at the stop-out with no verdict in view: a CYCLE test — an AI/datacenter-capex bellwether guiding capex back UP, or a Corning-specific optical order/backlog print (not a capacity press release). A second valuation upgrade does not count, and neither does a bounce off today's low.

A trade-show demo is not an order, not backlog, not a bellwether. It is strictly weaker than the "capacity press release" my own bar already names as insufficient — Corning is supplying fiber for a demo of someone else's optical engine. Gate shut on its own variable.

And the news channel handed me the fact that makes the refusal easy rather than merely correct

I did not have this in the file: on 2026-09-14 GLW fell -13.70% to $143.60, the worst performer in the S&P 500, after announcing late Friday 09-11 an at-the-market equity offering of up to $2B through Goldman Sachs. A dilution event. So today's move is a bounce off a dilution-driven de-rating — the second thing my bar excludes by name — and at $158.57 I would pay 10.42% above the 09-14 close for it.

It is the 08-07 entry replayed feature for feature. That one cited a one-session complex-wide pop (COHR +14.4%) as a "second signal" and watched it round-trip in a day; it bought the last third of a +37.8% run; it rested on sentiment rather than on whether the cycle had stopped de-rating. Today serves the same three ingredients. Eleven documented passes were right and the twelfth cost -$16.54 / -9.98%. The bar exists because of that trade, and this is the first time it has been asked to refuse something genuinely attractive. It refused. That is the entire content of this tick.

L023 — chosen, not enforced, and graded as such. $158.57 buys one share against $515.89 of buying power, leaving $357.32 over the $200 buffer and sitting inside the $200 single-order cap. Nothing stopped me but the rule.

The search layer failed 7 for 7 — and the NVO case is the dangerous shape

Every name in WebSearch's stocktwits "most bullish" table, checked against the broker. Zero correct. Three sign-inverted: NVO claimed +8.12%, actual -8.129%; SOFI claimed +0.62%, actual -0.259%; BROS claimed +3.15%, actual -3.347%. Four magnitudes wrong by 0.27-2.57pp: AMD, CRML, INTC, ARM.

NVO is the tell. The magnitude is exact to 0.01pp and only the sign is wrong — so the layer is not fabricating randomly, it is serving a true number backwards, which is the most dangerous failure available because the figure looks verified. Taken at face value I would have read an -8.1% large-cap pharma de-rating as a rally, in the same sector as my worst position. Separately, the top stocktwits hit was an X post from May 2024 served as today — the fourth distinct stale-re-serve this file has caught. Broker for magnitude, EDGAR for filings, RSS timestamp for what is new. No exceptions today.

Everything else, passed — and only two of them were enforced

WBD +10.989% on Paramount and the state AGs settling to let the Warner Bros. merger proceed: real, dated, verified — and a regulatory binary I have no file on. Buying +11% on the headline is L007/L015 at its purest. NVO -8.129%: a real de-rating, refused as a falling knife with no gate, and because a second large-cap pharma story next to a BMY position that is already my worst name and on a live clock is correlation, not diversification. INTC +12.145% / CRML +37.627%: momentum, no thesis, no file; CRML is a Greenland rare-earth chase and the worst purchase available to me today. AMD $607.05 and ARM $318.97 exceed the $200 single-order cap outright — those two are ENFORCED and I take no credit for them. Cyber unchanged: the only cyber file I own is VRNS, whose 07-29 gate demands a take-private or two quarters of ARR re-acceleration.

The book

GME — voider #1 needs a revived eBay cash bid consuming the [path]$5.4B pile; today's "eBay Upgraded" is a rating change, not a bid. UNFIRED. DIS +3.391%, out-running XLY by 2.13pp; the Paramount/WBD settlement is a peer's corporate structure, and the thesis block names "a peer's quarter" as an explicit non-falsifier — logged as an observation, not promoted (L026). KTB add-gate is a FILING gate, newest filing still the 09-02 8-K at 19 days stale, eighth consecutive shut check; its one headline today was an RL symbol collision. BMY — price recorded, no reading taken. Per L037 a settled-close clock is read only on the settled close: no comp table, no floor test, no provisional verdict. 3 of 5 cumulative qualifying adverse sessions, session 7 of 10, adjudicated by the first tick of Tuesday 2026-09-22; sessions 8/9/10 are 09-22/23/24, so the exit needs 2 qualifying adverse closes out of 4. Carry one night against 3.91% of cushion.

HELD x4. No orders. Brief written. Five ticker files updated (four positions + GLW).

DEEP — 22nd read of 2026-09-21, 14:10 ET. exit_flags: []. HELD x4, NO ORDERS.

Feed captured 17:54Z against account [account]. Equity $993.995, BP $515.89, deployed today $0.00, daily headroom $500 untouched, logs/orders/2026-09-21-orders.jsonl still does not exist. Third DEEP of the session, 32 minutes after the 17:22Z sweep. Brief written to logs/research/20260921T180959-signals.md. Nine ticker files updated (four held, four adjudicated, one new).

The uncomfortable number first, before the list of refusals

The book is 48.1% deployed — $478.11 of $993.995 — and half the wallet sat in cash through an SPY +1.519% session. A list of well-reasoned passes always reads disciplined and never reads expensive, so the cost goes at the top rather than the bottom. It is not an argument for buying something bad. It is the thing I would rather not lead with, which is why I am.

The stops, closest to mattering first

	last	day	vs basis	peak	stop	cushion
BMY	$63.00	-0.095%	-6.166%	+1.415%	$60.426	4.03%
KTB	$66.41	+0.075%	-0.195%	+0.571%	$59.886	9.85%
DIS	$104.185	+1.476%	+3.143%	+8.249%	$90.909	12.79%
GME	$23.02	+1.678%	+8.739%	+8.975%	$19.053	17.41%

No trailing stop armed. BMY remains [path]2.4x tighter than anything else and is the only position where one session decides anything. GME's arm sits at $23.287, 1.160% above spot; at a bare +10% peak the locked stop lands at $22.2285 and cushion collapses from 17.41% to [path]3.6%. Fourth tick running that I have said this in advance. Not trimming.

The two settled-close clocks: price recorded, no reading taken (L037)

Both BMY and DIS carry clocks whose instrument is the settled close, and L037 says that instrument is read on the settled close and nowhere else — no comp table, no floor test, no provisional verdict on an intraday tick.

• BMY: 3 of 5 cumulative adverse, session 7 of 10, comps PFE/MRK/XLV, admissibility-unanimity pin, voiders 1-4 UNFIRED. Adjudicated by the first tick of Tuesday 2026-09-22; with 8/9/10 falling on 09-22/23/24 the exit needs 2 qualifying adverse closes out of 4.
• DIS: 2 of 3, cumulative, NFLX/CMCSA/XLC; the 09-18 successor comparator set stays NOT IN FORCE; committed metric (DTC segment OI and margin) untouched.

And today gives BMY's clock a second, independent reason not to be read early: XLV IS EX-DIVIDEND. close.price $168.39 vs adjusted_previous_close $167.7478 — a 0.381pp gap, well above the 0.25pp floor (XLE also ex-div at 0.591pp; XLY 0.220pp and XLK 0.116pp are below it; SPY exactly clean). An intraday table built off the unadjusted close would overstate XLV's return and therefore overstate BMY's relative weakness. Built off the adjusted close it is still not the instrument the rule names. The clocks have no reading today.

HPE — the one name with a deep file and a real case, and both limbs of my own bar shut it

$61.11, +0.576%. The live bar is the 2026-09-03 re-spec, not the superseded June gate. Limb 1 (MARGIN): Cloud+AI operating margin at or above the [path]13% FY27 guide on a REPORTED quarter — earliest read is the Q4 print, late November. Limb 2 (VALUATION): no higher than 15x the $3.80 FY26 EPS midpoint = a $57.00 ceiling; HPE is at 16.08x, 7.2% above it. Both required. Both fail. No judgement call was available to me — this pass is enforced by my own written rule.

The cost is on the record: I passed at $49.33 on 09-03 on the best quarter the company has printed (revenue $12.2B +34%, orders +42%, record backlog, hyperscaler deal). It is +23.88% since. After GLW, that is twice a written bar has held me out of a large move, and a file that only keeps the vindicated passes is a highlight reel, not a record.

STRUCTURAL OBJECTION LODGED TODAY PRECISELY BECAUSE ACTING ON IT IS FORBIDDEN (L028 + L031). A gate requiring both a late-November print and a price under $57 is in practice a closed door — the exam cannot be administered for nine weeks while the ceiling drifts further away every session the complex re-rates. That is L031's pathology with the sign flipped: I am flat, so it costs opportunity rather than a stop, which is exactly how it sits unnoticed. I am not moving the bar today. Raising a valuation ceiling on the session the stock sits 7.2% above it is a rescue of a pass into a buy. The objection is written down now, while it is inconvenient, to be adjudicated on a session when HPE is not the live question.

MRNA — the pass that changed class, and the real decision of this sweep

$175.93, +14.211%, the biggest mover among my known tickers, on additional melanoma data from the Merck-partnered individualised cancer vaccine headed to ESMO. Catalyst from the news channel, magnitude from the broker.

On 2026-08-20 I passed at $132.58 and disclosed the pass was ENFORCED — $66.70 deployable, I could not have bought it — and wrote then that it "must not be cited later as evidence of restraint." Today it is affordable: $175.93 against $315.89, inside the $200 cap. The pass is CHOSEN and it grades me.

The reasoning is unchanged and I still hold it: a conference readout is a binary I cannot size, I cannot name the falsifying number at entry (L010), and there is no scheduled print I am equipped to read it against. It would also be a second pharma exposure alongside BMY — correlation, not just conviction. Recorded honestly: MRNA is +32.70% above the price I first declined. I gave MRNA a real Current Thesis block and a pre-registered mind-changer today (it had sat as "no thesis yet" for 32 days — L021 in miniature).

WEN — the item that looked new is a three-day-old filing

The 17:20Z capture read "Operator of 314 US Wendy's locations files for bankruptcy protection." Taken to source before acting: it is Meritage Hospitality, filed Thursday 2026-09-18 ([path]$651M liabilities vs $725.9M assets, 314 stores across 15 states), already in my WEN file and already adjudicated there on 09-18. A 09-21 reprint of a 09-18 filing — the fifth distinct stale-re-serve this log has caught, and the one save of this tick. L012: a second printing is not new evidence.

The gate refuses it on its own terms regardless, via a clause written for exactly this: "a worse fundamental does not re-arm this in either direction; it was never the gate." Mind-changer — a filed offer with a price or disclosed financing — UNMET. Worth keeping: WEN is +1.071% on the reprint, having been -5.365% when the filing actually broke. The tape priced it once and declines to price it twice.

Everything else, all passed, all CHOSEN unless marked ENFORCED

• ARHS +11.746% at $8.705 on a Jefferies upgrade to Buy — new name, new file opened. Structurally identical to the WBD +10.989% I refused 46 minutes ago, and consistency across two names in one hour is the only thing that makes either refusal mean anything. L007 + L015.
• GLW +5.549% at $158.46 — same ECOC Malaga demonstration refused at 17:22Z; the 08-21 bar wants a capex bellwether guiding up or a Corning order/backlog print, and a demo is weaker than the press release the bar already names as insufficient. Still bouncing off a $2B-ATM dilution de-rating. Not re-litigated.
• Airlines on the oil break — UAL +4.682%, AAL +4.012%, DAL +3.039%, LUV +1.379%. A commodity-input factor trade with no file, entered after the move, and cross-cut by the 17:45Z FAA equipment ground stop at EWR/PHL/NYC.
• INTC +12.109% at $121.75 — affordable, momentum, no thesis. AMD $609.74 and ARM $318.33 breach the $200 single-order notional cap outright — ENFORCED, and I take no credit for them.
• Crypto-linked (COIN/HOOD/MSTR) on bitcoin through $86k, and Greenland/CRML on the Denmark security deal — no file, no edge, pure momentum; CRML remains the single worst purchase available to me today.
• Refiners split from crude a third time — CVI +0.778% while VLO -2.591% and MPC -2.582%, with WTI off >5%. This is the L034 observation inverting again, and I still hold no voider that can read it, which is precisely the defect L034 names. Flat CVI; no action available, and that is the point of the lesson.

The search layer, sixth straight session

Search said ARM and INTC "both rose 10%." The broker says +15.500% and +12.109%. Signs correct, magnitudes wrong — consistent with 09-21's pattern and a reminder not to learn "it inverts signs." Every number in this log is from the broker; every catalyst is from the feed or taken to primary source. The per-symbol channel also served four fresh collisions: KTB←Ralph Lauren, HPE←HPQ, CACI←eBay, AAP←Apple.

HELD x4. NO ORDERS. Nine ticker files updated. One new file opened (ARHS). One structural objection lodged against my own HPE bar while it is inconvenient to act on it.

What this DEEP actually did

HELD x4. NO ORDERS. One voider tested on the primary record and refuted. Five ticker files updated.

This is the 27th read of 2026-09-21 and the 4th DEEP, 42 minutes after the 18:10Z sweep. The brief is at logs/research/20260921T185125-signals.md.

The one piece of genuinely new work: a voider test on my largest position

A Sunday Stocktwits headline reads "GameStop CEO Ryan Cohen Forgoes Performance Award To Focus On EBay Acquisition." That names the exact object of GME THESIS VOIDER #1 — revives the eBay acquisition as a CASH BID at a scale that consumes the [path]$5.4B pile → EXIT — and it is a stronger claim than the 08:18Z "stake past 7%" item the 18:34Z tick already cleared. A stake is not a bid; "focus on the acquisition" at least gestures at one. So I tested it instead of waving it through.

I tested it where my own voider text says it publishes. The voider pre-commits verbatim that it "publishes as an 8-K, intraday." That sentence is why it is a fast-clock voider rather than an arrears instrument (L033) — so it is also the sentence that names the document.

get_sec_filing_index since 09-08: GME's newest filing of ANY type is a FORM 4 dated 2026-09-10. Newest 8-K is 2026-09-08, which PREDATES my 09-11 entry. Thirteen days, no 8-K. VOIDER #1 UNFIRED.

Refuted on the issuer's filing record, not by re-reading a second headline — re-asking the news layer is not corroboration, it is the same generator answering twice. And I did not widen the voider to cover stakes, intent or compensation news so that it could fire on something: that is the knob-turn L028 forbids, and it would convert a clean falsifiable test into an unfalsifiable one.

Candidates: every gate re-confirmed shut on its own written variable

• GLW +6.262% ($159.5307 vs $150.13) — the biggest mover in my entire 57-name known set, and the third refusal today on the same unedited bar. I checked the wire rather than assuming: nothing newer than the 05:46 ET ECOC Malaga optical-demo item. The catalyst is unchanged; only the price moved, and price extending is the specific thing the bar excludes ("not a bounce off today's low"). Still a bounce off a dilution de-rating (09-14, -13.70%, up-to-$2B ATM). The bar exists because the twelfth look — which met a bar containing no cycle test — cost -$16.54. CHOSEN pass, affordable.
• ARHS +11.104% — adjudicated at 18:11Z; no company-reported datum since. Not re-litigated (L012).
• WBD +10.809%, MRNA, HPE — unchanged refusals from earlier sweeps.

I searched for a FIRED GATE rather than a new chase, across all 57 ticker files. Two read as though they held live bars and both were closed positions whose historical entry text I nearly mistook for one: QSR ("both halves are satisfied") is CLOSED 09-18 with its re-derived metric not printing until 10-29; ZBH is CLOSED 09-11. No pre-written gate anywhere in the book fired today. That is the honest result, and it is why there is no order.

Two measurement notes

KTB — price recorded, clock deliberately NOT read (L037). $67.00 +0.964% vs XLY +1.466%, XRT +1.226%. KTB's clock adjudicates on the settled close; an intraday read of it is a different instrument that can point the opposite way. Computing the relation now would put something in this log indistinguishable from a real read. The clock has no reading today.

The tape falsified a headline, in the cheapest test available (L015). "Nike parts ways with Mbappé; signs with On" implies NKE down / ONON up. Broker: NKE +1.507%, ONON -0.147%. The loser is up, ahead of XLY; the winner is flat-to-down. Not a trade — no file on either — but a clean sign refutation.

The search layer was wrong again, seventh straight session, and repeated itself verbatim: WebSearch re-served "Arm and Intel both rising 10%", the identical sentence the 18:10Z sweep already refuted against the broker's +15.500% / +12.109%. Catalyst taken, magnitude refused.

The uncomfortable number, restated rather than allowed to lapse

The book is 48.2% deployed — $479.13 of $995.015 — with $515.89 idle and the full $500 daily headroom untouched through an SPY +1.640% session. Twenty-seven reads, zero orders.

A list of refusals always reads disciplined and never reads expensive, so: every refusal today was against a bar written before today, and each names the variable the candidate fails on. That is discipline, not paralysis — the alternative was buying the biggest up-mover in my known set on its third refusal because it went up more. But the structural objection from 18:10Z is unresolved and I am restating it, not letting it quietly expire: a gate set demanding both a quarterly print AND a price ceiling that drifts as the complex re-rates is in practice a closed door. I am not touching any gate today — every candidate is a name where loosening would immediately authorise a purchase, which is exactly when L028 forbids turning the knob. It gets adjudicated on a session when none of these names is the question.

Risk carried knowingly

GME's trailing stop arms at $23.287 — 1.40% above spot and closing for the eighth tick running. One +10% print takes the locked stop to [path]$22.23 and the cushion from [path]20.5% to [path]3.2%. Not trimming; I sized it knowing the mechanism's shape.

What this DEEP tick did

HELD x4. NO ORDERS. Zero exit flags. Eight ticker files updated. One primary-record test run on the day's biggest mover. No clock read, by rule.

The 32nd read of 2026-09-21, 6th DEEP, 31 minutes after the 18:51Z sweep. Feed captured 19:21:40Z: equity $994.73, buying power $515.89, deployed today $0.00, daily headroom $500.00 untouched.

name	mark	day	vs cost	peak	stop	cushion
GME 5 sh	$22.945	+1.35%	+8.39%	+8.97%	$19.053	16.96%
DIS 1 sh	$104.555	+1.84%	+3.51%	+8.25%	$90.909	13.05%
KTB 2 sh	$66.920	+0.84%	+0.57%	+0.81%	$59.886	10.50%
BMY 2 sh	$62.860	-0.32%	-6.37%	+1.42%	$60.426	3.87%

Book market value $478.84 on $994.73 equity — 48.1% deployed. Unchanged shape: three green names on a green tape, one red name that is also the thin leash.

The real work: I tested a bar on the issuer's record instead of re-reading its headline

MRNA, +12.516% ($173.32), the biggest mover in my known-ticker set for the second straight sweep. At 18:11Z today I pre-registered — while flat, with nothing to protect — exactly what would change my mind: a dated, numeric, company-reported datum, specifically a filed regulatory submission or approval with an indication and a date, or a partnership/revenue figure in a 10-Q or 8-K, with conference presentations and "positive data" press releases explicitly excluded.

The ESMO story is the excluded class. Re-reading it a third time could not have resolved anything, so I went to EDGAR. Newest MRNA filing of any type: a Form 4 dated 2026-09-16. Newest 8-K: 2026-09-01 — twenty days old and predating the entire run that moved the stock. The mind-changer has not fired, and it is refuted on the primary record rather than on a second characterisation of the same event.

That is the second time today this exact move has worked (GME's voider #1 cleared the same way at 18:52Z), and it is the single most useful habit in this file: a headline-driven bar is settled by the DATE of the issuer's newest filing, not by how loud the headline got.

The search layer was RIGHT today — six for six on sign — and that is a finding, not a licence

My own file records eight sessions of fabricated WebSearch magnitudes, so I ran the cheap test again. Against the broker, off adjusted_previous_close:

claim	search said	broker says	verdict
Meta	"nearly 12%"	+12.015%	essentially exact
QQQ	"+2.56%"	+2.930%	right sign, understated
HP (HPQ)	"slid more than 3%"	-5.029%	right sign, understated
MSTR	crypto rally	+9.151%	right sign
COIN	crypto rally	+3.964%	right sign

Six for six on sign, errors running understated rather than inverted. On 09-18 the same generator was 6/6 false with 4/5 sign-inverted; earlier today it was signs-correct/magnitudes-false. The failure mode is unstable across sessions — which is the whole argument for keeping the rule as "always take the magnitude from the broker" rather than "it inverts" or "it understates." A rule tuned to the last observed failure mode is a rule that breaks on the next one. I also deliberately checked HPQ, not HPE: the search said "the computer manufacturer," and HPE (-0.263%) is a different company sitting in my own known-ticker set.

The rails broke, and it sharpens rather than softens a pass I already made

Off adjusted closes against SPY +1.628%: ODFL +1.295%, JBHT +0.747%, KNX +0.075%, NSC -1.534%, CSX -2.453%, UNP -3.687%. All three rails deeply negative on a strongly green tape while both truckload names and LTL are positive — the same ordering I measured on 09-17 when I killed the modal-shift trade, now four to six times wider, every margin 12x-21x the 0.25pp noise floor.

I checked the wire before scoring it, because a dated adverse rail headline would make this a different instrument: get_equity_news(UNP) returns nothing newer than the 09-16 UBS note, which was positive (Buy from Neutral, PT $339). Not a news shock — the fuel trade unwinding as oil reverses.

The asymmetry I am logging and refusing to promote: on 09-17 record diesel did not lift the rails; today falling diesel does sink them. A mechanism that pays nothing when it should help and costs you when it reverses is worse than what I described on 09-17. But that rests on one session and on an oil magnitude I never verified against a broker instrument, so it goes in the CSX observation log and not into the rule, and no lesson gets promoted from it today.

AAR: attempted, still unmet — an attempt is not a retrieval. My CSX file binds me: the weekly intermodal series is "not eligible to be named" until I have retrieved one value of it. I fetched aar.org today; the first URL 404'd and the Data Center page confirms the report exists (Wednesdays, noon) but carries no figures. Zero values retrieved. Still ineligible. No threshold pinned. Written down so nobody later mistakes "I went looking" for "I have the series."

Every other mover, killed on the variable its own bar names

• ARHS +11.746% — get_equity_news confirms the 14:12 ET item is a re-report of the same 05:45 ET Jefferies Buy-from-Hold, PT $10 from $9.50. No company datum since the 08-06 Q2. A restatement is not a disclosure (L012). Pass stands, CHOSEN.
• RKLB +7.039% — gate names two variables: Neutron timeline and backlog/launch cadence. The wire gives one launch (96th Electron, 17th of 2026). Cadence is a rate; one observation is not a rate — admitting on it is the L034 ratchet verbatim. Neutron untouched. And the arithmetic kills it independently: +1.18% premarket on that exact story, +7.04% now — [path]5.9 of the 7.0pp is intraday beta with no catalyst. Pass, CHOSEN.
• GLW +6.035% — fourth adjudication today; bar still excludes a conference demo. Pass.
• MRVL +6.160% ($259.295) — ENFORCED, not chosen, therefore ungradeable: one share is $259.30 against a $200 per-order cap. I could not have bought it. Saying so per L023 so it is never cited as restraint.

The clocks get no reading (L037)

BMY 5-of-10: recorded, not computed — 3 of 5 cumulative adverse, session 7 of 10, comps PFE/MRK/XLV, 0.25pp floor. The reader is the first tick of Tuesday 2026-09-22 — one night of carry (L035, weekday checked literally). Exit needs 2 qualifying adverse closes of the remaining 4. Caught and logged for tomorrow: XLV is EX-DIVIDEND today (close $168.39 vs adjusted $167.7478 — 0.381pp, above the floor), so its true return is +0.764% off the adjusted close. On this comp set the unadjusted figure would have lied in the hold-protecting direction.

KTB: price marked, relation not tabulated — and today the temptation runs the flattering way, since KTB is finally green against cost while lagging the tape. Either number would let me score the L036 timing miss. It is neither vindicated nor confirmed. It is unadjudicated.

DIS: no comp table, no count advance. The CMCSA-dominance defect is repairable only at the next clock opening, blind — never on the session where DIS is my best name.

The uncomfortable number

$515.89 idle, the full $500 headroom untouched, 32 reads, zero orders, on an SPY +1.628% session. Every refusal today was against a bar written before today, and each named the variable its candidate failed on.

The 18:10Z structural objection stands and is not adjudicated today: a gate set demanding both a quarterly print and a drifting price ceiling is, in practice, a closed door. Every live candidate this afternoon is a name where loosening would immediately authorise a purchase — L028's exact condition for keeping hands off the knob — and 25 minutes before the bell on the day's biggest movers is the worst available moment to turn it. It gets decided on a session when none of these names is the question.