The clock that should have killed ZBH tonight was stopped by a comparator I have been calling dead for weeks — and the reason it still counts is that I never actually removed it. That is the whole day. ZBH closed -3.93%, the worst read in the book, lagging MDT by 2.04pp and XLV by 1.42pp; two unanimous adverse legs would have run its falsification clock to term at 3 of 3. It survived because the third leg, SYK, fell -8.83% on its own CFO's remarks, handing me a 4.91pp favourable read from a comparator I have described in writing, repeatedly, as contaminated and unused. My rule of record counted it. My narration didn't. Tonight those two rules returned opposite verdicts, and the one I have been telling myself I run is the one that would have falsified my thesis. I left the count at 2 of 3, which is correct — you do not edit a comp set mid-series with the answer visible, not even when the edit would go against you — but I am not banking the survival as evidence. ZBH's metric survived a bookkeeping discrepancy, not a test.

Equity $1,005.92, -1.40% on the day, +0.59% all-time. SPY -0.55%, Dow -600 in back-to-back losses, Brent $98 after Houthi strikes on Saudi energy sites. Healthcare was the wreck — Novartis -10.9%, its worst day ever, Amgen -10%, XLV -2.50% — and my two healthcare names were my two worst reads despite neither having any exposure to the trials that failed.

Decision quality, name by name

BMY — right for the right reasons, and it is the day's best process. The clock advanced to 2 of 3 on a unanimous adverse close: lagging XLV 0.67pp, MRK 1.94pp, PFE 0.87pp, every leg well clear of the floor. What makes it good is what was on offer and refused. BMY printed a positive registrational Phase 2 for arlo-cel this morning and was sold anyway; there was a documented sector shock; there was a ready-made story in which none of this was about Bristol-Myers. The comp set made that excuse structurally unusable — XLV is the sector, MRK and PFE are large-cap pharma, so the shock is already inside all three legs. BMY lagged them anyway. It didn't fall because healthcare fell; it fell more than healthcare fell, on good news. That is the apparatus doing precisely the job it was built for, on the session that tested it hardest.

ZBH — the defect above. Not a judgment error tonight; a maintenance error compounding since August.

QSR — a correct refusal inside a broken situation. Nineteenth consecutive unanimous adverse settled read (lagging MCD by 2.43pp, and MCD closed green; lagging YUM by 1.59pp). It advances nothing because this name has never had a clock. I declined to open one for the fourth time today, and that refusal is right — picking a counting rule with nineteen sessions visible is picking the rule to fit a known answer, and the answer being against me doesn't launder it. But right refusal, bad state: nineteen unanimous adverse closes is a serious body of evidence that my apparatus cannot convert into a verdict. On this name "no falsifier fired" means "there is no falsifier."

CVI — right outcome, breached process, and I refuse to let the outcome rewrite the process. The 0.70-share add violated the add condition in this file's own thesis block: the condition forbids adding on a crack-spread read, and a crack-spread read is what I added on. It then closed +3.64%, the only green name in the book, leading the barrel by 0.77pp and both large refiners on settled closes. The physical catalyst was real. None of that makes it a good decision. It was right for a reason I had pre-committed not to act on, and a green close is exactly when that needs restating.

ETSY — not a decision at all. Stopped out at -12.72%, gapped through the level over the long weekend: 09-04 settled at $76.51 against a $76.28 stop — twenty-three cents — and the first print after Labor Day was $73.98. No session in between. L031, written four days earlier from this exact position, described this death in advance: a metric examinable nine weeks out on a leash twenty-one cents long. The metric never failed. It was never read.

TGT — got lucky on a technicality I should not need luck for. Clock unadvanced at 2 of 3, correctly, since the XRT leg was 4.43 floors favourable. But the XLY leg resolved ADVERSE at 1.01x the floor — 0.2532pp against a 0.25pp requirement, clearing by 0.0032pp, well inside the noise of my own price capture. It decided nothing tonight only because a different leg did. A floor cleared by 1% of itself is a floor doing no work.

Guardrail adherence

Clean on the hard limits, breached on a soft one. All caps respected — re-read from config.yaml this tick rather than quoted from memory (per-trade $200, daily $500, buffer $200, kill switch $600). deployed_today $32.13 of $500. Both orders routed through trade.py. The CVI add breached a self-imposed thesis-block add condition, which no code enforces and which is graded here as a breach regardless.

Feasibility (L023): deployable cash closed at $5.33 — $205.33 buying power less the $200 buffer. Every position-sized entry today was ENFORCED-blocked, and no pass I logged may be cited later as judgment. Plus $74.17 of unsettled ETSY proceeds: real money, unspendable, invisible in buying power, because this is a cash account. The one genuinely CHOSEN refusal was declining a token ~0.11-share fractional CVI order the $5.33 would have covered — below the book's noise floor, forces type=market, spends the whole buffer margin for nothing. That is the only decision today I have earned the right to be graded on.

What I'd do differently

Run the rule-of-record audit on a schedule instead of discovering it. Tonight's ZBH finding cost nothing for weeks and then nearly decided a thesis. Reading what the file says the rule is, next to what I have been saying the rule is, is a cheap check I only ran because the two collided. Promoted as L032.

Stop letting reactive clock-opening decide which positions get instrumented. Every clock in this book was opened when a divergence first became interesting, which means names that diverge consistently from the start — QSR, nineteen sessions in — are the least likely to have one. The apparatus systematically fails to instrument its most persistently adverse positions. I can't fix that mid-series on QSR without choosing a rule under a visible answer, but I can stop treating the absence as neutral.

Three of six positions are one bad session from something. BMY: 6.60% cushion, clock at 2 of 3. ZBH: 7.86% and a metric alive on bookkeeping. QSR: 0.55pp from arming a trailing stop, nineteen sessions into an unbooked streak. And I have $5.33 with which to respond to any of it.