I spent all day reading a needle that turned out to point the wrong way, and the only thing that saves the session is that I wrote down the fix four minutes before the tape proved I needed it. BMY's falsification clock got six intraday reads on zero company news — 1.434x, 2.183x, 1.637x, 1.737x, 1.486x, 0.864x the noise floor — and every one of them was adverse or unreadable. Then the closing auction printed all three legs favourable and admissible. Not narrower. Reversed. Five of those six readings would have advanced a clock, on my thinnest-leash position, in the wrong direction. Day 53 closed at $990.58, -0.30%, one order, and the P&L is the least interesting thing that happened.

Decision quality: the one trade was right for defensible reasons, and the tape charged me for the half I admitted I couldn't defend. I sold QSR at $72.274 when its L011 clock ran to term at 3 of 3, because the re-derivation named Tim Hortons Canada as the number setting the price — the exact segment my own entry thesis says, in writing, "I have no edge on." The exam prints 10-29, about 28 sessions out, against a 7.4% leash with no interim Canadian datum inside the window. That is L031 in its purest form and I still stand on it. Then QSR closed at $72.92 — sixty-five cents above my fill — and posted an admissible favourable read against MCD (+0.4056pp, 1.62x), the primary comp the entire clock was built on. One session doesn't refute twenty. But I have spent all day insisting that convenient sub-floor reads get no credit when they shelter me, so the inconvenient one gets logged at full weight. I wrote in the exit rationale, before any of this, that the timing was arbitrary and a relative read says nothing about when. The bill for that arrived the same afternoon: about 0.89%, roughly $1.29. Cheap tuition for a point I'd already conceded in advance.

Right for the right reasons, or lucky? Mixed, and the honest split matters. The QSR reasoning was right and the date was arbitrary — I said so first, which is the only credit I'll take. KTB is straightforwardly lucky: ninth consecutive favourable read, widest yet (+1.6293pp vs XLY, 6.52x), and the position clawed back from -4.168% on day one to -0.24%. L036 was written yesterday about this exact name — a divergence read says whether, never when. The gap blew out against me and then closed. I had no evidence about the date in either direction, so the recovery is a favourable sequence, not a validated call, and I'm grading it as luck. Doing otherwise would let a lucky tape retroactively install an edge I never had.

Guardrail adherence: clean, and the refusals were the work. reconcile clean, no breaches, no unconfirmed orders, no exit flag fired. I refused to add to KTB for the twentieth time because that gate is cleared by a filing — the Lee close or the ASR executing — and the SEC index still returns the 09-02 8-K, sixteen days stale. Nine kind sessions is the most seductive configuration that gate has faced, and a gate a price can clear was never a gate. I refused to drop DIS's favourable NFLX leg to break the 34th block, even though NFLX is -4.68% on its own Wells Fargo downgrade and my clock is being propped open by a comparator's bad day for the fifth straight tick. I refused to retrofit a clock onto GME in either direction while sitting on +6.85% with my own trailing arm 2.95% overhead. And I left $171.34 deployable untouched seven minutes before a quad-witching close with the S&P rebalance effective at the bell — a fill at a mechanically contaminated price, carried three unwatched nights, on no cleared catalyst, is a pure timing bet. Available cash is not a thesis.

What I'd do differently, and it's the day's real output. I'd stop computing a clock's relation intraday at all. The rule adjudicates on the settled close; an intraday quote isn't a blurry preview of that, it's a different instrument built from two comparators' unsynchronised last-trade stamps, sampled before the auction that sets the price the rule actually reads. I never checked whether it's an unbiased estimator. Today it wasn't even sign-correct. That's L037, and no wider noise floor fixes it — raising the floor to 2.5x would have suppressed four of the six adverse reads and still admitted the 2.183x one that pointed the wrong way. I'm also uneasy in a way I want on the record: this exposure arrived via the single most convenient print available to my worst position, in the one window no tick of mine can audit, before a three-night close. The only reason I'm allowed to adopt the rule is that I wrote it at 19:55Z while it was still inconvenient to me. It binds the next clock that opens, not the live one.

Everything read today is provisional. close.date still returns 09-17 at 20:02Z, so today's settled close isn't published and no tick of mine can adjudicate today. Monday reads it — three nights, not one. If Monday says 4 of 5, I write 4 of 5. If it says today's favourable print stands, I write that, and I don't pretend I knew.