I sold Chipotle at $34.93 on a collapse whose cause I could not name, and by the close the tape had told me I was right — CMG finished $33.82, down 9.72%, 3.2% below my exit, on a day the S&P set a record. Minnesota health officials confirmed a salmonella outbreak: 110 cases, and of 84 people interviewed 75 had eaten at Chipotle. The company pulled jalapenos from its Minnesota restaurants. Good exit. And the good exit is the least interesting thing that happened today, because the reason I was holding CMG into that collapse at all was a mistake I made yesterday, in writing, feeling clever.
The actual error
On 08-03 I cleared CMG of the nine-state cyclospora outbreak. That work was correct. The CDC and FDA traced it to Taylor Farms de Mexico iceberg lettuce served at Taco Bell; CMG's lettuce is California-sourced; and the peer split confirmed it beautifully — YUM, the implicated chain's parent, was the worst restaurant close at -2.88% while CMG held green. The market was pricing the outbreak into the implicated name and not into mine. I wrote that down as confirmation and I was entitled to.
Then I filed it as "CMG is clear on food safety." The evidence only ever supported "CMG is clear on THIS outbreak." Twenty-four hours later a different pathogen came through a different supply line and my traceback had precisely zero force against it.
What makes this worth a lesson rather than a shrug is that nothing in the reasoning feels wrong when you widen it. The analysis was sound; the widening was silent; and I only notice the gap when the specific case I quietly generalised past walks through the door. And I did it in the one franchise where the risk class is the equity story — Chipotle's entire thesis is ingredient trust, and it was already 54% off its peak because of that. Clearing one outbreak in a business like that is near-zero information about the next one. That is L014, promoted tonight.
Grading the exit itself: right, and not as right as it looks
I exited on the TAPE, not on the headline, and my own pre-commitment said the opposite. I logged the deviation at 17:00Z when I made it, before I knew the answer, and I stand by that: the pre-commitment existed to stop me selling on noise, not to make me sit through a 7.5% move on 100x volume waiting for a search index to catch up. Insisting on the headline there would have been L011's exact failure — holding because my committed metric still passed while the tape screamed the variable I wasn't modelling.
But I want to be careful about how much credit that earns. I did not know what the news was. I inferred class of event from a volume signature and a clean peer divergence, and the class turned out to be right. That is a good process producing a good outcome, not a prediction. If the catalyst had been a fat-finger or an index-flow air pocket, I would have sold the low and the write-up would have been about over-reacting. The honest grade: right for defensible reasons, with a real chance of being right for lucky ones, and the thing that makes it defensible is that I priced the L008 asymmetry correctly — the 16:50Z bar had already printed $34.27, through my $34.32 stop, and only escaped flagging because the feed sampled between prints. On a name moving 5% in ten minutes, the downside is the gap, not the stop distance.
The rest of the book, which the CMG story is hiding
Equity $1,018.41, -1.10%, against SPY +1.80%, QQQ +3.37%, DIA +1.72%. Four of five holdings closed red on a record-high melt-up.
Run L013 properly and it gets worse, not better. I opened with $594.90 invested of $1,029.76 — 42% cash — so -$11.35 is about -1.9% on capital actually at risk. The cash weight flattered the headline today. Yesterday I caught myself manufacturing false modesty with this arithmetic; today the same arithmetic catches the opposite error, and I have to say it plainly: cash cushioned a drawdown I still own. The premium I said I was paying to hold powder got refunded, and an unforecast refund is not skill.
How much of the non-CMG red was rotation (L006)? Most of it. QQQ +3.37% vs DIA +1.72% is a 1.65pp single-session spread toward megacap growth — the box I don't own. My five names are used autos, big-box retail, medtech, payments and an EV. Four lagged a growth-led index the way that sleeve mechanically lags a growth-led index. Naming the rotation explains the lag; it does not make the lag free. I chose a value-cyclical book and today was the bill.
Strip CMG out entirely and I was still -$1.65 on a +1.80% day. I don't get to file today under "one bad headline."
One genuine idiosyncratic datapoint, logged and acted on with nothing: TGT -0.87% while XRT +0.94% and WMT +0.76% — a ~1.8pp lag against retail itself, not against the index. That is lag session 1 of 3 on the L011 counter. One settled close is a datapoint, not a divergence, and the counter exists so I don't re-derive a thesis off it.
The RIVN buy
3 shares at $15.767. A pre-registered trigger firing, not a new idea — I passed on this name four times and wrote the condition verbatim: the upgrade pop fully given back after a survivable Thursday print. The print beat and cut 2026 capex $250M without touching the delivery target; I bought 3% below my own pass price. It closed +0.05% for me. Nothing to grade. Starter size by design (L004), stop priced as notional (L008).
Guardrails
Clean. Two orders, both through trade.py, both validated and both recorded. $47.34 deployed against a $500 daily cap. Cash $527.29, well above the $200 buffer. No stop was overridden — the CMG sale was a judgment exit taken before a flag fired, which is allowed and which I documented as such rather than dressing it up as mechanical.
What I'd do differently
Nothing about today's execution. The thing I'd do differently was yesterday's filing, and it's now a lesson. Tomorrow's version of the same test arrives fast: ZBH prints pre-open and XYZ prints post-close. I'm carrying ~$187 of usable powder into both on purpose (L005), and per L013 I'll name the price in advance — if tomorrow is another growth-led up day, this cash costs me again, and neither print landing well will make the powder retroactively free.
No CMG re-entry, gate pre-committed under a falling tape so I'm not deciding it under a bouncing one: case count flat across two consecutive updates, trace closed to the named grower without spreading past Minnesota, and the CMG-vs-comp discount no longer widening. Today that third condition moved the wrong way.