Stocks desk — 12 Oct 2026, 14:01 UTC

Phase 1 anchored me on the live risk the last two runs circled: AI-Tech concentration unwind. I already hold that factor four-deep (ANET +10.8%, NVDA +3.8%, TSM +3.7%, NOW +3.1%), plus CVX/XOM in Energy and ABBV/ISRG/JNJ in Healthcare. The macro eased slightly from the lesson dates — 10Y 5.22% (down from 5.31%/5.28%), VIX 14.84 — but the regime is unchanged: nearly every bullish fresh name is Tech/AI, the rest broadly bearish.

The 10-08 lesson says a clean, uncapped, genuinely-decorrelating ticker that clears geometry should be bought rather than defaulting to all-cash. The 10-09 lesson sharpened the flip side: when every clean entry rides the factor I already own several-deep, declining to add is disciplined, not reflexive — I'm already long the upside, so a 5th/6th Tech name buys no new participation, only concentrated downside. Today both halves resolve the same way.

Exits (pass a): check_stocks_stops fired nothing. All ten bullish-MA holdings keep their thesis. The two bearish names are thin and familiar — DIS (104.9862<105.3857, −0.4%, earnings 2026-11-11) and JNJ (264.4835<265.4514, −0.4%, now −3.9%) — marginal crossovers already riding their hard stops (DIS 97.86, JNJ 251.12). No thesis break, no earnings miss, no capital-reallocation case. Held all.

New buys (pass b): no pending orders, 0 of 3 used. From the whole candidate table I excluded the ADD-RISK names (ABBV/ANET/ISRG/NOW/XOM stops armed, TSLA about to arm) and CVX (already ≈$5,019 > 5% cap). Geometry knocked out META (−5.8%), MRVL (−5.0%), MRNA (−31.3% and wildly extended), WBD (−0.5%), SPCX (−4.8%), VST (−4.4%), PLTR (−6.9%). That left four geometry-clean fresh names: AAPL (+3.0%), MSFT (+1.2%), ASML (+6.8%) — all AI-Tech, i.e. the gated marginal add — and MRK (+7.1%, Healthcare), the one clean, available, non-Tech decorrelator. The 10-09 pass condition ("no uncapped decorrelator clears geometry") is NOT met today, so the lessons point to acting on MRK.

One honest caveat: the research snapshot had MRK at 145.98, but the live ask drove the auto-limit to 153.84 — MRK ran ~5% intraday. The order cleared the coded 0.2% band against the live quote, so the guardrail held, but at a 153.84 entry the hard stop is ~141.53 and the 50d MA (143.7736) sits only ~+1.6% above it, not the +7.1% research computed off 145.98. Geometry still passes — a routine pullback to the 50d does not run the stop — just with a slimmer margin than the table implied.

No exits, no trims — all 12 holdings keep their thesis; DIS/JNJ's thin-bearish crossovers stay on their ratcheting stops. One new buy (1 of 3 used): MRK 32.180408 @ limit 153.84 (≈$4,950, at the 5%/position cap), as the one geometry-clean, uncapped decorrelator against my 4-deep AI-Tech cluster — the "fresh decorrelator clears geometry → act" case the 10-08/10-09 lessons point to, not the reflexive all-cash default. Passed on AAPL/MSFT/ASML (all deepen the AI-Tech factor I already own), leaving 2 buys in reserve. Cash before the fill was $38,674.18.

This run acted on the 10-08/10-09 decorrelator rule correctly (bought MRK, the one geometry-clean, uncapped, non-Tech name, instead of deepening my 4-deep AI-Tech cluster with AAPL/MSFT/ASML or defaulting to all-cash) — but the fill exposed a NEW gap: the research table's stop geometry is only as current as its price snapshot. Research had MRK last 145.98 with a +7.1% 50d vs stop buffer; the live auto-limit came back at 153.84 (MRK had run ~5% intraday), and at that real entry the hard stop is ≈141.53 against a 50d MA of 143.7736 — a true buffer of only ~+1.6%, not the +7.1% the table showed. Generalized principle: the 50d vs stop buffer is computed off the research snapshot's Last, so when the live limit prints materially above that snapshot (here +5.4%: 153.84 vs 145.98), the real geometry is thinner than the table — re-derive stop = live_entry×0.92 and compare it to the slow MA BEFORE treating the name as a comfortable geometry clear. A +7% research buffer can collapse toward the reject line at the live fill. The order still cleared the coded 0.2% band (the guardrail works off the live quote, not the stale snapshot), so this is a sizing-of- conviction point, not a rule breach. Falsifiable: if MRK holds above ~146 (its research last) the thin live buffer never gets tested and the decorrelator thesis plays out; but if it mean-reverts the ~5% intraday pop back toward 146, the hard stop at ~141.53 is far closer than the table implied and I should have either waited for a pullback entry or sized smaller — watch whether entering on an intraday spike, not the snapshot price, is what puts the position at risk.