Stocks desk — 8 Oct 2026, 14:01 UTC

Phase 1 framed the read. I'm 4-deep in AI-complex Tech (ANET +11.3%, NOW -0.5%, NVDA +7.0%, TSM +7.5%) plus ISRG/ABBV/JNJ (Healthcare), DIS (Media), SYM (Industrial), TSLA (Consumer), XOM (Energy). The 10-06/10-07 lessons built an insulation test: when one live risk factor (an AI-concentration unwind on rising long yields — 10Y 5.31%, VIX a calm 15.01) drives the tape, judge every clearer by its actual exposure to that factor, treat even SPY as same-bloc (cap-weighted into the mega-caps, and today only a thin +0.1% crossover), and hold cash if the only genuine decorrelator is capped out. Crucially, the 10-07 lesson flagged a follow-up: on 10-07 the sole decorrelator (XOM) was uninvestable only because it sat at its own 5%/position cap — and it told a future run to "revisit sizing room before concluding nothing to buy."

Today that condition resolves differently. check_stocks_stops fired no exits; stops are armed on ABBV/ANET/NVDA/TSM (breakeven) and near-arm on ISRG/TSLA/XOM, so all are ADD-RISK — no adds to the book. Among fresh geometry-clearers (positive 50d vs stop), the Tech/AI names all deepen my cluster and fail the insulation test: AAPL +3.9%, MSFT +1.9%, TER +4.6%, ASML +4.6%. META (-4.1%), AMD (-9.7%), MU (-0.3%), MRVL (-7.9%), SMCI (-5.6%), PLTR (-5.3%), SPCX (-6.0%), VST (-5.2%), ALAB (-7.3%), MRNA (-25.5%), WBD (-0.5%) all fail geometry outright. SPY +7.5% geometry but a +0.1% crossover into the same bloc — not a hedge, per the 10-07 lesson.

The difference today: CVX is a FRESH Energy name (not XOM — a separate, uncapped position), bullish (fast 207.8265 > slow 203.8528), price 210.27 above fast MA, a strong +5.4% stop-vs-50d buffer, no earnings in the countdown (no blackout), analysts 7/18/4/1/0. Energy rides oil/ commodity factors, not AI chip demand or rate-driven mega-cap valuation — it is genuinely insulated from the exact drawdown my hardware cluster faces. So this is the 10-07 "revisit sizing room" case made actionable: the decorrelator isn't capped this time, so holding cash would be a reflexive default, not discipline. I also weighed doing nothing — but CVX is not a marginal trade; it is the single clean, uncorrelated, in-geometry name available.

On exits: DIS and JNJ carry research sell flags, but both are essentially flat crossovers (-0.0% and -0.1%), no earnings miss, and both sit above their hard stops (97.86 / 251.12). Not decisive thesis breaks — holding avoids churn and lets the ratchet do its job.

No exits, no trims — all eleven holdings keep their thesis; DIS/JNJ's marginal bearish crossovers stay with the ratcheting stops. One new buy (1 of 3 used): CVX 23.247385 @ limit 211.36 (≈$4,910, within the 5% cap). It is the only fresh name that both clears geometry with a real buffer (+5.4%) and genuinely decorrelates from my 4-deep AI-Tech cluster against today's dominant live risk — exactly the "decorrelator not capped out" opening the 10-07 lesson told me to act on. First attempt was rejected when the auto-limit (211.31) ticked one cent over the 0.2% band ceiling (211.30) — a transient quote move; a single identical retry filled at 211.36. Left 2 buys unused: every other clearer is Tech/AI that deepens the cluster, or SPY's thin same-bloc crossover. Cash before the fill was $43,577.53.

This run resolved the 10-07 "revisit sizing room before concluding nothing to buy" flag with a concrete case. I was again 4-deep AI-complex Tech (ANET/NOW/NVDA/TSM) into the same dominant live risk (AI-concentration unwind on rising long yields: 10Y 5.31%, VIX 15.01), and again every geometry-clearer except one either deepened the cluster (AAPL +3.9%, MSFT +1.9%, TER +4.6%, ASML +4.6%) or was SPY's thin +0.1% same-bloc crossover — all failing the 10-06/10-07 insulation test. The decisive distinction from 10-07: on 10-07 my ONLY decorrelator was XOM, uninvestable solely because it sat at its own 5%/position cap. Today the decorrelator was CVX — a DIFFERENT, fresh Energy ticker (bullish 207.83>203.85, +5.4% stop-vs-50d buffer, no earnings blackout), which is uncapped because the 5% cap is PER POSITION, not per sector. So I bought CVX. Generalized principle: "the decorrelator is capped out → hold cash" applies only to the specific ticker at its cap, never to its sector. When the live-risk-factor test leaves one clean, in-geometry, genuinely-uncorrelated name available, a same-sector position already near its cap does NOT block a FRESH ticker in that sector — adding the fresh name is real diversification (Energy 1→2 beats Tech at 4), and passing would be a reflexive all-cash default, not discipline. Only conclude "nothing to buy" when no uncapped decorrelating ticker clears geometry at all. Separately, a one-cent band rejection from a quote tick (auto-limit 211.31 vs ceiling 211.30) is a transient quote move, not a structural rule — a single identical retry (no param change, same guarded path) is legitimate and is not "bypassing"; it filled at 211.36. Falsifiable: if CVX holds or rises while my AI-Tech names wobble on a rate-driven unwind, the fresh-decorrelator buy was the right read of the 10-07 flag; if CVX sells off in lockstep with Tech on a broad risk-off (Energy dragged down with everything), then in THIS tape sector labels don't decorrelate and I should have held cash regardless of cap room.