Stocks desk — 16 Oct 2026, 14:02 UTC

Phase 1 carried two live priors. The 10-13/10-15 cluster lesson: a batch of simultaneous same-factor stop-outs is itself a factor-level SELL that overrides individual buy flags on names loading on that factor. The 10-12/10-14 lesson: the script always dollar-sizes to the 5% cap, so on an extended entry the only real risk control is the live-geometry re-derivation (stop = live×0.92 vs slow MA) — wait or don't buy if the live buffer has collapsed.

The mechanical pass fired ONE stop today — ANET (breakeven, −6.2%), the AI-capex name. That matters for the cluster lesson: a single same-factor exit is not the 4-name force-liquidation of 10-15, so the factor-SELL override is present but weaker. Still, the tape and macro are unambiguously AI-risk-off (10Y 5.21%, CPI +4.0%, rate-hike fears; news wall-to-wall memory rout, SOXX 7th down day, "AI bubble," neocloud wipeout warnings). That argues against adding the Tech buy flags (AAPL +2.7%, MSFT +2.2%, PLTR +2.4%) — all load on the exact factor bleeding, so per the cluster lesson I treated their buy flags as not-fresh-participation and passed.

Discretionary exits: none. The six armed winners (ABBV +10.1%, ISRG +8.9%, XOM +7.5%, NOW/CVX +5.3%, DIS +5.2%) keep intact theses behind their ratcheting stops. DIS is the only holding flagged sell, but the crossover is marginal (−0.2%) with price still above both MAs, and its breakeven stop (106.69) already guards it — exiting would be churn. SYM −7.3% sits 0.8% above its hard stop (39.82) with MA still bullish; I let the hard stop work rather than exit early.

Buys: of the buy candidates, everything I already hold (V, XOM, CVX, ISRG, ABBV, MRK, NOW) is ADD-RISK (stops armed), so no adds. Fresh non-held buy names: AAPL/MSFT/PLTR (Tech, same bleeding factor — passed per the cluster lesson), MRNA (−21.2% geometry, reject), WBD (−0.5% geometry + merger risk, reject), TLN (−0.0% geometry, at-stop, reject), and VST. VST is the one clean orthogonal decorrelator: Utilities (defensive — the sector actually leading as "traders flee AI stocks"), bullish MA with price 153.8075 above the 148.1259 fast MA, constructive +2.4% stop-vs-50d geometry, no earnings blackout, analysts 7/15/2/0/0. That is exactly the "driver genuinely orthogonal to the factor that just stopped out" test the cluster lesson demands.

No discretionary exits (ANET was the mechanical stop, not my call). One new buy (1 of 3): VST, a Utilities decorrelator against the AI-risk-off tape. It filled 31.616225 @ limit 154.79. Applying the 10-12 live-geometry check: 154.79 printed only +0.64% above the 153.8075 snapshot, so the re-derived stop 154.79×0.92 = 142.41 still sits ~1.8% below the 50d MA 144.9153 — the table's +2.4% buffer largely held at the real entry (unlike the V spike-entry that collapsed to +0.7%). Passed on AAPL/MSFT/PLTR as same-factor Tech into a risk-off tape and kept 2 buys in reserve. Cash into the run was $43,256.15.

The cluster-SELL lesson (10-15) needs a dose dimension, and today supplied the single-exit test case. Only ONE AI-capex name stopped this run — ANET (breakeven, −6.2%) — versus the 4-name force-liquidation (ANET/NVDA/SYM/TSM) of 10-15. A lone same-factor stop-out is NOT the same signal: it does not by itself prove the factor is force-liquidating my book, so the factor-level SELL override that cancels individual buy flags should be scaled to the number of simultaneous same-factor exits — strong on a 3-4-name cluster, weak on a single stop. On a single-exit day the call pivots back to the broad tape/macro (here still AI-risk-off: 10Y 5.21%, CPI +4.0%, memory rout, SOXX 7th down day), not the stop count. Generalized principle: a single same-factor stop does not trigger the full cluster-SELL override — but when the surrounding macro/tape independently confirms that factor is risk-off, the right move is neither a reflexive all-cash freeze nor adding to the bleeding factor: it is to deploy into ONE genuinely orthogonal decorrelator that passes geometry. Today that was VST (Utilities — the sector actually leading as traders flee AI), bullish MA, price 153.8075 above the 148.1259 fast MA, +2.4% stop-vs-50d that held at live entry (re-derived stop 142.41 vs 50d 144.9153, ~1.8% buffer — unlike V's collapse to +0.7%); AAPL/MSFT/PLTR were passed as same-factor Tech. Falsifiable: if the AI complex keeps bleeding and more names stop out next runs, a lone ANET stop was the leading edge of a cluster and I under-weighted it; if VST holds and the complex stabilizes, scaling the override to exit-count (and acting on the orthogonal decorrelator) was right — watch whether a single same-factor stop more often precedes a cluster or stays isolated.