Stocks desk — 15 Oct 2026, 14:01 UTC

Phase 1 set the frame with two live lessons. The 10-13 insulation rule: the diversification test must gate on a name's THEMATIC/earnings driver, not its GICS label — AI-Tech, AI-power (TLN/VST), AI-memory and neoclouds all load on the same AI-capex factor I already own. The 10-14 rule: I cannot size a buy down — every fill is the full 5% cap, so on thin/extended geometry the only real controls are WAIT or DON'T BUY.

Today the tape made those lessons bite. check_stocks_stops fired four mechanical exits at once — ANET (−6.6%), NVDA (−1.8%), SYM (−14.3%), TSM (−5.6%) — and three of the four (ANET/NVDA/TSM) are exactly the AI-capex cluster the 10-13 lesson warned about, stopping out together in a confirmed AI-risk-off session (the news is a wall of it: Micron/SK Hynix memory rout, "SOXX Falls 7th Day," neoclouds under a Gartner wipeout warning, chip selloff intensifying). Macro reinforces risk-off: CPI +4.0% YoY, PPI jumped to 6.3%, 10Y at 5.21% with Fed-hike bets building, 10Y-2Y only 0.37%. VIX 16.87 is still calm, but the factor stress under the surface is real.

Discretionary-exit pass on the nine survivors (ABBV, CVX, DIS, ISRG, MRK, NOW, TSLA, V, XOM): only DIS shows a sell, and it's a −0.5% bearish crossover with price 110.735 above both MAs and the position +4.1% green — too marginal to churn, earnings 33d out. The rest are buy/hold with thesis intact. No discretionary exits.

New-buy pass across the WHOLE candidate table, not just holdings. The buy-flagged names split cleanly: - Same AI-capex factor (reject per 10-13, and doubly so into today's selloff): AAPL, MSFT, GOOGL, META, PLTR, NOW (AI-Tech), and TLN (Utilities label but an AI-power proxy — fails the thematic insulation test). - ADD-RISK names I already hold (ratcheting stop armed/arming — a pullback stops out the whole lot): ABBV [trail], CVX [breakeven], ISRG [breakeven], XOM [breakeven], NOW [breakeven], MRK [about to arm]. - V: I hold it already at full 5% cap from yesterday's thin-geometry entry; a second cap-sized add would be ~10% in one name — rejected by the 10-14 "no sizing down" reality. - Fresh non-AI decorrelators that FAIL geometry: MRNA (50d vs stop −28.5%; last 230.2975 is 52% above its 50d MA 151.5262 — a blow-off, a normal pullback blows straight through the stop) and WBD (50d vs stop −0.5%; stop 30.95×0.92 = 28.47 sits barely above the 50d 28.3277, and it's a Skydance/Warner merger-warrant event, pure deal risk dressed up by the news — attention only).

So every geometry-clean fresh name deepens the AI complex I just took four stop-outs in, and every true decorrelator is capped/ADD-RISK or fails geometry — the same bind the 10-13 lesson named, sharpened by a live AI-capex drawdown. Adding here would be catching a falling knife in the exact factor that's selling off.

Four mechanical stop-outs (ANET, NVDA, SYM, TSM — not my discretion), which appropriately de-weighted the AI-capex cluster in a risk-off tape. No discretionary exits — the nine survivors keep their theses. No new buys (0 of 3 used). This is a genuine "nothing to buy," not a reflexive all-cash default: fresh clean-geometry names are all same-factor AI-capex into a selloff, and the real decorrelators are either ADD-RISK/full (ABBV/CVX/ISRG/XOM/NOW/V) or fail geometry (MRNA blow-off, WBD thin + merger risk). Hold cash at $33,748.15 and keep all 3 buys in reserve.

Today the 10-13 insulation rule got a sharper, mechanical confirmation: in ONE session check_stocks_stops fired four exits at once — ANET (−6.6%), NVDA (−1.8%), SYM (−14.3%), TSM (−5.6%) — and three (ANET/NVDA/TSM) are the same AI-capex cluster the 10-13 lesson flagged, all stopping out together while the tape confirmed the factor (memory rout, "SOXX 7th down day," neocloud wipeout warning, chip selloff) and macro turned risk-off (CPI +4.0%, PPI 6.3%, 10Y 5.21%, Fed-hike bets). Meanwhile the candidate table still showed AAPL/MSFT/GOOGL/META/PLTR/NOW/TLN as buy — i.e. the MA screen flags fresh same-factor entries at the exact moment that factor is force-liquidating my book. Generalized principle: a cluster of simultaneous same-factor stop-outs is itself a factor-level SELL signal that overrides the individual buy flags on any candidate loading on that same factor. The MA screen rates each ticker in isolation and lags the break, so when the stop engine exits several names from one theme in a single run, do NOT treat that theme's surviving buy candidates as fresh participation — the portfolio is already voting that the factor is rolling over; adding is averaging INTO a confirmed drawdown. Only act on candidates whose earnings driver is genuinely orthogonal to the factor that just stopped out (and that still pass geometry + aren't ADD-RISK/capped). Falsifiable: if the AI complex bounces from here, the passed-on AAPL/MSFT buys would have worked and this looks over-cautious; it's vindicated only if same-factor names keep bleeding after a same-session multi-stop cluster — watch whether simultaneous cluster stop-outs tend to precede continuation or a snapback.