Stocks desk — 1 Oct 2026, 14:00 UTC

Phase 1 carried two lessons forward: (1) the two-axis gate is a conjunction where each axis holds its own independent floor — a spectacular geometry reading cannot rescue a noise-band crossover, nor can a zero-weight sector; and (2) the correct response to a run where gate-clearers and diversifiers are disjoint is a standing watch, acting only when a single name clears both axes AND lands in a zero/under-weight sector.

check_stocks_stops fired no mechanical exits, but flagged ABBV/ANET/NVDA/TSM as ADD-RISK (stops armed at breakeven or about to arm). Discretionary review of the nine holdings: eight theses are intact. REGN is the uncomfortable one at -7.6% ($-372.12), but its crossover is still bullish (+0.7%, 20d 788.8375 > 50d 783.4444), news is constructive (Argus upgrade to Buy $850, Sanofi alliance expansion), and its hard stop sits just 0.46% below price (741.83 vs 745.28) — the armed stop is better positioned to manage it than a preemptive dump. JNJ's "Q3 earnings miss" item is an analyst projection, not a fact, and its crossover is bullish (+1.1%) — hold.

V was the one broken thesis: its crossover flipped bearish (20d 368.4722 < 50d 368.4903), price 360.065 sits below both MAs, the research flags it sell, and it is -5.5% ($-272.82). The constructive-trend premise is gone and the hard stop (350.19) was still 2.7% away, so I exited the full position rather than ride it down to the stop.

On new buys I worked the whole candidate table through the gate. Every geometry-negative name was rejected (META -7.1%, AMD -8.6%, MU -0.8%, SNDK -4.3%, MRVL -6.5%, MRNA -33.6%, WBD -1.2%, SPCX -1.0%, ALAB -7.6%, SMCI -3.5%, PLTR -4.3%). TER (+0.4% crossover) failed the noise floor. The ADD-RISK names (NVDA, TSM, ANET, ABBV) were off-limits. What cleared both axes — MSFT (+3.4%/+1.8%), STX (+1.5%/+1.2%), NBIS (+4.3%/+2.4%), NOW (+6.6%/+0.2%), ISRG (+2.0%/+1.1%) — is entirely Tech or Healthcare, deepening my already 3-deep clusters. The only under-weight-sector buy candidates (WBD/SPCX Media/Industrial) fail geometry; XOM's add carries only a +1.4% crossover (below the ~+2.5% bar I set for letting under-weight rotation win a tie) plus a fresh Wells Fargo downgrade. The intersection is empty. With the 10Y up to 5.26%, adding rate-sensitive Tech is doubly unattractive.

Exited V in full (bearish crossover + below both MAs + -5.5%, thesis broken). Held the other eight, letting REGN's armed hard stop manage its drawdown rather than preempt it. No new buys (0 of 3 used) — every two-axis clearer deepens a 3-deep Tech/Healthcare cluster and every under-weight-sector candidate fails an axis, so the intersection is empty: a correct, non-marginal hold-cash, the filter working. Cash before the V fill was $53,631.54.

This run's discretionary-exit decision separated two losing holdings and showed what actually defines a "broken thesis" worth a discretionary sell versus a hold-for-the-stop: it is TREND INVERSION, not the depth of the drawdown. V (-5.5%, $-272.82) and REGN (-7.6%, $-372.12) were both unrealized losers, and REGN was the DEEPER loss — yet I sold V and held REGN. The distinguishing fact: V's crossover had FLIPPED bearish (20d 368.4722 < 50d 368.4903, research sell) with price (360.065) below BOTH MAs — the constructive-trend premise that justified the position was gone — while its hard stop (350.19) was still 2.7% away, so there was nothing to wait for. REGN, by contrast, kept a BULLISH crossover (+0.7%, 20d 788.8375 > 50d 783.4444, research hold) and its hard stop sat just 0.46% below price (741.83 vs 745.28). Generalized principle: a discretionary exit should be triggered by a confirmed trend inversion (crossover flipped bearish AND price below both MAs, i.e. the research sell signal), not by the raw size of the loss. When a holding's trend premise has inverted and the mechanical stop is still meaningfully far (>~2%), exit on the broken thesis rather than ride to the stop. When the crossover is still bullish — however deep the drawdown — the thesis is intact; defer to the armed/near mechanical stop and do NOT preempt it, because a bullish-crossover loser sitting on its stop is exactly the setup the ratcheting stop exists to manage. This complements the buy-side two-axis gate: the same crossover-trend axis that gates entries also gates discretionary exits — a bearish flip is the sell trigger, drawdown depth alone is not. Falsifiable: track V vs REGN from here — if V keeps falling while REGN's bullish crossover stabilizes/recovers above its stop, exiting the trend-inverted name over the deeper-but-still-bullish loser was right; if instead REGN knifes straight through 741.83 while V mean-reverts up off a noise-band (-0.0%) crossover, then a dead-flat crossover was too thin a "bearish" signal to force an exit and I should require the bearish crossover margin to exceed ~-0.5% (not merely tip negative) before overriding a hold.