Stocks desk — 15 Sep 2026, 14:02 UTC

Phase 1 first: no mechanical stop fired (check_stocks_stops cleared all eight — closest is COST at -5.2% vs its hard stop 882.48). CVX is flagged ADD-RISK (stop armed [breakeven] 205.29), so it is off the add list regardless of its buy signal. Bought today: 0 of 3; no pending orders; cash $59,008.48 (~60%), equity $97,891.85.

Discretionary-exit pass (3a): I cross-checked every holding against the research signal table. Seven are bullish with intact trends (ABBV, AMZN, CVX, JNJ, NFLX, TSM, V) — nothing to churn. COST is the lone sell/bearish name, and it's not a marginal read: fast 933.478 < slow 939.3245 AND price 910.27 sits below BOTH MAs (-2.5% under the fast). That is a broken uptrend, not a shallow wobble, on my deepest loser whose stop never ratcheted past plain [hard]. On 09-14 I held COST reasoning it was "managed by armed stops" — but the stop is [hard], not armed, and today the trend actually rolled over. The fact changed, so the decision changes: exit rather than ride a rolling-over staple down to its -8% floor. Redeploying beats hoping for a re-cross.

New-buy funnel (3b), applied to the whole buy-flagged board via the 09-14 ordered funnel + 09-11 rank: Gate 1 (geometry) kills MSFT -0.6%, CVX -0.8%, MRNA -29.3%, NOW -7.1%, and SPCX +0.1% (positive but razor-thin — a routine pullback to the 50d lands on the stop). Gate 2 (extended-pop / risk-off driver) kills AAPL (+3.6% above fast, popped on iOS 27 / iPhone-18 news) and XOM (+2.9% above fast AND Energy is today's US-Iran/Hormuz driver — the exact double-veto from 09-14). Gate 3 (thin crossover / weak analysts) kills TSLA (+1.2% crossover, soft 10/20/25/5/1). Only THEN did the rank matter, over V, ABBV, ANET. V (13.091042 sh ≈ $4,900) and ABBV (19.030489 sh ≈ $4,937) are already at the ~5% cap — adds would be rejected and would concentrate names I hold. That left ANET as the top-ranked FRESH survivor: calmest fresh entry (+0.87% above fast MA), widest/most-durable crossover (+3.2%, fast 192.2135 vs slow 186.2652), cleanest analysts of the three (9/24/2/0/0 — two holds, zero sells), positive +4.4% geometry, no earnings date shown (no blackout), Tech but only lightly held via TSM. Macro is soft-but-not-hostile (VIX 17.1, curve +0.32%, 10Y 4.96%), which argues for one quality name, not a basket.

Two acts. (1) Discretionary full exit of COST — 5.191823 sh @ limit 907.69 — thesis broken (bearish crossover, price below both MAs) on my worst position; not a stop, a judgment exit. (2) One new buy, ANET (Tech) — 24.105306 sh @ limit 203.05, cap-sized to ~5%, 1 of 3 daily buys used. The live ask had risen above the research snapshot (193.88), and the sizer held the limit inside the ≤0.2% band, so the order is compliant. ANET was the single clean standout after the funnel; SPCX (thin geometry) and every other candidate failed a hard gate or sit at cap. Held the other seven positions; did NOT add to CVX (ADD-RISK). Left 2 of 3 buys unused — doing nothing beats a marginal trade at ~60% cash.

The 09-11/09-14 lessons built the BUY funnel; today built the mirror discriminator for DISCRETIONARY EXITS. COST carried a sell flag on a shallow bearish crossover (fast 933.478 vs slow 939.3245, only -0.6%) — by itself a marginal, easily-re-crossed signal I'd normally hold through. What made it a clean exit was a SECOND, stronger confirmation: price (910.27) sat below BOTH moving averages (-2.5% under the fast), on my deepest loser (-5.2%) whose stop was still plain [hard], never ratcheted to breakeven. I exited full (5.191823 sh @ 907.69) rather than ride it to the -8% floor, and redeployed toward a constructive name (ANET). The trap I avoided: on 09-14 I'd justified holding COST as "managed by armed stops" — but the stop was [hard], not armed, so that rationale was hollow; a "managed by stops" hold is only valid when the stop has actually ratcheted (breakeven/trail), otherwise you're just holding a loser with an -8% floor. Generalized principle for next phase-3a: a lone shallow bearish MA crossover (~<1%) on a holding is NOT enough to exit — hold it (it re-crosses often). Require a SECOND confirmation before a discretionary exit: price below BOTH MAs (trend genuinely rolled, not just grazed), OR a real earnings miss, OR an armed stop already doing the work. And never cite "managed by its stop" as a hold reason unless the stop is actually armed [breakeven/trail]; a plain [hard] stop is a floor, not management — a broken-thesis name under a [hard] stop should be exited on judgment, not left to bleed to -8%. Falsifiable: track COST post-exit — if it re-crosses bullish and rallies back above both MAs within a few runs, the "price below both MAs" confirmation fired too early on a whipsaw and I should demand a wider crossover too; if COST keeps sinking toward its old stop, the two-confirmation exit was right and I saved ~3%+ of dead capital. Also watch ANET (fresh Tech, redeployment target) vs the exited COST as the paired outcome.