Stocks desk — 17 Sep 2026, 14:02 UTC

Phase 1: 8 open positions, $59,044.02 cash (~60%), 0 of 3 buys used, no pending orders. Two live lessons framed the run. The 09-15 lesson set a two-confirmation bar for discretionary exits (a lone shallow bearish crossover isn't enough — require price below BOTH MAs, or a real earnings miss, or an armed stop). The 09-16 lesson graded stop-vs-50d geometry: a barely-positive cushion (~≤1-2%) is a soft veto, and compounds with volatility, imminent macro events, and sector concentration.

Phase 3a — exits. check_stocks_stops fired nothing; ANET (breakeven) and CVX (breakeven) are ADD-RISK. Every holding carries a bullish MA trend and none is flagged sell in the research. My deepest loser NFLX (-5.3%) is the one to test against the exit lesson: its trend is still bullish (fast 79.375 > slow 75.8078), price 76.16 sits below the fast MA but ABOVE the slow — so NOT "below both MAs," earnings aren't imminent (Oct 19, 32d), and there's no sell flag. That fails the two-confirmation test; it's a pullback inside an intact trend, not a broken thesis. Same read for AMZN (-3.9%, bullish hold) and V (-2.9%, bullish). No discretionary exits — no churn without a stated reason.

Phase 3b — buys. From the WHOLE table's buy flags, the geometry gate + ADD-RISK culled most: META (-1.7%), MRVL (-4.6%), MRNA (-34.2%), SPCX (-5.2%), SMCI (-8.1%), PLTR (-3.0%), NOW (-4.4%) all fail on negative geometry; ANET (+0.1%, held, ADD-RISK) and CVX (ADD-RISK, held) are out. That left positive-geometry fresh names: AAPL (+5.0%), NVDA (+6.1%), MU (+2.5%), SNDK (+1.4%), TSLA (+3.3%). Working the four framework questions: cash is ample ($59,044.02); open positions already include Tech (TSM, ANET) so a third Tech add raises concentration — the one real knock. Fast/slow: NVDA's +2.2% crossover (218.5915 > 213.8764) is genuine, not the shallow +0.5% of AAPL; price 219.02 is above its fast MA (constructive). Downside: NVDA's +6.1% stop-vs-50d is the THICKEST cushion on the buy list — the deliberate opposite of the 09-16 MRVL thin-geometry pass; the routine-pullback risk that vetoed MRVL simply isn't present here. Analysts 24/41/3/1/0 are the cleanest book on the board, no earnings blackout.

Weighing the 09-16 graded rule against NVDA: of its three compounding factors, geometry is strongly favorable (+6.1%), the Fed event that loomed on 09-16 has now been delivered (25bps hike, another projected — a headwind for high-multiple tech, but resolved rather than pending), and only sector concentration cuts against. One soft factor alone is not a veto. Macro is a genuine tightening tape (10Y 5%, CPI +3.7%), but VIX is calm (17.71) and I'm sizing one position at ~5% while holding ~60% cash. AAPL (shallow crossover, 15 holds/3 sells/1 sS), TSLA (25 holds/5 sells/1 sS, Consumer overlap with AMZN), MU and SNDK (thinner geometry, more semis piled on TSM) were all weaker. NVDA is the one clean standout — the case the funnel is built to catch.

One new buy: NVDA (Tech) — 22.350346 sh @ limit 219.35, cap-sized to ~5% ($4,902 of $98,046.15 equity), 1 of 3 daily buys used. It was the single clean standout: thickest geometry on the buy list (+6.1% stop-vs-50d), genuine +2.2% crossover with price above the fast MA, the cleanest analyst book on the board (24/41/3/1/0), and no earnings blackout — the deliberate mirror of the 09-16 thin-geometry pass. Held all 8 existing positions: none flagged sell, none with a broken thesis; NFLX/AMZN/V drawdowns sit under intact bullish trends and fail the two-confirmation exit bar. Did NOT add to ANET or CVX (both ADD-RISK). Left 2 of 3 buys unused and ~$54K cash — one high-conviction, well-cushioned name beats spraying marginal semis into a tightening tape.

The 09-16 lesson taught geometry as a SOFT VETO (reject thin stop-vs-50d into event/concentration); today ran its mirror and showed the same graded metric works as an active SELECTOR when several names clear the gate. The buy funnel left five positive-geometry fresh candidates — AAPL (+5.0% geom but shallow +0.5% crossover, 15H/3S/1sS), NVDA (+6.1% geom, genuine +2.2% crossover, 24/41/3/1/0), MU (+2.5%), SNDK (+1.4% thin), TSLA (+3.3% but 25H/5S/1sS). Rather than spread buys or default to nothing, I ranked on geometry-cushion × crossover-depth × analyst- quality and took the single best (NVDA, cap-sized ~5%), leaving 2 of 3 buys and ~$54K cash. The hard call was that NVDA is a THIRD Tech name (already hold TSM, ANET) — sector concentration, the 09-16 rule's third compounding factor. But I let strong geometry OUTWEIGH that lone soft factor because the other two factors were favorable: +6.1% is thick cushion (not the thin trap that vetoed MRVL at +0.9%), and the Fed event that loomed on 09-16 is now DELIVERED (25bps hike done, another projected — a resolved headwind, not a pending whipsaw). Generalized principle: the graded-geometry rule cuts both ways. (1) When multiple candidates clear the geometry gate, don't buy the widest- crossover name reflexively and don't freeze — rank the survivors by cushion depth AND crossover depth AND analyst quality together, and concentrate into the single cleanest rather than spraying marginal adds. (2) A lone adverse soft factor (sector concentration alone) does NOT veto a name whose geometry is genuinely thick (~≥5%) and whose macro event has already resolved; the 09-16 pass required TWO+ of {thin geometry, volatility/event, concentration} to compound. One factor is a note, not a gate. Falsifiable: track NVDA vs the passed-over AAPL/MU/TSLA over the next several runs — if NVDA holds its crossover and outperforms them, the geometry-led selection was right; if NVDA rolls over (Tech concentration + the projected second hike biting high-multiple names) while a passed name holds, I over-weighted geometry against the concentration/tightening risk and should demand geometry AND a non-overlapping sector before adding a third name to a cluster.