AI · US Tech — Trades

Dell is the book's highest-conviction print (88) and the only position carrying a real gain (+23%); Friday's 12% jump to an all-time high came from RBC initiating coverage at Outperform with a target, on top of record Q2 results (revenue $47bn, +58% y/y) and a $95bn AI-server order backlog, and the shares still trade near 21x forward earnings. We add to a winner on its own breakout rather than chase it: the order is a limit at the model's entry price, so it only fills if the stock comes back to us. OPERATIONAL: GTC buy-limit 564.47 (range 555.03-573.91, Entry_Price +/- 0.25 ATR); expires 2026-09-16. Sizing 713 x 112.34% = model -> x0.75 size modulator (MoS -2.48) = 601 -> capped by max_position_pct 60% headroom ( / 564.47) = 391 -> x0.5 STATED CHASE HAIRCUT = /. The haircut is a deliberate deviation from full model size, not a signal override, on three grounds: (i) UT-10 - this book has ADDed to DELL twice before (07-@411.73, 07-@419.03) and BOTH graded stop_hit; (ii) own-book BreakoutUp_52W|buy is 0-for-2 on 1wk and 4wk (n=2, report-only, cited not applied); (iii) L1 - BreakoutUp_52W 4wk targets overshoot +55-75% and mean-revert, which is why the +10.2% 1wk interpolation below should be read as an upper bound. L6 is applied in SPIRIT not letter: this is an analyst-initiation blowout, not a post-earnings gap, but a one-session +11.98% to an all-time high is structurally the same chase risk - hence the limit rather than market-on-open. Stop RAISED 469.94 -> 491.76 (2.0x ATR chandelier off the 567.29 close) - the ratchet moving up, which also protects the add. 12wk 808.39 exceeds the 446.05 DCF anchor by a wide margin - projection assumes growth beyond what is priced in. Macro caveat stated: August CPI printed core +0.3% and markets now price ~87% odds of a 25bp FOMC HIKE on 09-16, two sessions after the execution date, with triple witching 09-18 behind it. · news

decided @ 564.47 · stop 491.76 · t1w 622.06 · t4w 756.44 · t12w 808.39

A measured pyramid onto a one-day-old winner that the news flow has since validated. Micron gapped +6.1% as the NAND/DRAM pricing cycle re-accelerated, with Dell's record $95bn AI-server backlog cited as a direct demand driver - which is notable because Dell is already this book's largest holding, so the same catalyst is now paying twice. Micron carries the book's #1 quality rank and its 12-week projection points 11.6% higher. The add is kept small and risk-defined because conviction printed flat rather than rising. Raw 76, val_adj -6 (MoS -1.16) -> adjusted 70. Held @ 971.06 (+4.69%, filled at this morning's open). ADD sizing (held-name path): 50.71% x = 40.6, x0.75 size mod = =. Entry range 1005.27-1027.91. Stop 969.00 = -4.68%; 4wk 1107.13 = +8.91%, R:R 1.90:1. 12wk 1134.12 exceeds DCF anchor 772.94 - projection assumes growth beyond priced-in. ASSUMPTION STATED: the scan CSV was written at 06:00:17 before the 06:00:35 executor fill, so it prints MU as unheld (Held=N) with a fresh-entry; per UT-7 the 80-share holding from input/ustechportfolio.json governs and the conservative ADD formula is used rather than the fresh-entry. L6 check passed: MU FQ4 earnings are 09-30, so this is a sector re-rating not a post-earnings gap; ATR expanded only +8.6% (41.7->45.3), well short of the UT-10 spike bar. L7: MU also signals in us/ussemicon - execute once across books. · news

decided @ 1016.59 · stop 969.00 · t1w 1043.75 · t4w 1107.13 · t12w 1134.12

TSMC is the highest-conviction fresh signal in the book: it broke out of a three-session stall to a Buy_Rank #1 STRONG BUY on the back of a monthly sales update showing robust AI demand, record 2026 capex guidance of -64bn and full-year revenue growth above 40%. The technical setup is clean rather than stretched - the close sits in a perfect EMA10>20>50>200 bull stack with RSI 57, so this is an entry into strength, not a chase. Valuation is the offset: at a deeply negative margin of safety the position is deliberately taken at half size. Raw conviction 86, val_adj -10 (MoS -13.82) -> adjusted 76. Size mod 0.5x (MoS <= -3). Sizing: 0.6695 x equity x 20% buy_pct =, x0.5 = / 428.91 = =. Entry range 426.50-431.32 (+/-0.25 ATR). Stop 419.45 = -2.21% (0.98 ATR); 4wk target 448.19 = +4.49%, R:R 2.03:1. CAVEAT: Target_12Week 416.00 is BELOW the entry (-3.01%) - treat this as a 4-week swing, not a 12-week hold, and it is a second reason the size is halved. DCF_per_share 46.21 is a unit-mismatch artifact on the ADR (systemic finding 2), so the DCF anchor annotation is not meaningful here. Book has sold TSM twice before (07-@414.91, 08-@409.54); re-entry is 3.4% above the last exit. L7: TSM likely also signals in us/ussemicon - execute once across books. · news

decided @ 428.91 · stop 419.45 · t1w 434.69 · t4w 448.19 · t12w 416.00

We are cutting this position in full. The company announced a million at-the-market equity offering on top of the million already raised - open-ended dilution that the market read immediately, sending the shares down 12% even against a record quarter. The chart agrees: the stock has closed below its risk line for four straight sessions and below its 200-day average for four, with every shorter moving average stacked bearishly above it, and our own twelve-week model now projects 43% lower from here. A record quarter funded by issuing stock into a falling price is not a thesis we want to hold through. Operational: STRONG SELL - REDUCE cv76 (rising -> 76), Sell_Rank #1, engine 90.04 with rationale 'target full exit'. Close 100.38 vs recorded stop 111.93 = -10.32% (deepest yet, 4th session); 4th consecutive close below EMA200 (106.26) with EMA10 109.70 / EMA20 114.95 / EMA50 122.95 stacked above; Target_12Week 57.22 = -43.0%; VWAP_Distance -22.26%. Both L2 legs armed 4 sessions. Realized - (-33.09%) vs 150.03 cost; proceeds. This is a LOSS-CUT not a profit-take - the PROFIT_ARMED print is the arm-on-loss mislabel and is INERT under UT-9 (Gain_ATR -6.43). TENTH re-fire under L2+L3+UT-3/UT-11. CRITICAL: execution verb corrected to 'market' - the ten prior AAOI exit cards were written 'market_on_open', which paper_executor.py _EXEC_KINDS (line 112) / _eligibility (line 554) rejects BEFORE the skip-log path, making them invisible; ledger tally is a perfect separation, ('sell','market') 5 cards -> 5 fills, ('sell','market_on_open') 22 cards -> 0 fills ever. Estimates: 1wk/4wk n/a - the REDUCE row prints a blank Target_Price, no number synthesized. SCORECARD CAUTION stated not buried: RallySell_BearTrend|sell is n=4 at 0.00/0.00 on this book's own data - every prior exit in this family looked early at both horizons; report-only so nothing is adjusted, and the exit is taken anyway because a -10.32% broken stop with a -43% forward view and open-ended dilution is exactly the case L2 exists for. Valuation originates nothing (Val_Signal NA, FV_per_share -3.38). · news

decided @ 100.38 · stop 111.93 · t12w 57.22

We are taking a third of this position off the table on target exhaustion, and holding the rest. The stock has run to within 9% of our own twelve-week projection while momentum has started to diverge from price - the classic signature of a move that has done most of its work - so we bank a slice and raise the risk line under the remainder. This is a risk trim, not a change of view: Nvidia just announced a roughly billion acquisition of Hugging Face, brokers sit at 46 Strong Buys out of 51, and every forward horizon we model is still positive. That is precisely why the trim is 30% and not more. Operational: ADR-0012 profit-take, ARMED + one tick FIRED. Armed = PROFIT_ARMED printed with T12_Progress_Pct 91.34 (>=90); Gain_ATR +0.21 is POSITIVE so UT-9 does NOT make the arm inert (contrast AAOI at -6.43 where it does). Fired = tick (a), a second Exit_Warning code RSI_DIV on the same row. Ticks (b)/(c)/(d) all checked, none fires: conviction ROSE; Target_12Week 250.12 > 1.02 x close (233.02); row is HOLD not NO TRADE. G1 softening does NOT reach it - it protects fresh BreakoutUp_*/MomentumContinuation_* at conviction >=80 and this signal is AnticipationUp_VCP, outside those families - so the trim is MANDATORY, not optional. One tick -> profit_take_1 = 30%. Valuation amplify does NOT apply (MoS_FV -2.400 is not <= -3), portion stays 30%. 168 x 0.30 =, proceeds, realized + (+0.73%). Residual, stop RAISED 213.25 -> 215.33. Estimates rendered as upside forgone (f=0.30 per L5); 12wk exceeds DCF anchor 163.98. TENSION RECORDED: this banks on a position 0.21 ATR above cost, armed entirely by the T12 exhaustion branch not by gain magnitude; the identical card was issued 08-28 (SELL @227.98) and withdrawn 08-29 when Gain_ATR flipped negative. The pending 08-29 PROPOSED block (X2) proposes Gain_ATR >= 1.0 to arm, which would suppress this card - PROPOSED is inert by design and is NOT applied. Watch NBIS warrant lock-up 09-11 (~21M shares). · news

decided @ 228.45 · stop 215.33 · t1w 243.64 · t4w 279.08 · t12w 250.12

A measured pyramid onto a one-day-old winner that the news flow has since validated. Micron gapped +6.1% as the NAND/DRAM pricing cycle re-accelerated, with Dell's record $95bn AI-server backlog cited as a direct demand driver - which is notable because Dell is already this book's largest holding, so the same catalyst is now paying twice. Micron carries the book's #1 quality rank and its 12-week projection points 11.6% higher. The add is kept small and risk-defined because conviction printed flat rather than rising. Raw 76, val_adj -6 (MoS -1.16) -> adjusted 70. Held @ 971.06 (+4.69%, filled at this morning's open). ADD sizing (held-name path): 50.71% x = 40.6, x0.75 size mod = =. Entry range 1005.27-1027.91. Stop 969.00 = -4.68%; 4wk 1107.13 = +8.91%, R:R 1.90:1. 12wk 1134.12 exceeds DCF anchor 772.94 - projection assumes growth beyond priced-in. ASSUMPTION STATED: the scan CSV was written at 06:00:17 before the 06:00:35 executor fill, so it prints MU as unheld (Held=N) with a fresh-entry; per UT-7 the 80-share holding from input/ustechportfolio.json governs and the conservative ADD formula is used rather than the fresh-entry. L6 check passed: MU FQ4 earnings are 09-30, so this is a sector re-rating not a post-earnings gap; ATR expanded only +8.6% (41.7->45.3), well short of the UT-10 spike bar. L7: MU also signals in us/ussemicon - execute once across books. · news

decided @ 1016.59 · stop 969.00 · t1w 1043.75 · t4w 1107.13 · t12w 1134.12

Palantir's AI platform keeps winning enterprise distribution - it rose 7.9% on an expanded PwC US partnership that opens another channel into large corporate, on top of a business compounding at a 155% Rule-of-40 with 157% net dollar retention. The technical setup is a pullback-to-VWAP buy at strong conviction rather than a chase, so we queue a limit below the market and let the stock come to us; the offset is that the shares are expensively priced against any intrinsic anchor, which is why we size at half. Operational: adjusted conviction 86 -10 (val_adj, MoS_FV -13.405 clamped at floor) -5 (scorecard_bias, PullbackBuy_VWAP|buy n=5 at 0.25/0.25, newly ACTIONABLE) = 71, clears the 65 floor by 6. Quality_Pass True, Quality_Rank 6/31 - gate clears. Entry_Distance_ATR -0.68 (extended past the limit) so QUEUED not bought; action price is the anchor snapshot's own Entry_Price, no carried-forward level. Size 0.6695 x 1,016,240.86 equity x 20% buy_pct =, x0.5 size modulator (MoS_FV <= -3) = / 175.81 =. Risk if stopped 387 x 9.87 = = 0.38% of equity. Estimates use f=0.30 per L5 (calib_1w -0.141, degenerate). 12wk 260.11 exceeds DCF anchor 46.62 - projection assumes growth beyond priced-in. expires_on 2026-09-09 = 3 trading sessions from the 09-04 anchor (09-07 Labor Day). Re-arm trigger if expired: a fresh PullbackBuy_VWAP print with adjusted conviction >= 65. Countervailing view noted: Michael Burry publicly short-biased, overvaluation critique is exactly what val_adj -10 encodes. · news

decided @ 175.81 · stop 165.94 · t1w 180.22 · t4w 190.50 · t12w 260.11

We are placing a small buy order in Apple that only triggers if the shares keep rising. Apple was one of just three stocks in the book to close higher on a day when twenty-seven fell, on the day John Ternus formally took over as chief executive, and it is the only high-quality name in the portfolio currently showing a genuine buy setup. His first public outing is the iPhone 18 Pro and first foldable iPhone launch on 9 September, five trading days away. We are deliberately buying only half our normal amount, because analysts' average price target is essentially where the stock already trades - the setup is good, the starting price is not cheap. OPERATIONAL: ANTICIPATE slot (4th slot, one per book, NOT one of the <=3 ACQUIRE). Eligibility satisfied on the letter of the rule - last three snapshots BUY-ANTICIPATE 83 -> WATCH 55 -> BUY-ANTICIPATE 83 =; conviction 83 well above the 55 slot floor; the 08-29 card pre-committed to exactly this trigger. Quality_Pass TRUE, Quality_Rank 3/31 - clears the gate that blocked SNOW/SAP/MSFT. Entry: GTC buy-stop @323.17, entry range 321.17-325.18 (+/- 0.25 x ATR 8.02). Close 325.13 sits ABOVE the stop (Entry_Distance_ATR -0.24, inside the -0.25 no-chase bar) so it should fill at/near the 09-02 open. RSI 61.1, VWAP_Distance +4.47% - not extended. Trend 94 / Mom 50 / Vol 33, Consolidation_Bullish. Sizing: Portion 50.00% x equity x buy_pct_of_portfolio 20% =; x0.75 size modulator (MoS_FV -2.975 in the -3<MoS<=-1 band, from the REGENERATED fairvalue file); x0.5 anticipate-slot factor =; / 323.17 = 110.07 -> = = 3.75% of equity. max_position_pct 60% and both non-binding. Stop 299.10 = -7.45% / ~3.0 ATR; risk = 0.28% of equity. Estimates f=0.096 (Step 7 fallback - CSV has only a single Est_Low/High band, no ADR-0013 Est_*_Point columns). 12wk 332.66 EXCEEDS DCF anchor 237.39 - projection assumes growth beyond priced-in. WARNING - forward-curve inversion: 4wk 371.31 sits ABOVE 12wk 332.66; the model expects the move given back. Counterfactual argues against and is why size is halved: anticipate n=8, median 1wk -0.73% / 4wk -8.81% ['saved']. Scorecard cell AnticipationUp_VCP|buy n=1 hit_1w 0.00 - report-only, no bias. Catalyst: 09-09 event is 5 trading sessions out (09-07 Labor Day holiday) - ON the <=5-day boundary. expires_on 2026-09-08 = anchor + 3 trading sessions. · news

decided @ 323.17 · stop 299.10 · t1w 327.79 · t4w 371.31 · t12w 332.66

Salesforce delivered the quarter that settles the argument about whether AI helps or hurts it: profits came in 80% above expectations, management raised full-year earnings guidance by 18.5%, and its AI products are now a billion recurring revenue line growing over 200% a year, alongside a new partnership embedding Anthropic's Claude across Slack and Salesforce. The shares jumped 23% and have kept climbing since. This is one of only two names in the book our valuation work still calls fairly priced rather than expensive, and the only one whose 12-week projection sits inside its own discounted-cash-flow anchor. We buy at half size, and only on a further move up. OPERATIONAL: QUEUED as the model's own Buy Stop @257.66 (Entry_Distance_ATR +0.11) - it fills ONLY on continued strength above Friday's range, self-limiting in the same way the anticipate-slot buy-stop is. REPRICED 213.78 -> 257.66 per the Step 7 provenance rule: the 08-28 GTC limit was placed for L6 gap discipline, CRM never traded within 15% of it, and the rule explicitly covers a previously-queued name that has run past its limit AND is overbought (RSI 80.79 > 75) - reprice to the live Entry_Price or demote to WATCH. Repriced, not republished stale. L6 WEIGHED AND ANSWERED: the 08-26 +22.58% gap is an earnings-power re-rating (adj EPS vs est = 80% beat; FY guide +18.5% to), not a sentiment pop of the UT-1/MU kind, and the stock has HELD and extended (+1.82%, then +1.57% on a -1.71%-median tape). Cautions priced in via size: RSI 80.79, VWAP_Distance +20.31% (genuinely extended), and UT-12 folds MomentumContinuation_Up|buy into L1's discount-on-ties list (own-book cell n=2 at 0/2 on 1wk, median err -4.19%). SIZING: Portion 66.95% x equity x buy_pct 20% =; x0.50 =; / 257.66 = 252.48 -> = = 6.68% of equity. THE MODULATOR IS 0.50, NOT THE FILE'S 0.75, AND THIS IS STATED NOT SILENT: fairvalue_T0 prices CRM at 209.06 vs a 256.00 anchor close (+22.5% stale), and MoS_FV recomputed live (256.00 / FV_per_share 59.55 = P_FV 4.30 -> MoS -3.30) crosses the <= -3 tier. Quality gate PASSED (Quality_Pass True, F 7, Quality_Rank 15/31). Stop 211.73 is WIDE - -17.8% / ~3.0 ATR, chandelier_trail on a post-gap name, sitting just above the pre-gap base; correct invalidation level but of risk on - flagged deliberately. Entry range 253.83-261.49 (+-0.25 x ATR 15.310). Estimates (f = 0.096): 1wk 266.48 (+3.42%), 4wk 349.52 (+35.66%), 12wk 364.80 (+41.58%) - NO DCF annotation, DCF_per_share 379.39 is ABOVE the 12wk target, unique among this book's buy candidates. Expires 2026-09-02; withdraw automatically if conviction falls below 65. · news

decided @ 257.66 · stop 211.73 · t1w 266.48 · t4w 349.52 · t12w 364.80

Applied Optoelectronics is the physical layer of the AI data centre, and this quarter it stopped being a promise: record Q2 revenue of (+86% year-on-year), the first return to non-GAAP profitability, and a Q3 guide of -290M that implies 130% growth, with management saying manufacturing capacity - not customer demand - is what limits them through mid-2027. First 1.6T qualification with a major hyperscaler completes within weeks and a + order ships from late Q3. We buy it on a fresh 20-day breakout, but on a limit rather than at the close, because the shares rose 15.5% in a single session and trade 25% above their volume-weighted average - we want the position at our price, not the crowd's. Half the normal size: the company is still unprofitable on a statutory basis, the stock has already trebled this year, and our own valuation model cannot put a positive intrinsic value on it. OPERATIONAL: L1/UT-6 no-chase - GTC limit at the model Entry_Price 150.03, entry range 145.71-154.35, unfilled on a gap-up by design; expires 2026-08-19 (3 sessions), withdraw or reprice at each EOD. L6 checked and NOT applicable (Q2 print was 2026-08-06, eight sessions prior). Size modulator 0.5x applied by stated assumption: Val_Signal NA arises from a NEGATIVE FV_per_share (-3.38) with F_Score 4, read as the MoS_FV<=-3 tier rather than neutral 1.0x; full would have been (10.1% of equity, 3.0% at defined risk). Sizing arithmetic: 50.71% x x 20% x 0.50 = / 150.03 = = (5.07% of equity); defined risk = 1.49% of equity. WARNING: Target_12Week 148.34 sits BELOW the action price while Target_4W is +58.6% - treat as a 1-4 week trade on a hard chandelier trail at 106.06, not a hold. All four graded buy families are 0-for-8 on 4-week targets. Scorecard bias 0 (no actionable cell, UT-8); f=0.134. · news

decided @ 150.03 · stop 106.06 · t1w 161.81 · t4w 237.96 · t12w 148.34

Nvidia set up for a second consecutive session at the highest conviction this book has ever recorded (96, up from 93), and analysts have just raised their three-year earnings growth expectation from 33% to 44% a year on the view that hyperscaler AI spending was under-modelled - with a median price target of across 65 analysts. Crucially the stock did not gap on that news, it rose only 0.5%, so there is a real setup to buy rather than a spike to chase; and unlike every other buy candidate in the book today, Nvidia passes the quality screen outright at rank. We commit only on strength, at half the normal anticipation size, with risk capped at a third of a percent of the book. Operational: ANTICIPATE slot ARMED - eligibility verified 2-of-3 (08-12 WATCH cv54, 08-13 ANTIC cv93, 08-14 ANTIC cv96) and cv96 clears the 55 slot floor by 41. GTC buy-stop at the live snapshot Entry_Price 225.76, ED +0.07 = +0.20% above the 225.30 close - provenance-clean, no chase, and NEVER at the expired 08-08 level of 224.39. Sizing: 0.50 x equity x 20% =, x0.75 size modulator (MoS -2.52 in the -3..-1 band) =, x0.5 anticipate slot = -> = (3.73% of equity) against; caps non-binding. Defined risk 168 x (225.76-206.05) = = 0.33% of equity. 1wk uses f=0.30 per L5 (calib_1w 0.134, 9 graded buys below the 10 floor, 12th session). Both 4wk 265.16 and 12wk 252.32 exceed the DCF anchor 163.98 (by 61.7% and 53.9%) and the 12wk sits BELOW the 4wk - forward-curve inversion, second occurrence on this ticker, so read the 4wk as the trade objective. Horizon: queue expires 08-18, AHEAD of 08-19 FOMC minutes, 08-21 opex and the 08-26 Q2 print - a clean window by construction. Bear counterpoint on the record: Burry's vendor-financing critique of the ~$500bn chip-buyer financing consortium, which is why this is 3.7% and not 7.5%. AnticipationUp_VCP|buy is n=1 in this book's scorecard (report-only, bias 0). Auto-withdraw if conviction <65, on expiry unfilled (reprice from the then-current snapshot), or on a settled close below EMA20 212.37. · news

decided @ 225.76 · stop 206.05 · t1w 237.58 · t4w 265.16 · t12w 252.32

Dell broke to a 52-week high, closing up 9.9% at after Super Micro guided FY2027 revenue to -72bn against a 5bn consensus - the market read that as the AI-server pie growing rather than Dell losing share, directly overturning the share-loss thesis that had knocked the stock down the day before. With a $43bn AI-server backlog and AI revenue up 757% last quarter, the position is now +7.9% and the model flipped from HOLD to BUY-ADD as conviction jumped. We are adding on further strength only, at half the model's size, because the shares have already converged onto our own DCF anchor and the late-August margin print is the unresolved risk. Operational: GTC buy-stop above the session high 484.71, so no fill without a new high and no cash committed otherwise; = 63.29% x 542 held = 343, halved by the MoS -3.02 size modulator; from. Position stop stays at the 444.14 ratchet for all post-fill (never lowered, and deliberately NOT raised - 1.14x ATR below the close is already tight, and tightening into day one of a breakout converts noise into an exit); defined risk on the add = 0.73% of equity. L1 applied as a discount not a veto - BreakoutUp_52W overshoots 4wk targets and the own-book cell is n=1/hit_1w 0.0, so read the +36.8% 4wk and +41.7% 12wk for direction, not magnitude (both exceed the DCF anchor). UT-10 tested and does not bind: it gates PullbackBuy_VWAP buy-LIMITS into an UNDERWATER name; this is a BreakoutUp_52W buy-STOP on a +7.86% position. Only the high-ATR leg holds (35.28 = 7.3% of price), which is why size is halved. f=0.30 per L5 (9 graded buys, below the 10 floor). Withdraw triggers: conviction <65, expiry 08-17 unfilled (reprice from the then-current snapshot, never republish @487.08), or a settled close below EMA20 435.27. · news

decided @ 487.08 · stop 444.14 · t1w 540.88 · t4w 666.40 · t12w 690.41

Exit the last of a losing TSMC position the model can no longer form any view on. The signal engine prints NO TRADE with conviction 0 and an Indeterminate pillar consensus ('consensus conflict + low conviction') for the third time in four sessions, while its 12-week projection of 376.25 sits 9.1% BELOW the 414.00 close — a directionless name with a negative forward view, held at a 5.5% loss. TSMC's business news is genuinely strong (Q2 revenue 2bn +36% y/y, net income +77%, Q3 guided +37%, 2026 capex raised to -64bn, a further $100bn committed to Arizona, and roughly of open-market buying by the CEO/CFO/COO since early July) — but that is a thesis argument, not a timing signal, and it does not override. If TSMC deserves owning here it deserves a fresh entry on a live buy signal with a real stop, not the remainder of a position already 61% sold. OPERATIONAL: 9th arming (07-25, 07-28, 07-29, 07-30, 07-31, 08-01, 08-04, 08-05, today). ROOT CAUSE OF THE 8 PRIOR NON-FILLS FOUND AND FIXED IN THIS CARD: every arming since 07-25 was written execution='market_on_open', which is outside paper_executor._EXEC_KINDS {market,gtc_limit,gtc_stop}; _eligibility returns None and drops the card BEFORE the skip-logging path, so it produced no output at all. The four TSM sells that did fill (07-17 to 07-22) all used execution='market'. This card uses 'market'. Basis is signal-based per CLAUDE.md s9 (sustained NO TRADE on a held position + negative 12wk, 16th session) and L2/UT-11 (an unfilled armed exit is an incomplete exit - re-fire, do not re-litigate on price action). Stop-break leg retired (close +8.20% above the 382.61 stop). PROFIT_ARMED at T12_Progress 110.03 with Gain_ATR -1.46 is the UT-9 arm-on-loss mislabel, inert, ~21st session - this is a loss-cut, not a profit-take. Valuation amplify suppressed by G2 (DCF 45.15 vs price 414.00, ~9.2x off) so val_adj forced to 0, 17th session. Proceeds at the 414.00 anchor close ref (08-05 open printed 419.32); realised ~- (-5.48% vs 438.02 cost). Est band 375.43-425.76: top only +2.84% above the exit, floor -9.32%. 08-10 TSMC monthly revenue lands two sessions after the exit, stance 'none' - not a sale into a catalyst. RE-ENTRY: BUY/STRONG BUY conv >=65 with a live Entry_Price and a Target_12Week ABOVE entry - not a price level, and not this position. · news

stop 382.61 · t12w 376.25

Dell broke to a fresh 52-week high on the back of an AI-server business that is now the whole story: last quarter's revenue grew 88% year on year, the company booked 4bn of AI orders, and management lifted its full-year revenue guidance to -169bn and its AI-server target to about $60bn. The model's read is the strongest this book has produced - a 52-week breakout at conviction 88, ranked first, with trend and momentum both near maximum - and the holding has just turned profitable for the first time since it was opened in July. We are adding to a winner we already own, at the model's own trigger price, with the risk line defined. The valuation screen is the one thing pushing back: on an earnings-power basis the shares look richly priced, which is why the add is deliberately half the size the signal alone would justify. Operational nuance: ADD path (112.34% x 347 =) x 0.5 size modifier (MoS_FV -3.14, <= -3 tier) =; GTC buy-stop at the printed Entry_Price 465.04 (ED -0.06, so it fills on the open unless DELL gaps below the trigger - that is the intended protection). Post-fill @ blended 448.63 = 25.3% of book, max_position_pct 60% not binding. L6 tested and does NOT bind: +8.92% is sector beta (SOX +6%, S&P record close, PLTR +30%), not a post-earnings gap - Dell last reported FQ1 2027 on 2026-05-28 and next reports late August. L1 overshoot flag: Target_Price 667.22 and even the f=0.30 1wk of 525.69 print ABOVE the model's own Est band top of 509.17; treat 509.17 as the credible 4-week ceiling. 12wk 639.35 exceeds the DCF anchor 485.61. UT-10 examined not merely cited: its not-underwater gate now passes ( +6.34%, Gain_ATR +0.70) while its non-spiking-ATR gate fails (ATR 29.22->39.77), and its stated scope is pullback-pyramiding, so it discounts size rather than vetoing. RSI_DIV on the row is neutralised as a trim trigger by G1 (fresh BreakoutUp at conv >= 80) and is a trail instruction, not a buy blocker. STOP RATCHET: the engine stop LOWERED 390.11 -> 385.49 on an up day (ATR expansion outran the anchor); stop is held at 390.11 and must never be lowered. Tighten on 08-07 NFP and 08-12 CPI (horizon stance tighten-trail). · news

decided @ 465.04 · stop 390.11 · t1w 525.69 · t4w 667.22 · t12w 639.35

TSMC's exit case has half-repaired, so we take profits on the de-risk rather than complete it: after a +5.5% session the shares closed at 424.61, back above the 50-day average and only 0.5% under our stop, which is no longer a decisive break - but the model's 12-week view (407.07) is still 4.1% BELOW the price and the engine still reads the name bearishly (AVOID, conviction 64 and rising) on a position we are underwater on. Q2 was a blowout (EPS +74.5%, FY26 revenue growth guided above 40%) yet the rally is a sector-wide AI bid on Taiwan's +83.6% US export orders, not a TSMC re-rating, and the capex raise to -64B is a real margin overhang. So: halve the position into strength at a price 5.5% better than the level the previous verdict was chasing, and keep the other half working with a raised stop. Operational: L2 re-tested and found half-satisfied (stop-break leg healed, negative-12wk leg survives) -> exit SCALED from a full reduce-to-30-residual down to a 50% trim; L3/UT-11 escalation of a FOURTH consecutive 30-share partial fill (txn ustech-20260722-0001) - act at the right size, do not drip a fifth session; G2 SUPPRESSES the valuation amplify (DCF vs price is ~10x broken) so the MoS<=-3 tier bump is deliberately NOT applied; UT-9 - PROFIT_ARMED prints at T12_Progress 104.31 on a -3.06% underwater position (Gain_ATR -0.74) = arm-on-loss mislabel, inert, this is a loss-trim not a profit-take; L5 f=0.30 (calib -0.299 degenerate); L6 no post-rebound chase. Residual on a raised stop 406.37 (close - 1xATR, ratcheted up from 385.00). Proceeds. · news

decided @ 424.61 · stop 406.37 · t12w 407.07

TSMC's exit case has half-repaired, so we take profits on the de-risk rather than complete it: after a +5.5% session the shares closed at 424.61, back above the 50-day average and only 0.5% under our stop, which is no longer a decisive break - but the model's 12-week view (407.07) is still 4.1% BELOW the price and the engine still reads the name bearishly (AVOID, conviction 64 and rising) on a position we are underwater on. Q2 was a blowout (EPS +74.5%, FY26 revenue growth guided above 40%) yet the rally is a sector-wide AI bid on Taiwan's +83.6% US export orders, not a TSMC re-rating, and the capex raise to -64B is a real margin overhang. So: halve the position into strength at a price 5.5% better than the level the previous verdict was chasing, and keep the other half working with a raised stop. Operational: L2 re-tested and found half-satisfied (stop-break leg healed, negative-12wk leg survives) -> exit SCALED from a full reduce-to-30-residual down to a 50% trim; L3/UT-11 escalation of a FOURTH consecutive 30-share partial fill (txn ustech-20260722-0001) - act at the right size, do not drip a fifth session; G2 SUPPRESSES the valuation amplify (DCF vs price is ~10x broken) so the MoS<=-3 tier bump is deliberately NOT applied; UT-9 - PROFIT_ARMED prints at T12_Progress 104.31 on a -3.06% underwater position (Gain_ATR -0.74) = arm-on-loss mislabel, inert, this is a loss-trim not a profit-take; L5 f=0.30 (calib -0.299 degenerate); L6 no post-rebound chase. Residual on a raised stop 406.37 (close - 1xATR, ratcheted up from 385.00). Proceeds. · news

decided @ 424.61 · stop 406.37 · t12w 407.07

Cut the loss on TSM: sell 206 of and keep only a 30-share tail. TSMC's Q2 was a genuine blowout - record revenue, profit +77% and full-year growth guidance lifted above 40% with Arizona spending raised to - but the market has spent five sessions selling that news, knocking the shares ~7% from the print on fears that the -64B capex bill compresses margins. Our own forward view has now flipped decisively negative: the 12-week projection has slid ->, i.e. 6% BELOW today's price, while the stock sits 5.7% under its protective stop and 8.2% underwater on cost. Strong fundamentals the tape refuses to pay for are not a reason to keep holding a broken chart, so we step out and leave a small tail in case the AI cycle reasserts. OPERATIONAL: third re-fire of the same armed exit - the 07-17 and 07-18 proposals each executed as only, so the de-risk is still incomplete; UT-11/L3/UT-3 say escalate the residual now, do not re-defer. L2 fully satisfied (decisive break of the 426.80 stop + negative 12wk + bearish/absent signal); row label is NO TRADE (conv 0, consensus conflict) but the label is not required. Residual on stop 385.00 (close - 1xATR); close 402.30 still above EMA200 359.77 so not a full below-EMA200 exit. Valuation val-blind: DCF vs price is broken -> G2 suppress, MoS -16.05 not applied. UT-9: PROFIT_ARMED prints (T12_Progress 106.51) on a -8.15% underwater position (Gain_ATR -2.10) = arm-on-loss mislabel, inert - this is a loss-cut, not a profit-take. Estimates: 1wk/4wk n/a (no Target_Price on a NoTrade row; model band 388.83-447.08), 12wk 377.71 (-6.1%); f=0.30 per L5 (calib_1w -0.299 degenerate). Execution pending the next US open 2026-07-21; reprice if the open gaps >~1.5%; grade from the execution price. L2/L3/L4/L5/L6/L8, UT-3/UT-9/UT-11 applied. · news

decided @ 402.30 · stop 385.00 · t12w 377.71

Complete the armed L2 exit: sell 236 more of TSM (reduce -sh residual on stop 380.00). The prior 07-17 SELL 90% only PARTIALLY executed - the auto-executor sold just @ 392.12, leaving 266 held - so the de-risk is incomplete and per L3/UT-3 escalate rather than re-defer. The bear case has STRENGTHENED: row is bearish AVOID/RallySell_BearTrend (conv 49), Target_12Week flipped from +0.4% flat (07-17) to -3.8% NEGATIVE (383.29 vs close 398.37), and the close is a decisive break of the documented 426.80 stop (GainATR -2.15, -9.05% underwater). L2 fully satisfied (stop-break + negative 12wk + bearish signal). Keep a 30-sh residual (TSMC's record Q2 warrants a small tail; close 398.37 is still above EMA200 359.04 so not a full below-EMA200 exit) on stop 380.00 (close - 1xATR). News NEUTRAL for the trade: TSMC Q2 was a blowout (rev +36% YoY, EPS +77%, HPC 66%, + Arizona, FY26 guide >40%) but the stock SOLD the news (-2.77% post-print, still sliding) - fundamentals already priced, price action rejected them; news does not cancel the signal exit. Valuation GROWTH_PRICED_IN MoS -16.05 but DCF is broken -> G2 suppress, val-blind. PROFIT_ARMED prints but underwater (GainATR -2.15) = arm-on-loss mislabel, not a profit-take. Execution pending next US open 2026-07-20; reprice if the open gaps >~1.5%; grade from the execution price. Proceeds L2/L3/L4/L5/L6/L8 applied. · news

decided @ 398.37 · stop 380.00 · t4w 403.50 · t12w 383.29

DELL prints STRONG BUY - ADD conv 100 on PullbackBuy_VWAP with a AI-server backlog, FY27 EPS guide up, 19-up/0-down EPS revisions and a Street 12mo target - a genuinely strong AI-infrastructure name back above our cost. But it just popped +8% and closed 1.32 ATR above the model entry (RSI_DIV), so per no-chase (L6/UT-6) we queue the ADD at the 419.03 model entry rather than chase the pop; fills only on an ~8.4% pullback. Held-name ADD so quality gate N/A; val_adj -10 (MoS -3.06), size-mod 0.5x (MoS<=-3). Stop 420.12 (ratchet, do not lower to model 360.70). 12wk 651.99 exceeds DCF (growth-beyond-priced-in). Auto-withdraw if conv<65 or unfilled by 2026-07-17. · news

decided @ 419.03 · stop 420.12 · t1w 432.06 · t4w 462.46 · t12w 651.99

Dell flips to a fresh STRONG BUY on MomentumContinuation_Up (Buy_Rank #1, conv 78) with a self-limiting buy-STOP above market at 450.88 (close 431.97, +4.4% away) -> it fills only if momentum resumes, so no chase (UT-6/L1). Quality gate PASSES (F-Score 8, FAIR_VALUE) so the new entry is allowed, but the name is priced above intrinsic (MoS_FV -2.50 -> val_adj -10, adj conv 68) and Street is cautious: avg PT (~+12% from close) yet shares -8.7% over 30d and UBS says the 2026 near-doubling caps further upside -> news NEUTRAL, the stop-entry structure is the right hedge. Sized 0.6695*1M*0.20=133,900 x0.75 size-mod (MoS in -3..-1) = 100,425 ->. Stop 351.50 is wide (3 ATR, -22%). Executes next US open 07-09 as a GTC buy-stop, expires 07-14 (=CPI day). CAVEATS: model 4wk 649.65 / 12wk 615.56 both exceed DCF & Street anchor (annotate); 4wk target sits ABOVE the 12wk (model inconsistency); PROPOSED-P1 (inert) flags MomentumContinuation_Up as a weak 1wk cell (0/3, -18.9% overshoot) -> watch the fill. f=0.30 (calib degenerate, L5). L8: pending next open, reprice if the open gaps >~1.5%. · news

decided @ 450.88 · stop 351.50 · t1w 510.51 · t4w 649.65 · t12w 615.56

AI-foundry momentum upgraded to STRONG BUY (conv 76, Buy_Rank #3) off the settled 07-06 close: Barclays Overweight PT, consensus Strong Buy (13/17), 3nm capacity expansion (TW/AZ/JP) + wafer price hikes, +93% 52wk. Model prints a Market order at 451.79 but converted to a GTC buy-limit at the same price to avoid chasing into the 07-08 FOMC minutes (overlay: no new leverage into the print), 07-14 CPI, and Q2 earnings 07-16 -> fills only without a gap-up; expires 07-10, before earnings. FailedBreakdown_Up_20D is an L1-discounted chase cell and this is a single fresh print (L4) -> modest starter, not a max position. NA_CCY valuation -> val_adj 0, gate off. f=0.30 (L5). Sized 0.6695*1M*0.20 ->. · news

decided @ 451.79 · stop 438.91 · t1w 467.25 · t4w 503.31 · t12w 538.11