Hedge Fund Research Desk · updated October 3, 2026
📢 Follow & Share
This is a live, public $1M fund — every trade posted in real time. If it helps you, send it to someone who trades.
Circuit breaker
🟢 ALL CLEAR
Rules
day ≤ -2.5% / week ≤ -5% → halt entries · -10% off peak → full stop + manual restart
💼 Positions
Live paper-fund book across the three accounts · auto-refreshed with the page
Fund equity
$1,044,428
Day P&L
−$35
Total P&L
+$44,428 (+4.4%)
Fund acct
$1,044,428 (+4.4%)
| Symbol | Book | Qty | Entry | Last | Value | Unrealized P&L |
|---|---|---|---|---|---|---|
| TQQQ | Fund | 3,115 | $73.82 | $81.01 | $252,346 | +$22,396 (+9.7%) |
| SPY | Fund | 235 | $767.43 | $769.64 | $180,865 | +$519 (+0.3%) |
| MPC | Fund | 130 | $310.56 | $422.33 | $54,903 | +$14,530 (+36.0%) |
| AMD | Fund | 84 | $479.50 | $633.91 | $53,248 | +$12,970 (+32.2%) |
| DELL | Fund | 87 | $455.00 | $562.52 | $48,939 | +$9,354 (+23.6%) |
| EFA | Fund | 451 | $106.27 | $103.96 | $46,886 | −$1,044 (-2.2%) |
| ANET | Fund | 211 | $189.03 | $207.35 | $43,751 | +$3,865 (+9.7%) |
| GH | Fund | 240 | $169.92 | $177.80 | $42,672 | +$1,891 (+4.6%) |
| SNOW | Fund | 121 | $334.07 | $341.04 | $41,266 | +$843 (+2.1%) |
| ET | Fund | 1,956 | $20.50 | $20.47 | $40,039 | −$59 (-0.1%) |
| RVMD | Fund | 193 | $206.24 | $206.64 | $39,882 | +$77 (+0.2%) |
| FIX | Fund | 23 | $1,706.58 | $1,728.01 | $39,744 | +$493 (+1.3%) |
| CAT | Fund | 47 | $841.00 | $845.42 | $39,735 | +$208 (+0.5%) |
| ECG | Fund | 292 | $136.61 | $123.35 | $36,018 | −$3,872 (-9.7%) |
| IESC | Fund | 102 | $388.52 | $339.34 | $34,613 | −$5,016 (-12.7%) |
| GE | Fund | 108 | $370.00 | $309.56 | $33,432 | −$6,528 (-16.3%) |
| ATI | Fund | 172 | $230.80 | $192.42 | $33,096 | −$6,601 (-16.6%) |
| EEM | Fund | 376 | $66.60 | $67.67 | $25,444 | +$401 (+1.6%) |
| DBC | Fund | 762 | $31.02 | $32.54 | $24,795 | +$1,156 (+4.9%) |
| TER | Fund | 19 | $409.91 | $449.04 | $8,532 | +$743 (+9.5%) |
| PANW | Fund | 20 | $382.09 | $403.24 | $8,065 | +$423 (+5.5%) |
| OSCR | Fund | 234 | $32.68 | $30.94 | $7,240 | −$407 (-5.3%) |
| NBIS | Fund | 29 | $271.63 | $242.81 | $7,041 | −$836 (-10.6%) |
| STX | Fund | 8 | $964.72 | $848.99 | $6,792 | −$926 (-12.0%) |
| QQQ 261106 P710 | Fund | -2 | $7.94 | $6.12 | $-1,224 | +$364 (+22.9%) |
| MPC 261016 C450 | Fund | -1 | $4.70 | $5.90 | $-590 | −$120 (-25.5%) |
| QQQ 261030 P675 | Fund | -2 | $8.80 | $1.72 | $-344 | +$1,416 (+80.5%) |
| ANET 261016 C230 | Fund | -2 | $2.53 | $0.95 | $-190 | +$316 (+62.5%) |
🧾 Recent Trades
| Date | Symbol | Book | Side | Qty | Fill |
|---|---|---|---|---|---|
| 2026-10-01 | GH 261016 C210 | Fund | BUY | 2 | $0.45 |
| 2026-10-01 | EEM | Fund | BUY | 133 | $66.74 |
| 2026-10-01 | SPY | Fund | BUY | 19 | $762.70 |
| 2026-10-01 | EFA | Fund | BUY | 128 | $103.08 |
| 2026-09-30 | QQQ 261106 P710 | Fund | SELL | 2 | $7.94 |
| 2026-09-30 | QQQ 261016 P678 | Fund | BUY | 2 | $0.91 |
| 2026-09-30 | Fund | 2 | $0.74 | ||
| 2026-09-28 | VNQ | Fund | SELL | 338 | $91.32 |
| 2026-09-28 | SNOW 261016 C410 | Fund | BUY | 1 | $0.56 |
| 2026-09-22 | SPY | Fund | BUY | 14 | $774.30 |
| 2026-09-18 | GE 261016 C370 | Fund | BUY | 1 | $0.43 |
| 2026-09-17 | GH 261016 C210 | Fund | SELL | 2 | $2.30 |
🎯 Allocation Dial
Mechanical exposure ladder: trim when the market is overextended, deploy the reserve into big dips, defensive below the 200-day
SPY vs 200-day
+7.3% · NORMAL BULL
Target exposure
110%
Off 20-day high
-0.5%
SPY RSI(2)
87
Proprietary exposure ladder: leverage deployed in early trends and hard dips, trimmed when extended, defensive below trend — exact thresholds private
🤖 Machine Learning Loop
Weekly cycle: every sleeve is graded against its sealed backtest expectation (never retuned on the same data);
losers go on probation (half size) or halt; factor weights learn walk-forward from realized returns. Last cycle: 2026-10-03
swing
🔵 LEARNING
0 trades
tqqq
🔵 LEARNING
0 trades
csp
🔵 LEARNING
2 trades
supp
🔵 LEARNING
3 trades · PF 5.52
dip
🔵 LEARNING
0 trades
leaps
🔵 LEARNING
0 trades
pead
🔵 LEARNING
1 trades · PF 0.00
condor
🔵 LEARNING
1 trades
moon
🔵 LEARNING
0 trades
trend
🔵 LEARNING
1 trades · PF 0.00
cc
🔵 LEARNING
3 trades
Factor weights: equal until 8 weekly snapshots accumulate
🗣️ Social Arb Radar
Camillo-style attention anomalies: Google Trends + Wikipedia pageviews + Reddit velocity vs baseline,
across a hand-curated product→ticker map. WATCH ONLY — no trades until a signal proves itself. Scan: 2026-10-01
| Ticker | Term | What | Trends x | Wiki x | Heat |
|---|---|---|---|---|---|
| CAVA | Cava | mediterranean fast casual | 0.84 | 65.52 | 🔥 HOT (29.59) |
| NFLX | Monster Lizzie Borden | Netflix Ryan Murphy S4, Sep 2026 | 3.03 | 53.4 | 🔥 HOT (25.42) |
| NFLX | Unabomber movie | Netflix Russell Crowe film, released 9/25/26 | 13.07 | 17.01 | 🔥 HOT (14.82) |
| AMZN | The Love Hypothesis | Amazon MGM / Prime Video, released 9/23/26 | 3.03 | 21.69 | 🔥 HOT (11.32) |
| PSKY | Heart of the Beast | Paramount / Brad Pitt survival film, wide 9/25/26 | 4.48 | 9.84 | 🔥 HOT (6.86) |
| SONY | Resident Evil movie | Sony Pictures release 9/18/26 | 2.48 | 12.24 | 🔥 HOT (6.81) |
| — | Crumbl | cookies (private — trend context) | 0.63 | 14.2 | 🔥 HOT (6.66) |
| CMCSA | Forgotten Island | DreamWorks/Universal animated film, wide 9/25/26 | 4.33 | 8.81 | 🔥 HOT (6.32) |
| WBD | Digger movie | Warner Bros / Tom Cruise / Inarritu, 10/2/26 wide release | 7.38 | 3.98 | 🔥 HOT (5.87) |
| DIS | American Horror Story | FX/Hulu AHS S13, premiered 9/24/26, weekly eps | 3.15 | 7.35 | 🔥 HOT (5.02) |
🏛️ Capitol Radar
Recent congressional stock BUYS (STOCK Act disclosures filed within 60 days of trade). WATCH ONLY. Scan: 2026-09-30
| Ticker | Buyers | Detail |
|---|---|---|
| MSFT | 5 | D; a |
| GOOGL | 4 | D; a |
| AVGO | 3 | D; a |
| AMAT | 3 | J; o |
| BRK/B | 3 | M; a |
| ACN | 3 | S; c |
| HCA | 3 | R; o |
| PG | 3 | D; a |
| NVDA | 3 | J; o |
| GS | 2 | D; a |
| MRVL | 2 | R; o |
| MLM | 2 | A; p |
- 28 Sept batch: no confluence with the attention layer (NFLX/META/CAVA/DECK/ONON flagged) or the fund’s books. Hern’s CB/KO are defensive 1K-15K odd lots from a muni-heavy PTR. Blumenthal’s large legs are spouse private real-estate/LLC vehicles – no tradable signal. Standing read: slow, low-signal feed.
🏗️ Theme Radar
FRIDAY 10/2 READING (10/1 close): THE FIRST LEGAL SCORING SCAN since the 9/17 print. AI Data-Center Build. Census C30 says the build is ACCELERATING, not peaking: data-center construction $85B SAAR, +73% y/y, +7.5% m/m, ~149% annualized since March, while power-plant construction grows only +8.5% – the load is outrunning the supply build by an order of magnitude. Micron’s capex raise moved money from the shortage owner to the cure in one session. AI Data-Center Build. The 10-year printed a new 19-year high and the funding sort got sharper: ORACLE SENT A FORCE-MAJEURE NOTICE on Project Jupiter (the Bloom-fuel-cell-powered New Mexico campus developed by a Blue Owl unit), seeking to defer rent if the site misses its 2028 target. Both parties say it is on schedule, but the most leveraged borrower in the build has now formally reached for an escape clause. The Texas permit pause now HAS AN END DATE: ERCOT’s impact report is due ~Dec 10 (TCEQ compliance report 10/19). — 10/2 ADDS: Census C30 August – data-center construction $14.4B in the month / $85B SAAR, +7.5% m/m, +73.2% y/y, +823% since Jan 2021, ~149% annualized since March; >10% of private nonresidential. Power construction $186B SAAR, +0.8% m/m, +8.5% y/y. Factory construction still ~2x data centers (~$19B/month). Forgent Power Solutions Q3 orders +268% y/y, $1.5B backlog (~18 months of output); MV switchgear sold out through 2028, gas-turbine lead times past 5 years (DC POWER eXchange 9/30). NOTE C30 counts the shell and integrated MEP – the cheap part – not servers or electrical gear. 9/29 ADDS: global data-center capex outlook raised above $1T for 2026; the 14 largest operators ~$750B vs <$450B in 2025 (Amazon ~$200B, Google $195-205B). Liquid cooling guided +25-40%/yr (Vertiv $9.5B backlog; AAON $2.13B backlog, mostly BASX data-center liquid cooling). Oracle force-majeure notice on Project Jupiter. Texas TCEQ pause end-date: ERCOT report ~12/10. DOE $1.9B loan backs a third nuclear restart. 10y 5.24-5.27%, a 19-year high. PRIOR (9/27): SPEND KEEPS ACCELERATING, BUT MORE OF IT IS PRICE AND LESS OF IT IS PERMISSION. Dell’Oro: worldwide data-center capex +92% YoY in Q2 2026, with neoclouds and model builders growing fastest — and Dell’Oro itself flags that a large part is DRAM/NAND price inflation raising server ASPs, not deployed capacity. GE Vernova: data-center electrification orders in one quarter exceeded all of 2025, electrification orders doubled to $7.1B. Caterpillar: record backlog on data-center power gen. Transformer lead times 3-5 years, power transformers 128 weeks, HV breakers ~125 weeks. SemiAnalysis pushes back on the ’30-50% of 2026 capacity canceled’ stat we carried on 9/22: hyperscaler self-build YE2026 forecasts moved only ~1% and colo <5% over six months; what slips is financing-less announcements; the real bottleneck is equipment lead times, with COOLING named as the next one. THE NEW CONSTRAINT IS POLITICAL: Abbott ordered TCEQ on 9/21 to stop issuing any data-center permit until an ERCOT/TWDB audit is done (no end date; TCEQ reports back 10/19), expanding the August grid-approval moratorium and aimed specifically at behind-the-meter developers bypassing ERCOT; Massachusetts EO 658 (9/8) bars permits for new sites over 25 MW; California has a seven-bill energy/water package; ~$68B of Q2 projects were blocked by local opposition. AND THE COST OF CAPITAL JUMPED: 10y 5.13%, 30y 5.53%, oil back above $103 at one point; utilities were the worst S&P sector. Dated catalyst: MICRON REPORTS WEDNESDAY 9/30 after the close (guided $50.0B revenue, 86% GM); Census C30 August construction spending due ~10/1. Scan: 2026-10-02
| Ticker | Tier | Evidence | Momentum |
|---|---|---|---|
| NDSN | precision dispensing plus electronics test and inspection (SCORED 10/1: NOT CONFIRMED, NOT FAILED – missed the since-scan leg by 0.5) | FIRST LEGAL SCORE, AND IT IS A NEAR-MISS ON OUR OWN LETTER. Relative mom1 +4.5 (mom1 +5.0 vs SPY +0.5): PASS. Relative return since the prior scan (9/28 close): -0.5: FAIL by half a point over three sessions. Relative return since the 9/17 proposal print, which is the genuinely non-overlapping number: +5.1. The rule says both legs positive, so this is NOT CONFIRMED – and we do not bend the rule the first time it costs us. It is also not a failure: the trend is intact (-1.4% off high, template TRUE, vol 22, mom3 +13.6). Carried to the next legal score, the 10/15 close (Fri 10/16 scan). RULE QUESTION FOR THE REVIEW TASK: the ‘since-scan’ leg should arguably be measured from the TESTED print (9/17), not from a 3-session-old intermediate scan; written down here, not applied retroactively. | SCORED: rel mom1 +4.5 PASS, since-9/28 rel -0.5 FAIL, since-9/17 rel +5.1. -1.4% off high, mom3 +13.6, TRUE, vol 22, $113M/day. Next legal score 10/15 close |
| AMAT / LRCX / KLAC / TER / CRDO | THE SEMICONDUCTOR TOOL CHAIN – Micron funded the cure and the tape paid it; LRCX and TER now template TRUE (trigger clock STARTED, AVOID lifts only on the second print) | THE WAVE HANDOFF HAPPENED IN THE OPEN. Micron raised FY27 capex above prior plans (H1 ~$25B, Q1 ~$11.5B, heavier construction in H2) and its own stock faded after hours, while the suppliers of that capex ran: since the 9/17 print AMAT +26.4 rel, LRCX +26.0, KLAC +18.1, TER +17.3, CRDO +24.5; since 9/28 AMAT +9.0, LRCX +8.4. That is exactly the shortage-owner-to-cure transfer the wave framework predicts. Trigger status: LRCX template RECLAIMED (TRUE, -21.4% off high, rel mom1 +16.8) and TER TRUE (-14.0%, rel mom1 +23.5) – this is the FIRST qualifying print for both; AMAT (-26.7%) and KLAC (-33.5%) still FALSE. Second print needed at the 10/15 close. HONEST OBJECTION: mom1 +17 to +24 is not modest momentum; whatever this is, it is no longer the next-wave signature, it is the wave being recognized. Fund’s TER ($7k) held, not added to. | FIRST PRINT: LRCX TRUE -21.4% mom1 +17.3; TER TRUE -14.0% mom1 +24.0; AMAT FALSE -26.7% mom1 +19.8; KLAC FALSE -33.5% mom1 +17.2; CRDO FALSE -30.5%. Since 9/17 rel +17 to +26. Second print 10/15 |
| GNRC / CMI / CAT / POWL | BACKUP POWER AND BEHIND-THE-METER GENERATION — Texas just aimed its permit freeze at the reroute itself (WATCH, THESIS DAMAGED) | CORRECTION TO THE WAVE-4 MECHANISM. We wrote that grid scarcity plus state large-load tariffs FORCE self-supply. On 9/21 the largest data-center state halted every TCEQ permit for data centers, and the stated reason is to stop behind-the-meter developers from bypassing ERCOT — on-site gas generation needs exactly those air permits. So the reroute is now itself a political target, and the pattern is spreading (Massachusetts 25 MW cap, California package, $68B of Q2 projects blocked locally). Orders already booked still stand (POWL $400M+ phase-1 BTM order, record $2.4B backlog; Amazon’s up-to-$8B Generac framework; CAT record backlog; Kodiak’s 76 MW West Texas BTM job). But the tape agrees with the downgrade: KGS -30.1% off high, mom1 -12.5; CMI -27.6%, mom1 -9.2. The fund holds CAT (47 sh). Watch, no adds, and any future trigger must survive the ERCOT audit outcome. | BROKEN, TEMPLATES FALSE — POWL -41.1% off high, since 9/21 +1.1 (rel +1.3). GNRC -29.6%, +1.6 (rel +1.8). CMI -27.6%, -1.8, mom1 -9.2. CAT -22.7%, +0.6, HELD. KGS -30.1%, mom1 -12.5. Watch |
| ENTG | SEMICONDUCTOR CONSUMABLES – paid per WAFER RUN (WATCH; the signature is turning into a chase before it can be scored) | Micron’s FY27 capex raise is the catalyst this entry named, and ENTG responded: +5.4 rel since the 9/28 close, +17.2 rel since 9/17. Template TRUE, 13.8% off its high. THE PROBLEM IS THE OTHER LEG OF THE SIGNATURE: mom1 is now +22.4 (rel +21.9) – the ‘momentum still modest’ condition that made it a candidate is gone. First legal score remains the 10/5 close (Tue 10/6 scan); if it passes, it passes as a momentum name, and a proposal from here would be buying the move after it happened. Vol 64. No proposal, no chase. | TEMPLATE TRUE, NO LONGER MODEST – -13.8% off high, mom1 +22.4, mom3 +8.1, mom6 +33.7. Since 9/28 rel +5.4, since 9/17 rel +17.2. Scores at 10/5 close |
| VICR / AEIS / MPWR | POWER CONVERSION — more compute per PERMITTED megawatt (NEW WATCH, not a signature) | NEW FORWARD READ. If megawatts become politically rationed (Texas permit halt, Massachusetts 25 MW cap, California package, $68B of projects blocked), the scarce unit shifts from the MW to compute-per-MW, and the payout moves to whatever raises it: 48V/800VDC power conversion, vertical power delivery, liquid cooling retrofits (Airsys: ~22 MW of every 100 MW wasted by legacy cooling). Vicor is the live example: an 18% day on 9/17 on a non-exclusive Vertical Power Delivery license to an unnamed OEM, Q3 sequential growth guidance raised from ~10% to >20% on royalties, and two New Hampshire sites bought for Fab-2/Fab-3 to nearly triple capacity. Royalty revenue is the pricing-power test our framework asks for. BUT IT IS NOT THE SIGNATURE: VICR +25.9% in four sessions, mom1 +43.0, template FALSE, vol 88, $195M/day — that is a chase. AEIS (+6.4, FALSE, -28.1% off high) and MPWR (+7.0, FALSE, -19.0%) are the more ownable expressions and neither has repaired. Watch the tier; no trigger until a template reclaim. | STILL A CHASE – VICR mom1 +75.3, -18.7% off high, FALSE, since 9/17 rel +42.2. AEIS -24.4% FALSE (below SMA). MPWR -19.2% FALSE. Watch the tier, no trigger |
| TRGP / WES / ET / OKE | Permian gathering and processing — hit by the rate shock, not by the thesis (HOLD ET, NO ADDS) | Midstream traded as a yield proxy into a 5.13% 10-year: TRGP -4.0 (rel -3.7), ET -3.4 (rel -3.1), OKE -3.5, WMB -3.3, KMI -3.1, WES -2.1 over four sessions. OKE’s template has now FAILED (-9.1% off high); TRGP, ET and WES remain TRUE. Nothing in the mechanism changed — but note the Texas permit halt also blocks behind-the-meter gas plants in the Permian’s backyard, which trims the ‘molecules to campuses’ volume upside for Texas-sited load. ET is held and behaving as a 17-vol income name; hold, no adds, score at the first legal scan. | SOLD WITH RATES, STEADIED 9/28 – TRGP -8.2% off high TRUE; ET -7.5% TRUE, HELD (9/28 rel +0.3); WES -9.1% TRUE; OKE -9.1% FALSE. Since 9/17 rel -3 to -6 |
| WTS / IEX / AIT / GWW | the COOLING LOOP AND MRO TIER — stood down on 9/18, and a two-day scan may not un-stand it (STAND DOWN, UNCHANGED) | Stand-down unchanged and not renegotiable on four sessions. For the record the tier firmed: AIT +4.1 (rel +4.3), IEX +3.4 (rel +3.7), WTS +2.0 (rel +2.3), GWW -2.7. mom1 still negative across all four. SemiAnalysis names COOLING as the next equipment bottleneck after transformers, and the Airsys analysis says legacy cooling wastes ~22 MW per 100 MW site — relevant if power becomes permit-capped (see VICR entry). | TEMPLATES TRUE, STOOD DOWN – AIT -6.8% off high, 9/28 +1.2 (rel +2.0), since 9/21 rel +6.3; IEX -3.9%, since 9/21 rel +4.9; WTS -8.6%, +3.1; GWW -10.3%, -0.7. mom1 still negative across all four |
| EXPD | AIRFREIGHT FORWARDING — the physical logistics of the AI build (HOLD AND TRIM, unchanged) | Gave back 2.4% (rel -2.2) and mom1 slipped to -1.4. Still 3.3% off its high with template TRUE; this is extension bleeding off, not the exit condition, which remains airfreight tonnage decelerating two months running. HOLD AND TRIM unchanged. | PASSES – -2.7% off high, 9/28 +0.6 (rel +1.3), mom1 -1.1, TRUE. Hold and trim unchanged |
| ANET / APH | interconnect and east-west networking — content per rack (ANET HELD AND BEHAVING; APH stays removed) | ANET flat (since 9/28 rel 0.0, -2.9% off high, mom1 +8.0, TRUE) – held and behaving. APH FIRST LEGAL SCORE: rel mom1 +4.7, since-scan rel +1.5, template TRUE at -2.7% – PASS, print 1 of 2 toward reinstatement; second print 10/15 close. It stays removed until then. | ANET HELD -2.9% off high, mom1 +8.0, TRUE. APH PRINT 1 OF 2: -2.7%, mom1 +5.2, since-scan rel +1.5, TRUE |
| MOD | Airedale data-center cooling — the customer prepayment is unchanged and the stock just had its best two days in months (REPAIR BENCH, trigger NOT yet met) | 9/28 -11.6% (rel -10.9) to $175.04. CAUTION – LIKELY MECHANICAL, NOT VERIFIED: the drop fell on an ex-date (a $1.90 cash distribution listed ex 9/28) while the Performance Technologies separation / Gentherm combination moves toward closing; if a spin value went ex, adjusted bars will show a fake loss. Check the distribution terms before reading this print. Demand case (>$4B Airedale 2027-29 capacity agreement with $165M upfront cash, data-center sales +90%) unchanged. Trigger unchanged: template reclaim plus two positive relative prints scored at least 10 sessions apart. | BROKEN, PRINT CONTAMINATED BY EX-DATE – -43.0% off high, 9/28 -11.6 (rel -10.9), mom1 -6.0, mom3 -31.7. Template FALSE. Bench |
| ETN / NVT / HUBB / VRT / GEV | factory-built electrical systems and grid equipment – ETN and NVT score their FIRST positive legal print (clock started, not a proposal) | FIRST LEGAL SCORE: ETN rel mom1 +11.4, since-scan rel +1.6, template TRUE (-4.9% off high) – PASS, print 1 of 2. NVT rel mom1 +12.0, since-scan rel +3.5, TRUE (-10.0%) – PASS, print 1 of 2. Second legal print: 10/15 close. Supporting fundamentals: Forgent Q3 orders +268%, MV switchgear sold out through 2028, C30 power construction growing only +8.5% against data centers at +73%. HUBB (-15.8%), VRT (-34.6%), GEV (-16.0%) still FALSE. ATKR still at its high. | PRINT 1 OF 2 – ETN -4.9% off high, mom1 +11.9, since-9/28 rel +1.6, TRUE. NVT -10.0%, mom1 +12.5, rel +3.5, TRUE. HUBB/VRT/GEV FALSE. Second print 10/15 |
| BE | solid-oxide fuel cells — the behind-the-meter reroute (WATCH ONLY, STILL UNOWNABLE AT OUR SIZE) | Bloom gave back 2.8% (rel -4.5) over the two sessions, which is the first thing it has done that is not vertical, and the Senate block removes the specific federal catalyst we cited for it last week. The state-tariff layer keeps the mechanism alive. Everything else is unchanged and so is the verdict: roughly $20 BILLION of total backlog against ~$3.6B of projected 2026 revenue, $7.65 BILLION of data-center contracts signed inside a single 90-day window, a 100 MW Equinix order, a 12.5 MW utility-grade power block for data centers — and 109% annualized volatility on a twelve-month base of +150%, which means a position sized to matter would dominate this book’s risk. Correct mechanism, unownable security at our size. Recorded a fourth time, not sized | TEMPLATE TRUE, UNOWNABLE, HIT ON 9/28 – -8.9 (rel -8.2) on Oracle’s Project Jupiter force-majeure notice (a Bloom-powered site) plus securities litigation; -24.0% off high, mom1 +20.7, vol 109. Watch only |
| ORCL | THE FUNDING TELL — fell another 7.7% into a 5% 10-year (diagnostic, deteriorating) | The clearest single reading of the rate shock. Oracle fell 7.7% (rel -7.4) in four sessions to 55.8% off its high while AMD rose to a $1T cap and META rose 10%. Nothing about AI demand changed; what changed is the price of the ~$40B of debt and equity Oracle still has to raise against $90-95B of capex and $664B of RPO concentrated in one counterparty. Private credit sold with it (BX -6.3, OWL -7.0). This is theme 6 showing up in theme 1: with the 10-year at 5.13%, own the parts of the build that are cash-funded and can raise price, not the ones that borrow to build. | BROKEN, WORSENING – -57.2% off high, 9/28 -3.3 (rel -2.5), since 9/17 rel -12.6; force-majeure notice on Project Jupiter. mom1 -12.7, mom12-1 -47.2. Diagnostic only |
| STICK-BUILT / EPC | AVOID — wave 1 and 2, NINTH WEEK, and the labour statistics explain WHY rather than rescuing it (a rule, not a ticker) | Eleventh week. C30 is the best demand print this tier has had (data-center construction +73% y/y, accelerating), and the tier bounced on it (PWR +3.1 rel, IESC +2.7, FIX +1.8 since 9/28) – but templates are still FALSE for PWR (-15.6%), FIX (-18.4%), EME (-18.5%); IESC TRUE at -16.8% on vol 69. Our reason for the avoid was never demand; it was that the scarce input is the labour the contractor buys. A strong C30 does not change that. Avoid stands; fund still owns FIX and IESC. | n/a – discipline note. PWR -15.6% FALSE mom1 +8.4; FIX -18.4% FALSE; EME -18.5% FALSE; IESC -16.8% TRUE. Bounced on C30 |
Crowded consensus (beta, not alpha): WEEK OF 9/21-9/25: QQQ +3.85%, SPY +1.15%, DIA -0.17%, IWM -1.21%; technology +3.6%, communication services +1.9%, UTILITIES -4.3%, ENERGY -3.8%, seven of eleven sectors down. Leaders were price-setters and platforms: META +10.2% (Muse agent, ~30% in September, best month since 2013), AMD +9.9% to a $1T cap on a reported 10% Q4 price hike, MSFT +3.7; in our coverage VICR +25.9, SANM +7.3, MPWR +7.0, AVT +8.0 to an all-time high, ENTG +6.1, AEIS +6.4. Losers were whatever borrows or yields: ORCL -7.7, OWL -7.0, BX -6.3, XLU -2.8 (-5.2 since 9/17), CCJ -5.5, midstream -3 to -4, tankers (STNG -5.5, FRO -4.2). The narrowness is the warning: the index rose on a handful of mega-caps while the average stock and small caps fell into the highest long yields in ~20 years.
Wave map: LIQUIDATED, NINTH WEEK: land, site work and shells. The 499,000-worker shortfall and electrical work at 45-70% of build cost explain the de-rating rather than reversing it — the scarce asset is the labour the contractor buys. PWR -18.2% off high, EME -19.7%, FIX -21.0%, IESC -19.0%, STRL -48.4%, DY -45.5%, MTZ -49.1% →→ STOOD DOWN BY OUR OWN RULE AND NOT UN-STOOD BY A TWO-DAY SCAN: mechanical/electrical fit-out plus the cooling loop and MRO tier. WTS -6.7, IEX -6.5, AIT -7.6, GWW -3.4 relative, templates still TRUE, order books still inflecting. Returns only on a template-intact reclaim plus two positive relative prints scored at least 10 sessions apart →→ BROKEN, AND THE SHORTAGE BENEATH IT GOT WORSE THIS WEEK: transformers, switchgear and the grid. Global transformer capacity 98% utilized, US lead times 4-5 years, switchgear sold out through 2028, US grain-oriented electrical steel self-supply only ~20%, electrical steel prices +20%, and 30-50% of 2026 data centers facing delay. Every listed owner still fails: HUBB, VRT, GEV, AZZ, POWL, CLF, and the whole regulated utility and IPP complex →→ THE REROUTE, NOW A POLITICAL TARGET: behind-the-meter self-generation. Federal Ratepayer Protection Act dead in the Senate (9/17); state large-load tariffs in ~24-25 states; and on 9/21 Texas halted all TCEQ data-center permits specifically to stop BTM developers bypassing ERCOT. Booked orders stand (POWL $400M phase-1, GNRC/Amazon $8B, CAT record backlog) but every listed expression fails the template and KGS is -30% off high →→ NOW / WHERE THE TAPE PAID: the CURE. Micron raised FY27 capex and faded; AMAT/LRCX/KLAC/TER/ENTG/CRDO rose 17-26% relative since 9/17. LRCX and TER reclaimed templates (print 1 of 2). Price-setters still hold (AVT -0.8% off high, LITE -0.7%, MU -9.6%). Sold: borrowers and yield proxies, but the 10y reversal from 5.34% on 10/1 gave them a one-day reprieve
- 10/2 FRIDAY NOTE – THE SCORING SCAN (10/1 close = 10 sessions after the 9/17 print, so the 9/18 proposals and the ETN/NVT/APH prints are legally scoreable for the first time). Tape since the 9/28 close: SPY -0.2, QQQ +0.7. The 10-year touched 5.344% on 10/1 – highest since 2002 – then reversed to ~5.21%; S&P +0.19% to 7,666. Brent ~$97-104 intraday, Hormuz still disrupted. A stopgap CR averted an October shutdown. MICRON (9/30): Q4 revenue $54.2B (guided $50.0B), EPS $33.42, Q1 FY27 guide $61.5B at ~86% GM; FY27 capex RAISED above prior plans (Q1 ~$11.5B, H1 ~$25B, heavier construction in H2) on ~$150B of long-term agreements. MU faded after hours on the capex; the tool chain did not. CENSUS C30 AUGUST (10/1): data-center construction $85B SAAR, +7.5% m/m, +73% y/y, ~149% annualized since March; power construction only +8.5% y/y; total construction unexpectedly up.
- SCORECARD 10/1 (first legal scan for the 9/17 prints): NDSN NOT CONFIRMED (mom1 leg PASS, since-scan leg -0.5) – carried to 10/15; ETN, NVT, APH: print 1 of 2 PASS; LRCX, TER: template reclaimed, print 1 of 2 on the tool-chain trigger; ENTG scores 10/5 but has lost the ‘modest momentum’ leg (mom1 +22). PROPOSALS: NONE NEW. No X watch-note: nothing has the next-wave signature AND a confirmed two-print record; the names that ran are no longer modest.
- WAVE READ: Micron is the cleanest wave-transfer evidence the file has recorded. The constraint owner reported a $54B quarter and FELL on raising capex; its suppliers (tools, consumables, test) rose 17-26% relative in ten sessions. Money is leaving the shortage owner for the cure. The NEXT tier after front-end tools is what a construction-heavy H2 FY27 capex pays: clean-room build-out, fab MEP, and the per-wafer consumables once the fabs run – which brings our banned EPC tier back into the conversation via fabs, not data centers (C30: factory construction ~2x data centers). Not a reversal of the avoid; a reason to watch whether FAB construction lines in C30 inflect next month.
- 9/29 TUESDAY NOTE (one session, nothing scoreable): the 10-year made a fresh 19-year high at 5.24-5.27% intraday on 9/28, 30-year 5.55%; SPY -0.7, QQQ -1.1, VIX 16.1; WTI back to ~$93 (from >$103). Fed governor Cook: AI and oil keep inflation up, further hikes data-dependent. The rate-shock sort continued in one day: borrowers and long-duration fell (ORCL -3.3, BE -8.9, VRT -3.6, CRDO -8.7, AMD -3.6) while cash-rich names rose (NVDA +1.7 on a new $150B buyback). Scoring calendar unchanged: NDSN/TECK/ETN/NVT/APH at the 10/1 close (Fri 10/2 scan); ENTG at the 10/5 close (Tue 10/6 scan). MICRON reports WED 9/30 after the close.
- PROPOSALS 9/29: NONE NEW, and no X watch-note – nothing can be scored one session after a reading, and the day’s big movers in the file were down (MOD on an ex-date, BE on Oracle’s Jupiter notice). NEW FACTS FOR THE WAVE MAP: (1) the Texas permit wall has an END DATE (ERCOT report ~12/10), which bounds the wave-4 damage to a known calendar; (2) Oracle’s force-majeure notice is the first time a borrower in the build has formally reached for an escape clause – the funding-tell thesis gained a document; (3) liquid cooling is the one sub-tier whose disclosed growth (+25-40%/yr) is still accelerating while its listed owners (MOD, AAON, VRT) sit 35-43% off their highs with templates FALSE – a watchlist, not a signature.
- NOTHING IS SCORED, AGAIN, AND THIS TIME THE REASON IS A MISSED RUN. The Friday 9/25 scan did not write to this file, so this Sunday catch-up covers four sessions after a Tuesday reading. The 9/18 proposals (NDSN, TECK) were made on the 9/17 print, six sessions ago; the ENTG trigger was written on 9/21, four sessions ago. SCORING CALENDAR: NDSN, TECK, ETN, NVT, APH first legal test = 10/1 close (Friday 10/2 scan). ENTG = 10/5 close (Tuesday 10/6 scan). The non-overlapping evidence so far: NDSN positive (+3.4 rel since 9/17), ENTG strongly positive (+11.6 rel since 9/17), TECK flat (+0.1) with a deteriorating mom1.
- THE RATE SHOCK SORTED THE AI BUILD BY FUNDING, NOT BY TIER. 10-year 5.13% (highest since 2007), 30-year 5.53% (highest since 2004), bond-volatility gauge +19% on the week. Within one theme, the debt-funded and yield-proxy pieces fell (ORCL -7.7, private credit -4 to -7, utilities, midstream) while cash-rich price-setters rose (AMD, MU, AVT, ENTG). This is the 9/15 fee-taker/risk-taker principle and the 9/18 pricing-power amendment showing up in a single week of tape. Practical rule while the 10-year is above 5%: no new positions whose thesis requires cheap capital (IPPs, utilities, neocloud lessors, BTM developers).
- CORRECTION TO THE WAVE-4 MECHANISM: THE REROUTE IS NOW THE TARGET. We treated behind-the-meter generation as the market’s way AROUND grid and political constraints. Texas’s 9/21 order halts TCEQ permits specifically so BTM developers cannot bypass ERCOT, stacked on the August grid-approval moratorium, with no end date. With Massachusetts, California and ~$68B of locally blocked Q2 projects, ‘social licence’ is a new binding constraint the wave map did not have. It hurts wave 1 (starts), wave 4 (BTM gen) and Texas-sited midstream volume; it helps anything that raises compute per permitted MW (power conversion, cooling retrofit) and sites outside the blocking states.
- DISCOUNT THE 92% HEADLINE. Dell’Oro’s +92% Q2 data-center capex is real spending but its own authors attribute a large part to memory and storage PRICE inflation raising server ASPs. It is evidence for the memory theme (who is charging more) more than for the physical build (how much is being built). Do not cite it as a construction statistic. Likewise, SemiAnalysis’s satellite work argues the ’30-50% of 2026 capacity delayed or canceled’ line we carried on 9/22 is overstated for funded hyperscaler builds — the delays are concentrated in financing-less announcements.
- PROPOSALS FOR THE REVIEW TASK: NONE NEW. Carried: NDSN, TECK (the latter weakening). No X watch-note posted: nothing meets the next-wave signature under our own scoring rule, and VICR, the one breakout, is extended (mom1 +43) with a failed template. Time-sensitive event rather than entry: MICRON EARNINGS WED 9/30 AFTER THE CLOSE — the FY27 capex figure bears directly on ENTG, the tool-chain trigger, and the AVT exit condition.
FRIDAY 10/2 READING (10/1 close). Memory / HBM and the Optical Chain. MICRON PRINTED: Q4 revenue $54.2B vs $50.0B guided (+379% y/y), EPS $33.42, FY26 revenue $133.2B with DRAM >$100B; Q1 FY27 guide $61.5B at ~86.25% GM; ~$150B across 26 long-term supply agreements; 2027-28 supply-demand ‘much tighter than 2026’; and FY27 CAPEX RAISED above prior plans. The stock faded on the capex; the cure rallied. Memory / HBM and the Optical Chain. Micron reports WEDNESDAY 9/30 after the close against guidance of $50.0B revenue, ~86% gross margin, $31 EPS; previews cite ~$100B of binding multi-year HBM take-or-pay contracts and UBS models FY27 blended HBM ASP +76%. The memory and optics makers fell into the print (MU -2.6, SNDK -3.7, MRVL -3.8, COHR -4.5, CRDO -8.7) while the tool chain held (KLAC +0.7, TER +0.7, AMAT +0.4). — THE SHORTAGE IS UNCHANGED AND THE CURE IS NOW FUNDED. Upstream tightness as recorded: memory product prices surged 40%-plus in Q1 2026 with a further ~50% forecast through Q2; conventional server DRAM contract prices still rising 13-18% quarter-over-quarter in Q3 though TrendForce frames this as MODERATING; supplier inventories at historic lows; DRAM +171% year-over-year; DDR5 contract pricing ~$19.50; supply-demand gaps of 4.9%/4.2%/5.1% for DRAM/NAND/HBM; SK hynix sold out for 2026 with 2027 booked and warning the shortage may run PAST 2030; Deloitte modelling the crunch as not easing until 2029. MICRON’S OWN GUIDANCE IS THE MOST EXTREME SINGLE DISCLOSURE IN THIS FILE: fiscal Q4 guided to a RECORD $50.0 BILLION OF REVENUE (plus or minus $1.0B) at an 86.0% GROSS MARGIN and $31.00 OF EPS, with HBM3E and HBM4 FULLY BOOKED THROUGH CALENDAR 2027 and demand stretching into 2028. WHAT IS NEW IS THE SUPPLY SIDE. JP Morgan now models WAFER FAB EQUIPMENT SPENDING +31% IN 2026 AND +38% IN 2027. Micron has already raised FY26 capex to ~$20B from just over $13B in FY25, and FY27 is expected to EXCEED $45 BILLION — more than a doubling, from the most aggressive investor in DRAM. Industry DRAM capex goes $53.7B in 2025 to $61.3B in 2026 (+14%); NAND $21.1B to $22.2B; SK hynix $20.5B (+17%) on HBM4 capacity at M15x; Samsung $20B (+11%) on 1C-process HBM. AND THE CHARACTER OF THE SPEND MATTERS MORE THAN THE SIZE: TrendForce describes the emphasis as shifting from raw capacity toward PROCESS TECHNOLOGY UPGRADES, HIGHER-LAYER STACKING, HYBRID BONDING and high-value products such as HBM — that is MORE STEPS PER WAFER rather than more wafers, which is a different payout than a classic capacity cycle and favours consumables and back-end over raw front-end capacity adds. The optical shortage below it is unchanged: Lumentum’s CEO calls the indium phosphide gap worse than memory while shipping MORE THAN 30% BELOW demand across five InP fabs; McKinsey models 800G output 40-60% below demand through 2027 and 1.6T 30-40% short through 2029; Nvidia put $2B each into Coherent and Lumentum; EML lead times run past 2027; indium is a ZINC BYPRODUCT with China at ~70% of refined supply and US-bound exports down 77%. Scan: 2026-10-02
| Ticker | Tier | Evidence | Momentum |
|---|---|---|---|
| TECK | zinc refining and the Western indium/germanium supply (SCORED 10/1: FAILED – PROPOSAL WITHDRAWN) | FIRST LEGAL SCORE: relative mom1 -2.7 (mom1 -2.2) FAIL; since-scan rel +0.3; since the 9/17 proposal print rel -0.4. Ten sessions after the proposal the stock is exactly where it was while the market and the theme moved on. The mechanism (Trail indium/germanium, Canada’s $400M equity-like support) is unchanged, but the tape has refused it twice and our rule says a failed score withdraws the proposal. WITHDRAWN. Back to watch; returns only on two new positive scored prints. | FAILED – -9.3% off high, mom1 -2.2 (rel -2.7), since 9/17 rel -0.4, TRUE. Proposal withdrawn |
| AMAT / LRCX / KLAC / TER | THE TOOL CHAIN — three scans of AVOID meet a forward capex number and an 8-12% two-day rip (AVOID STANDS, TRIGGER NOW WRITTEN) | This entry exists to record a challenge to our own position rather than to change it. For three scans we have said the market will own the shortage and will not own the cure, and the cleanest support was that the tool chain did not participate in the post-hike rally. In the two sessions since, it was the best group in the entire file: LRCX +12.2 (rel +10.6), AMAT +11.2 (rel +9.5), KLAC +8.9 (rel +7.2), TER +8.1 (rel +6.4), CRDO +11.3 (rel +9.6). The reason given is forward capex — WFE +31% in 2026 and +38% in 2027, Micron FY27 above $45B — which is precisely the mechanism our own wave framework would predict: the shortage owner’s cash becomes the next tier’s revenue. TWO THINGS KEEP THE AVOID IN PLACE AND BOTH WERE WRITTEN BEFORE THE OUTCOME. First, the measurement rule: two sessions cannot score anything, and this is the third leadership flip in five weeks, the previous two of which reversed inside a scan. Second, the damage is still enormous and the templates still FALSE — AMAT 35.7% off its high, KLAC 39.0%, LRCX 30.2%, CRDO 38.1% — and the trade press itself attributes the move to money returning to an oversold group with no fresh catalyst. WRITTEN TRIGGER, EFFECTIVE NOW: the tool chain comes off AVOID only on a full trend-template reclaim plus two positive relative prints scored at least 10 trading sessions apart, each with a positive since-scan relative return. The fund’s small TER position ($7.2k) is not an endorsement and is not being added to | BEST TWO-DAY GROUP, TEMPLATES STILL FALSE — since-scan LRCX +12.2 (rel +10.6), CRDO +11.3, AMAT +11.2, KLAC +8.9, TER +8.1. Off highs AMAT -35.7, KLAC -39.0, LRCX -30.2, CRDO -38.1, TER -21.1. Templates FALSE except TER. AVOID stands |
| ENTG | SEMICONDUCTOR CONSUMABLES — the per-wafer toll on a capex wave whose disclosed character is MORE STEPS, not more wafers (NEW, WATCH) | NEW CANDIDATE, AND IT IS THE FIRST THING IN WEEKS THAT PASSES THE AMENDED FRAMEWORK’S PRICING TEST BY CONSTRUCTION RATHER THAN BY LUCK. Our 9/18 amendment says stand where the spend arrives next AND where the supplier can price. The tool makers (AMAT, LRCX, KLAC) are paid once per TOOL SOLD, which makes them a pure capex-cycle bet and explains a 30-40% drawdown on a capex scare. Entegris is paid per WAFER RUN — filtration, specialty materials handling, CMP consumables, contamination control — so it collects on a capacity BUILD and on a utilization RAMP alike, and it collects more per wafer as nodes shrink and as HBM stacking multiplies process steps. That matters specifically now because the disclosed direction of memory capex is NOT raw capacity: TrendForce describes the shift as process upgrades, HIGHER-LAYER STACKING and HYBRID BONDING, which is more steps per wafer rather than more wafers. THE TAPE IS THE ONLY THING IN THE FILE WITH A GENUINELY MODEST PROFILE: template TRUE, 22.1% off its high, mom3 -22.1, mom6 +25.0, mom12-1 +51.2 — no quintupled base to deflate — at $286M a day. THE HONEST OBJECTIONS: mom1 is -1.0 (rel -2.7), so it does NOT pass today and is NOT a proposal; annualized vol of 65 is high for this book; and we screened the whole packaging tier around it and it is a wreck (UCTT -48.8% off high, ICHR -50.0, MKSI -42.2, ACLS -40.9, ONTO -27.1, AMKR -44.1, KLIC -37.0, all templates FALSE), with only COHU passing on $51M/day which is too thin for us. Written trigger, tested no earlier than 10 sessions from today: positive relative mom1 AND a positive relative return since the prior scan, twice | TEMPLATE TRUE, MOM1 NEGATIVE, NOT A PROPOSAL — -22.1% off high, mom1 -1.0 (rel -2.7), mom3 -22.1, mom6 +25.0, mom12-1 +51.2. Vol 65, $286M/day |
| MU / LITE | the CONSTRAINT OWNERS — a $50B quarter at an 86% gross margin, and still not proposals (WATCH) | MICRON DELIVERED AND THE STOCK DIDN’T CARE. $54.2B vs $50.0B guided, Q1 guide $61.5B, ~$150B of LTAs, FY27 capex raised. MU +2.9 rel since 9/28, -9.6% off high, mom12-1 +459 – fully paid, context only. LITE is the stand-out: +13.7 rel since 9/28, -0.7% off high, mom1 +20.3 – the InP shortage owner at a high, extended. Neither is a proposal. | MU -9.6% off high, mom1 +17.6, since-scan rel +4.3. LITE -0.7%, mom1 +20.3, since-scan rel +13.7. Fully paid |
| AVT | electronic component DISTRIBUTION — paid on the PRICE of the scarce good (HELD, exit AMBER, print contaminated) | Held. -0.8% off high, mom1 +14.8, since-scan rel +1.0. EXIT STAYS AMBER: Micron’s capex raise is the supply response that eventually ends price-driven distribution margins, but Micron also said 2027-28 will be TIGHTER than 2026 – the supply arrives with a lag measured in clean-room years. Hold, no add. | PASSES, HELD – -0.8% off high, mom1 +14.8, since-scan rel +1.0. Exit AMBER |
| CIEN | coherent optical systems and DATA-CENTER INTERCONNECT — reprieved by the calendar, not by the evidence (BENCH, final scan restated) | Scheduled to retire at the 10/1 score; instead it rose 10.4 rel since 9/28. But template is FALSE, below its SMA, -39.5% off high. A bounce is not a reclaim. RETIRED as scheduled; re-enters only on a template reclaim. | RETIRED – -39.5% off high, below SMA, FALSE, since-scan rel +10.4 (bounce) |
| NTAP | enterprise storage systems — stood down on 9/18 and a two-day scan may not un-stand it (STAND DOWN) | NTAP’s reported mom1 has flipped positive to +2.8 (rel +1.1) and its template is TRUE at 4.3% off its high. That is one overlapping reading four days after the third consecutive negative print that fired the trim rule, and it is not grounds to reverse anything. The logic was always sound — when memory is over half a server’s bill of materials, list prices are up 17-30% and OEMs respond with component downgrades, existing arrays get run longer and tiered rather than replaced — but the memory attribution was never broken out in NTAP’s disclosure and we have no coverage edge at $475M a day. Stood down. Returns only on two positive relative prints scored at least 10 sessions apart | STOOD DOWN, PRINT TURNED BUT NOT SCOREABLE — -4.3% off high, mom1 +2.8 (rel +1.1), mom3 +25.5, mom6 +97.6, mom12-1 +55.2. $475M/day. Template TRUE. No change |
| SWKS / STX / SNDK / WDC / MRVL / COHR | the makers and the RF layer — SWKS finally gave something back and we still have no mechanism for it | SKYWORKS was the best tape in the file for two consecutive scans and we said repeatedly that we had never underwritten a mechanism for it. It fell 2.8% (rel -4.5) over these two sessions while the group around it rose 6-12%, which is the first evidence that agrees with our refusal to reach for a story after the fact; it is still 2.8% off its high with a full template and a NEGATIVE twelve-month base of -8.9, so whatever it is, it is a re-rating off a low base and not a shortage we identified. The storage and optics complex around it rose hard: SNDK +9.4 (rel +7.8) though still 24.3% off its high on a twelve-month base of +1,519% that makes it uninvestable on the modest-momentum leg, STX +9.2 (rel +7.6), WDC +5.7 (rel +4.1), COHR +8.6 (rel +7.0), MRVL +6.9 (rel +5.2). The fund holds a small STX position ($6.9k, -10.0%) whose template is now TRUE again. None of this is scoreable and none of it is a proposal | MIXED, NOT SCOREABLE — SWKS -2.8% off high, since-scan -2.8 (rel -4.5), mom12-1 -8.9, template TRUE. STX -19.8%, since-scan +9.2 (rel +7.6), template TRUE, HELD. SNDK -24.3%, +9.4 (rel +7.8), mom12-1 +1,518.9. WDC -39.9%, +5.7. COHR -24.7%, +8.6, FALSE. MRVL -18.6%, +6.9, TRUE |
| PACKAGING TIER | UCTT / ICHR / MKSI / ONTO / ACLS / KLIC / CAMT / AMKR — screened for the first time because memory capex is shifting to hybrid bonding, and it is a wreck (AVOID) | We screened this tier for the first time this scan, and the reason was a specific disclosed fact: TrendForce describes the character of 2026-27 memory capex as shifting away from raw capacity toward process upgrades, HIGHER-LAYER STACKING and HYBRID BONDING. That should pay the advanced-packaging and subfab equipment names before it pays raw front-end capacity. IT IS PAYING NOBODY. UCTT is 48.8% off its high, ICHR 50.0%, MKSI 42.2%, ACLS 40.9%, AMKR 44.1%, KLIC 37.0%, ONTO 27.1%, CAMT 27.9%, VECO 46.7%, AEHR 33.4% — every single template FALSE. The only two passes are COHU (template TRUE, 18.0% off its high, mom1 +12.0) at just $51M a day, which is too thin for this book to size, and ENTG, which is a consumable rather than a packaging tool and is recorded separately above. A tier that should lead a stacking-led capex cycle and is instead 30-50% off its highs is either the next wave being ignored or evidence that the capex cycle is not believed. Recorded so we do not rediscover it; no proposal | BROKEN, ENTIRE TIER — UCTT -48.8% off high, ICHR -50.0%, MKSI -42.2%, ACLS -40.9%, AMKR -44.1%, KLIC -37.0%, ONTO -27.1%, CAMT -27.9%, VECO -46.7%, AEHR -33.4%, all templates FALSE. Passes: COHU (TRUE, -18.0%, $51M/day, too thin), ENTG (TRUE, -22.1%, watch) |
Crowded consensus (beta, not alpha): The two-session rotation ran straight into this theme and it was the most concentrated move in the file: LRCX +12.2, CRDO +11.3, AMAT +11.2, SNDK +9.4, MPWR +9.2, STX +9.2, KLAC +8.9, COHR +8.6, TER +8.1, MRVL +6.9, LITE +6.8, MU +6.8, WDC +5.7, CIEN +6.3, against SPY +1.7. The exceptions are informative: SWKS, the best tape of the last two scans and the one we never had a mechanism for, FELL 2.8%; and TECK, the theme’s proposal, rose 2.4% while the whole materials complex around it fell 5-9%. WHAT IS NOT CROWDED IS THE PART OF THIS THEME THAT SHOULD BENEFIT FROM THE DISCLOSED DIRECTION OF THE SPEND: the advanced-packaging and subfab tier that hybrid bonding and higher-layer stacking ought to pay is 27-50% off its highs with every template FALSE, and the per-wafer consumable (ENTG) sits 22.1% off its high with mom3 -22.1. Either the market does not believe the capex numbers it just bought the tool makers on, or this is where the file should be standing in a month.
Wave map: PAID AND UNOWNABLE: the constraint owners. MU (FQ4 guided to $50.0B revenue, 86% gross margin, $31.00 EPS, HBM booked through 2027 into 2028) and LITE (shipping 30%+ below demand across five InP fabs). Twelve-month bases of +478% and +413% disqualify both on the modest-momentum leg. Permanently removed from candidate status; retained as context →→ PAID ON PRICE, EXIT AMBER: distribution. AVT, with ~one third of its growth attributed by management to memory price increases, held but not added, and a funded capacity response is exactly what ends a price-driven run →→ THE CURE, NOW FUNDED AND NO LONGER IGNORABLE: the tool chain. WFE +31% in 2026 and +38% in 2027 per JP Morgan; Micron FY27 capex above $45B against ~$20B in FY26; DRAM industry capex $53.7B to $61.3B. AMAT, LRCX, KLAC, TER, CRDO rose 8-12% in two sessions on it and all remain 21-39% off their highs with templates FALSE. AVOID stands with a written trigger →→ WHERE THE CHARACTER OF THE SPEND ACTUALLY POINTS, AND NOBODY IS THERE: consumables and advanced packaging. The disclosed shift is to process upgrades, higher-layer stacking and hybrid bonding — more STEPS per wafer, not more wafers. ENTG (per-wafer toll, template TRUE, 22.1% off high, mom3 -22.1) is the watch candidate. The packaging equipment tier around it is 27-50% off its highs with every template FALSE →→ THE MATERIAL BENEATH THE LASER: TECK, the one large Western-listed separator of high-purity indium and germanium, with up to $400M of Canadian federal and Growth Fund money to double germanium capacity against Chinese indium exports to the US down 77%. Proposal carried from 9/18, not scoreable today
- 10/2 FRIDAY NOTE – THE SCORING SCAN (10/1 close = 10 sessions after the 9/17 print, so the 9/18 proposals and the ETN/NVT/APH prints are legally scoreable for the first time). Tape since the 9/28 close: SPY -0.2, QQQ +0.7. The 10-year touched 5.344% on 10/1 – highest since 2002 – then reversed to ~5.21%; S&P +0.19% to 7,666. Brent ~$97-104 intraday, Hormuz still disrupted. A stopgap CR averted an October shutdown. MICRON (9/30): Q4 revenue $54.2B (guided $50.0B), EPS $33.42, Q1 FY27 guide $61.5B at ~86% GM; FY27 capex RAISED above prior plans (Q1 ~$11.5B, H1 ~$25B, heavier construction in H2) on ~$150B of long-term agreements. MU faded after hours on the capex; the tool chain did not. CENSUS C30 AUGUST (10/1): data-center construction $85B SAAR, +7.5% m/m, +73% y/y, ~149% annualized since March; power construction only +8.5% y/y; total construction unexpectedly up.
- TECK FAILED its first legal score (rel mom1 -2.7) and the proposal is WITHDRAWN. CIEN retired as scheduled despite a bounce. The memory theme’s payout has moved down the chain to the tool makers and consumables (see the AI-build theme); the constraint owner is now the funding source, not the trade.
- 9/29 TUESDAY NOTE (one session, nothing scoreable): the 10-year made a fresh 19-year high at 5.24-5.27% intraday on 9/28, 30-year 5.55%; SPY -0.7, QQQ -1.1, VIX 16.1; WTI back to ~$93 (from >$103). Fed governor Cook: AI and oil keep inflation up, further hikes data-dependent. The rate-shock sort continued in one day: borrowers and long-duration fell (ORCL -3.3, BE -8.9, VRT -3.6, CRDO -8.7, AMD -3.6) while cash-rich names rose (NVDA +1.7 on a new $150B buyback). Scoring calendar unchanged: NDSN/TECK/ETN/NVT/APH at the 10/1 close (Fri 10/2 scan); ENTG at the 10/5 close (Tue 10/6 scan). MICRON reports WED 9/30 after the close.
- MICRON 9/30 IS THE WEEK’S ONLY DATED CATALYST FOR THIS THEME. What to read: FY27 capex (the tool-chain and consumables thesis), HBM4 booking into 2028, and any hint of DRAM contract price moderation (the AVT exit condition). Do not trade the print; read it into the 10/2 scan.
- AMD’S 10% Q4 PRICE HIKE (reported 9/18 via TrendForce, TSMC cost pass-through) is the pricing-power test passed at the compute layer, and it came in the same week the 10-year hit 5.13%. Price-setters absorbed the rate shock; borrowers did not.
- The packaging/back-end tier bounced hard (VECO +12.3, COHU +10.5, ACLS +8.8, KLIC +8.4, UCTT +6.4) but every name except COHU still fails the template, 24-48% off highs. A bounce off a wreck is not a signature; scoreable from 10/1.
- THE OVERLAPPING-WINDOW BUG: A TUESDAY SCAN CANNOT CONFIRM OR REFUTE A FRIDAY SCAN, AND THIS FILE ALMOST ACTED ON THE ARTIFACT. Last scan measured to the 2026-09-17 close; this one measures to 2026-09-21. That is TWO TRADING SESSIONS. A 21-bar mom1 window therefore shares 19 OF ITS 21 BARS with the print we already recorded, so the two numbers are not two observations, they are one observation read twice. The proof is NDSN. Nordson ROSE 0.8% over the two sessions and its mom1 relative print FELL from +3.4 to -7.5 — an 8.6-point collapse produced entirely by the two bars that dropped off the BACK of the window, which were worth +10.0%. Had we read this naively we would have recorded ‘NDSN failed on its third print’ and killed a proposal because the stock went UP. The same artifact runs the other way and is just as fake: ETN’s mom1 went -5.1 to +4.8, but 3.7 of those 9.9 points are dropped bars; AVT went +0.1 to +10.4 with 7.0 points of it dropped bars; ANET went +3.3 to +11.8 with 4.9 points dropped. THIS IS THE SAME DISEASE AS THE 9/11 mom3 FINDING (a fixed 64-bar window faking +3 to +13 points of momentum market-wide), in a new place, and we caught it on a name we had already proposed. THE FIX, EFFECTIVE NOW AND BINDING: (1) A CONFIRMATION OR A FAILURE MAY ONLY BE SCORED ON A SCAN AT LEAST 10 TRADING SESSIONS AFTER THE PRINT IT IS TESTING. Tuesday scans four calendar days after a Friday scan score NOTHING — they cannot satisfy a trigger and they cannot reset one. (2) The genuinely NEW information in a short-gap scan is the RELATIVE RETURN SINCE THE PRIOR SCAN, which shares no bars with anything. That number, and the news, are what a Tuesday scan is for. (3) Going forward a two-scan confirmation requires BOTH a positive relative mom1 AND a positive relative return since the previous scan, so that every confirmation contains at least some non-overlapping evidence. CONSEQUENCE FOR THIS SCAN: NDSN and TECK, proposed on 9/18, are NEITHER confirmed NOR failed today. Their triggers are CARRIED UNCHANGED to the Friday scan. Nothing else may be scored either.
Grid Equipment Bottleneck — ORDERS DOUBLED, UTILITIES FELL 5%, AND THE DIVERGENCE NOW HAS A NAME: THE COST OF CAPITAL. GE Vernova’s data-center electrification orders in one quarter exceeded all of 2025; the regulated utilities that must finance the grid build fell into a 5.13% 10-year. GAP: last reading 9/21 close, this one 9/25 close = FOUR sessions; the 9/18 proposals’ print (9/17) is SIX sessions old. Neither reaches the 10-session scoring floor. The Friday 9/25 scan never wrote to this file, so this Sunday catch-up is the only reading of the week. — THE PHYSICAL NUMBERS, WHICH ARE THE POINT OF THIS THEME, DETERIORATED AGAIN. IEA’s 2026 Grid Report puts GLOBAL TRANSFORMER MANUFACTURING CAPACITY AT 98% UTILIZATION with backlogs still growing. US large power transformer lead times have gone from 12-18 months in 2019 to FOUR TO FIVE YEARS in 2026 — 128 weeks for standard units and 144 WEEKS for generator step-up transformers. High-voltage circuit breakers run roughly 125 WEEKS. Medium-voltage switchgear, substation-class transformers, large generators, UPS systems and automatic transfer switches all sit at TWO TO FOUR TIMES pre-pandemic lead times, and SWITCHGEAR IS EFFECTIVELY SOLD OUT THROUGH 2028. THE INPUT BENEATH IT IS THE REAL CHOKE AND IT IS BARELY COVERED: the United States produces only about 20% OF ITS OWN GRAIN-ORIENTED ELECTRICAL STEEL; Japan, Korea and Germany hold roughly 70% of global GOES output; one of the largest mills (NLMK, Russia) was sanctioned in 2022; GOES capacity additions take two to three years; and electrical steel prices are up more than 20% in recent months. THE CONSEQUENCE IS NOW BEING QUANTIFIED IN PROJECTS RATHER THAN IN WEEKS: 30-50% OF 2026 DATA CENTERS FACE DELAY from the transformer and electrical equipment shortage, and nearly half of planned US developments could be delayed or cancelled on power infrastructure. Layer the demand on top — six hyperscalers at $1.3 TRILLION of 2027 capex, ~$1T on the Moody’s, JP Morgan and Goldman numbers, and a $1.3 TRILLION US utility capex programme across 2026-2030 — and this is, on paper, the most acute and best-documented physical shortage in the entire file. Scan: 2026-10-02
| Ticker | Tier | Evidence | Momentum |
|---|---|---|---|
| HUBB / AZZ / NVT / VRT / ETN / GEV | transformers, switchgear, busway and e-houses — the shortage worsened and the equities did not care (BROKEN, no adds) | Two names moved and neither can be scored: ETN +6.3% (rel +4.6) with its template TRUE at 5.3% off its high, and NVT +5.4% (rel +3.7) with template TRUE at 13.4% off. ETN’s headline mom1 improvement from -5.1 to +4.8 is 3.7 points dropped bars and must not be quoted as a repair. The rest are unchanged and broken: HUBB 17.5% off its high with mom1 -2.3 and template FALSE, AZZ 16.4% off with mom1 -4.2 and FALSE, VRT 33.3% off with mom1 -5.2 and FALSE, GEV 19.5% off with mom1 -2.0 and FALSE. THE CONTRADICTION IS THE FINDING: this scan produced the strongest physical shortage evidence the theme has ever carried — 98% global transformer utilization, four-to-five-year lead times, switchgear sold out through 2028, 30-50% of 2026 data centers facing delay — and not one listed owner of that shortage holds a trend. Two readings remain open and we cannot yet choose between them: either the market is pricing these as fixed-margin order-takers whose backlog inflates with their input costs (the same amendment that explains the EPC tier), or this is a genuine dislocation. We do not add on an open question | BROKEN WITH TWO EXCEPTIONS, NOTHING SCOREABLE — ETN -5.3% off high, since-scan +6.3 (rel +4.6), template TRUE. NVT -13.4%, +5.4 (rel +3.7), TRUE. HUBB -17.5%, mom1 -2.3, FALSE. AZZ -16.4%, -4.2, FALSE. VRT -33.3%, -5.2, FALSE. GEV -19.5%, -2.0, FALSE |
| POWL | custom switchgear and BEHIND-THE-METER electrical houses — the single best disclosed order in the file this week (REPAIR BENCH, TOP) | Powell takes the top of this theme’s bench on the strength of a disclosure that states our wave-4 thesis as a customer’s purchase order. Q3 FY26: a company-record backlog of $2.4 BILLION with project visibility through fiscal 2028, new orders +158% YEAR-OVER-YEAR to $934 MILLION, a quarterly BOOK-TO-BILL OF 3.0x and 2.2x fiscal year to date, and — the part that matters — DATA-CENTER ORDERS EXCEEDING $400 MILLION REPRESENTING PHASE 1 OF A MULTIPHASE BEHIND-THE-METER DESIGN FOR ON-SITE GENERATION ASSETS. That is the self-supply reroute being bought in stages by a named customer, against a grid where switchgear is sold out through 2028 and transformers take four to five years. AND THE TAPE IS STILL WRECKED: 41.7% off its 52-week high, mom1 -5.0 (rel -6.7), mom3 -39.0, template FALSE, on $100M a day. It rose 6.1% (rel +4.4) in the two sessions, which is news and not a score. Trigger, non-overlapping: reclaim the full trend template, then two positive relative prints at least 10 sessions apart | BROKEN, BEST DISCLOSURE, TOP OF BENCH — -41.7% off high, since-scan +6.1 (rel +4.4), mom1 -5.0 (rel -6.7), mom3 -39.0, mom6 +9.1, mom12-1 +95.0. Vol 61, $100M/day. Template FALSE |
| CLF | the ONLY US producer of GRAIN-ORIENTED ELECTRICAL STEEL — the input two tiers beneath the transformer (BENCH, and it just fell hard) | This scan finally quantified why CLF sits on this bench: the US produces only about 20% OF ITS OWN grain-oriented electrical steel, Japan, Korea and Germany hold ~70% of world output, one of the largest mills was sanctioned in 2022, capacity additions take two to three years, and electrical steel prices are up more than 20% in recent months — into global transformer capacity running at 98% utilization. Cleveland-Cliffs is the sole domestic GOES producer and is therefore the most direct listed expression of the tightest input in the theme. THE TAPE DOES NOT AGREE AND JUST GOT WORSE: CLF FELL 5.4% (rel -7.1) over the two sessions in the worst group in the file, is 25.4% off its high, and its template is FALSE. Its headline mom1 of +12.4 is a rear-view number that is being unwound in real time. The steel complex around it was sold hard (NUE -8.6, STLD -6.3, CMC), which reads as a macro or a rate move rather than anything about GOES. Stays on the bench with the standard non-overlapping trigger; a 12.4-point mom1 attached to a name that just dropped 5.4% is precisely the kind of number this scan exists to distrust | BROKEN AND DETERIORATING — -25.4% off high, since-scan -5.4 (rel -7.1), mom1 +12.4 (stale), mom3 +1.5, mom6 +54.3, mom12-1 -6.0. Vol 68, $155M/day. Template FALSE. Bench |
| FCX / SCCO / TECK / NUE / STLD | copper and the metals beneath the conductor — the whole complex was the worst group in the file (no adds) | Copper holds and steel does not. FCX is 9.7% off its high with a template TRUE and mom1 +1.3, SCCO 9.8% off and TRUE, TECK 7.1% off and TRUE — but the ferrous side was the single worst group over the two sessions: NUE -8.6% (rel -10.3), STLD -6.3% (rel -7.9), CLF -5.4% (rel -7.1), with CF -7.9% alongside. NUE and STLD still hold templates on six-month momentum of +53.8 and +41.0, so this is a sharp give-back rather than a break, but it removes any temptation to treat the electrical-steel story as a tape we can act on today. TECK is theme 2’s carried proposal and rose 2.4% through the sell-off, which is the right kind of divergence and still not scoreable | COPPER HOLDS, STEEL SOLD OFF — FCX -9.7% off high, mom1 +1.3, TRUE. SCCO -9.8%, -0.3, TRUE. TECK -7.1%, +1.1, TRUE, since-scan +2.4 (rel +0.7). NUE -11.8%, since-scan -8.6 (rel -10.3), TRUE. STLD -18.5%, -6.3 (rel -7.9), TRUE. CLF -25.4%, -5.4, FALSE |
| UTILITIES / IPP | the regulated rate base and merchant power — a $1.3 TRILLION capex programme and still nothing to own (AVOID, unchanged) | Unchanged from 9/18 and not re-screened in full, because four calendar days cannot move a regulated rate base. S&P Global Regulatory Research Associates forecasts roughly $1.3 TRILLION of aggregate US energy-utility capex across 2026-2030, a record, and on 9/18 every single regulated name failed the trend template with a negative one-month relative print (AEP, D, SO, ETR, PPL, EXC, XEL, NEE, PEG, DTE, WEC, AEE, NI, CNP, EVRG, OGE, POR, IDA) while the merchant half was a wreck (VST 33.8% off its high, CEG -34.6%, TLN -34.2%, NRG -41.9%). The Senate block of the Ratepayer Protection Act removes the specific de-risking catalyst we noted for the regulated names; the state tariff layer partially replaces it. Either way the readings stand: bond proxies in a hike cycle that is not finished, and an IPP power-price trade that is over. NOT A PLACE TO STAND | BROKEN, ENTIRE COMPLEX, UNCHANGED — every regulated name template FALSE as of 9/18; IPPs VST -33.8% off high, CEG -34.6%, TLN -34.2%, NRG -41.9%. Avoid |
| SKILLED LABOR / HIGH-VOLTAGE OEMs | the people who actually install it — a 499,000-worker shortfall that is a COST, not a margin (a rule, not a ticker) | The labour data found this scan is dramatic and must not be misread. The sector faces a shortfall of up to 499,000 WORKERS; electrical work is 45-70% OF TOTAL DATA-CENTER CONSTRUCTION COST per the IBEW; Microsoft’s president has called the electrician shortage the single biggest bottleneck slowing data-center expansion; Northern Virginia journeyman wiremen are at $72-75 an hour all-in under the IBEW Local 26 agreement with data-center specialists north of $280k; and the shortage now extends to mechanical contractors, controls specialists, commissioning teams and high-voltage field talent, making labour a SITE-SELECTION CONSTRAINT. THE INVESTABLE CONCLUSION IS NEGATIVE, NOT POSITIVE, AND IT FOLLOWS DIRECTLY FROM OUR OWN AMENDMENT: the scarce asset is the labour, and the contractor BUYS it. Scarcity in a purchased input is cost inflation. That is why a record pipeline has coincided with nine weeks of de-rating across PWR, EME, FIX, IESC, STRL, DY and MTZ. There is no listed pure-play on the wage itself at a size this book can carry. Recorded as a rule | n/a — discipline note. The contractors that buy this labour: PWR -18.2% off high, EME -19.7%, FIX -21.0%, IESC -19.0%, STRL -48.4%, DY -45.5%, MTZ -49.1% |
Crowded consensus (beta, not alpha): Nothing in this theme is crowded and that is the whole problem. The best-documented physical shortage in the file — transformers at 98% global capacity utilization and four-to-five-year lead times, switchgear sold out through 2028, high-voltage breakers at 125 weeks, US grain-oriented electrical steel self-supply at ~20% with prices up 20%, and 30-50% of 2026 data centers facing delay as a result — is owned by a set of equities that are uniformly 13-42% off their highs with templates FALSE. Over the two sessions since the last scan only ETN (+6.3, rel +4.6), POWL (+6.1, rel +4.4) and NVT (+5.4, rel +3.7) moved, and none of that is scoreable. Meanwhile the money went to the semiconductor tool chain. The uncomfortable possibility this file has to keep open is that a documented shortage in a COMPETITIVE, FIXED-MARGIN manufacturing tier produces backlog rather than earnings, and that backlog measured in years is a working-capital problem as much as it is a revenue opportunity.
Wave map: SPENT: the interconnection queue and the permits. 474 GW queued, four-to-seven-year waits, state moratoria. Pays nobody directly; forces everything downstream →→ THE DOCUMENTED CHOKE, AND NOBODY IS BEING PAID FOR IT: transformers and switchgear. 98% global capacity utilization, 4-5 year US lead times, switchgear sold out through 2028, breakers at 125 weeks, 30-50% of 2026 data centers facing delay. HUBB, AZZ, VRT, GEV all FALSE; only ETN and NVT hold templates and neither is scoreable →→ THE INPUT BENEATH THE CHOKE, WHICH IS WHERE THE AMENDMENT POINTS: grain-oriented electrical steel and copper. US GOES self-supply ~20%, 70% of world output in Japan/Korea/Germany, a sanctioned Russian mill, 2-3 year capacity lead times, prices +20%. CLF is the sole US producer and is 25.4% off its high and FALSE after falling 5.4% this week. Copper holds: FCX, SCCO, TECK all template TRUE →→ THE REROUTE AROUND THE CHOKE: behind-the-meter switchgear and on-site generation. POWL’s $400M+ phase-1 behind-the-meter data-center order inside a record $2.4B backlog at a 3.0x book-to-bill is the best disclosure in the theme, and the stock is 41.7% off its high. Federal catalyst dead in the Senate; state large-load tariffs (~24-25 states approved) alive and spreading →→ NOT HERE. Over the two sessions since the last scan the money went to the semiconductor tool chain and memory (LRCX +12.2, AMAT +11.2, SNDK +9.4, MU +6.8). This theme’s best two-day movers were ETN +6.3, POWL +6.1 and NVT +5.4, none of which is scoreable
- 10/2 FRIDAY NOTE – THE SCORING SCAN (10/1 close = 10 sessions after the 9/17 print, so the 9/18 proposals and the ETN/NVT/APH prints are legally scoreable for the first time). Tape since the 9/28 close: SPY -0.2, QQQ +0.7. The 10-year touched 5.344% on 10/1 – highest since 2002 – then reversed to ~5.21%; S&P +0.19% to 7,666. Brent ~$97-104 intraday, Hormuz still disrupted. A stopgap CR averted an October shutdown. MICRON (9/30): Q4 revenue $54.2B (guided $50.0B), EPS $33.42, Q1 FY27 guide $61.5B at ~86% GM; FY27 capex RAISED above prior plans (Q1 ~$11.5B, H1 ~$25B, heavier construction in H2) on ~$150B of long-term agreements. MU faded after hours on the capex; the tool chain did not. CENSUS C30 AUGUST (10/1): data-center construction $85B SAAR, +7.5% m/m, +73% y/y, ~149% annualized since March; power construction only +8.5% y/y; total construction unexpectedly up.
- GRID 10/2: C30 power construction +8.5% y/y vs data centers +73% – the supply build is not keeping up, which is the theme. Forgent Q3 orders +268%, $1.5B backlog (~18 months). ETN and NVT passed print 1 of 2 (see theme 0). Utilities got a one-day reprieve as the 10y reversed from 5.34%; the cost-of-capital argument is unchanged with the 10y above 5.2%. CLF +4.0 rel on 10/1 is one day, still -10.5 rel since 9/17.
- 9/29 TUESDAY NOTE (one session, nothing scoreable): the 10-year made a fresh 19-year high at 5.24-5.27% intraday on 9/28, 30-year 5.55%; SPY -0.7, QQQ -1.1, VIX 16.1; WTI back to ~$93 (from >$103). Fed governor Cook: AI and oil keep inflation up, further hikes data-dependent. The rate-shock sort continued in one day: borrowers and long-duration fell (ORCL -3.3, BE -8.9, VRT -3.6, CRDO -8.7, AMD -3.6) while cash-rich names rose (NVDA +1.7 on a new $150B buyback). Scoring calendar unchanged: NDSN/TECK/ETN/NVT/APH at the 10/1 close (Fri 10/2 scan); ENTG at the 10/5 close (Tue 10/6 scan). MICRON reports WED 9/30 after the close.
- The four-scan puzzle (‘a real, worsening shortage whose every listed owner is sold’) got its likeliest answer this week: the grid build is financed by the regulated rate base and IPP balance sheets, and long yields at 20-year highs raise that financing cost faster than the shortage raises allowed returns. XLU -5.2% since 9/17; utilities the worst S&P sector. The equipment makers that sell into the shortage without financing it (NVT +8.6 since 9/17, ETN) are the ones repairing.
- Texas’s permit halt and the ERCOT audit add a regulatory layer to the grid-connection queue; the audit has no end date. Expect utilities with Texas large-load exposure to trade on audit headlines through at least TCEQ’s 10/19 report-back.
- THE OVERLAPPING-WINDOW BUG: A TUESDAY SCAN CANNOT CONFIRM OR REFUTE A FRIDAY SCAN, AND THIS FILE ALMOST ACTED ON THE ARTIFACT. Last scan measured to the 2026-09-17 close; this one measures to 2026-09-21. That is TWO TRADING SESSIONS. A 21-bar mom1 window therefore shares 19 OF ITS 21 BARS with the print we already recorded, so the two numbers are not two observations, they are one observation read twice. The proof is NDSN. Nordson ROSE 0.8% over the two sessions and its mom1 relative print FELL from +3.4 to -7.5 — an 8.6-point collapse produced entirely by the two bars that dropped off the BACK of the window, which were worth +10.0%. Had we read this naively we would have recorded ‘NDSN failed on its third print’ and killed a proposal because the stock went UP. The same artifact runs the other way and is just as fake: ETN’s mom1 went -5.1 to +4.8, but 3.7 of those 9.9 points are dropped bars; AVT went +0.1 to +10.4 with 7.0 points of it dropped bars; ANET went +3.3 to +11.8 with 4.9 points dropped. THIS IS THE SAME DISEASE AS THE 9/11 mom3 FINDING (a fixed 64-bar window faking +3 to +13 points of momentum market-wide), in a new place, and we caught it on a name we had already proposed. THE FIX, EFFECTIVE NOW AND BINDING: (1) A CONFIRMATION OR A FAILURE MAY ONLY BE SCORED ON A SCAN AT LEAST 10 TRADING SESSIONS AFTER THE PRINT IT IS TESTING. Tuesday scans four calendar days after a Friday scan score NOTHING — they cannot satisfy a trigger and they cannot reset one. (2) The genuinely NEW information in a short-gap scan is the RELATIVE RETURN SINCE THE PRIOR SCAN, which shares no bars with anything. That number, and the news, are what a Tuesday scan is for. (3) Going forward a two-scan confirmation requires BOTH a positive relative mom1 AND a positive relative return since the previous scan, so that every confirmation contains at least some non-overlapping evidence. CONSEQUENCE FOR THIS SCAN: NDSN and TECK, proposed on 9/18, are NEITHER confirmed NOR failed today. Their triggers are CARRIED UNCHANGED to the Friday scan. Nothing else may be scored either.
Munitions & Missile Rearmament — STOOD DOWN A FOURTH WEEK; the primes kept falling (GD -4.9, HII -4.4, LHX -3.7, NOC -3.1 in four sessions) and ATI’s trim rule stands. GAP: last reading 9/21 close, this one 9/25 close = FOUR sessions; the 9/18 proposals’ print (9/17) is SIX sessions old. Neither reaches the 10-session scoring floor. The Friday 9/25 scan never wrote to this file, so this Sunday catch-up is the only reading of the week. — NO NEW PROGRAMME DATA THIS SCAN AND NONE IS EXPECTED IN A FOUR-DAY GAP; THE TAPE IS THE ONLY NEW INFORMATION AND IT IS BAD. Against SPY +1.7% over the two sessions and measured on the 21-day window, the defence complex is the weakest broad group in the entire file after the EPC tier: KRMN 69.5% off its high with mom1 -35.2 (rel -36.9), KTOS 62.8% off with -13.5, AVAV 60.7% off, HII 38.6% off, LDOS 35.0% off, LHX 33.9% off with -8.3, MRCY 32.1% off, NOC 30.8% off with -6.2, CW 28.3% off, ESE 26.0% off, VSEC 26.0% off with -16.0, AIR 24.4% off with -12.2, TDG 23.5% off, HWM 22.3% off with -16.9, HEI 19.1% off with -13.6, DCO 18.8% off with -11.1, ATI 18.4% off with -9.6, GD 10.6% off with -8.3, CACI 6.9% off. ONLY TWO NAMES IN THE ENTIRE COMPLEX HOLD A FULL TREND TEMPLATE — ATI and DCO — and both print double-digit negative one-month relative numbers. This is not rotation, it is a sector-wide de-rating, and it has now run for three consecutive scans. Scan: 2026-10-02
| Ticker | Tier | Evidence | Momentum |
|---|---|---|---|
| ATI | titanium and nickel superalloys — TRIM RULE FIRED 9/18, POSITION STILL HELD, RULE STANDS (ACTION: TRIM) | The trim rule fired on 9/18 on a third consecutive negative relative month and the fund still holds 172 shares worth $32.5k at -18.0%. Over the two sessions since, ATI rose 3.0% (rel +1.3). THAT IS NOT A REASON TO RENEGOTIATE AND WE ARE RECORDING IT EXPLICITLY SO THE REVIEW TASK IS NOT SURPRISED BY THE PRICE. A fired rule is not un-fired by a two-day move that our own measurement fix says cannot be scored; if we allow a bounce to reverse a trim we have no rules at all, only narration. The name still holds a full trend template at 18.4% off its high with mom1 -9.6 (rel -11.3) and mom12-1 +160.0, which is exactly the profile we have said we will not own: a fully-paid twelve-month base with deteriorating one-month relative strength. THE TRIM PROPOSAL STANDS AND IT IS THE FUND’S CLEANEST SOURCE OF CASH against a book running MINUS $87,145 | TRIM RULE STANDS – -17.5% off high, mom1 -4.9, since-scan rel +6.0 (one bounce), TRUE. Action: trim |
| DCO / AIR / HWM / HEI / CW | aerostructures, components and the aftermarket — stood down on 9/18 and deteriorating (STAND DOWN) | The stand-down holds and the evidence got worse rather than better. DCO still carries a full trend template at 18.8% off its high but prints mom1 -11.1 (rel -12.8). Around it the aftermarket and component tier broke badly: HWM mom1 -16.9 (rel -18.6) at 22.3% off its high, HEI -13.6 (rel -15.3) at 19.1% off, AIR -12.2 (rel -13.9) at 24.4% off, CW -12.4 (rel -14.1) at 28.3% off, VSEC -16.0 (rel -17.7). Templates FALSE on all but DCO. Nothing here is addable and nothing here is scoreable today | BROKEN — DCO -18.8% off high, mom1 -11.1 (rel -12.8), template TRUE. AIR -24.4%, -12.2, FALSE. HWM -22.3%, -16.9, FALSE. HEI -19.1%, -13.6, FALSE. CW -28.3%, -12.4, FALSE. VSEC -26.0%, -16.0, FALSE |
| CACI / LDOS / BAH | defence IT and services — the least damaged corner and still not a trend (WATCH) | CACI is the only defence name in the file within 10% of its high (6.9% off) and it carries mom3 +40.7 with mom1 -2.1 (rel -3.8) and a template that is still FALSE. BAH prints mom1 +1.3 with mom3 +23.1 at 23.4% off its high; LDOS mom1 -9.0 with mom3 +22.7 at 35.0% off. The pattern across all three — a strong three-month number with a weak one-month number — is the same shape the June base effect produced elsewhere in this file, and we declined signals of exactly that shape on 9/8 and were right. No proposal | MIXED, TEMPLATES FALSE — CACI -6.9% off high, mom1 -2.1 (rel -3.8), mom3 +40.7, FALSE. BAH -23.4%, +1.3, mom3 +23.1, FALSE. LDOS -35.0%, -9.0, mom3 +22.7, FALSE |
| KRMN / AVAV / KTOS / MRCY / ESE / NOC / LHX / GD / HII | drones, missiles and the primes — a sector-wide de-rating, third consecutive scan (AVOID) | The drone and small-cap defence complex is the worst-performing group in the file that is not an EPC contractor, and the damage widened again. KRMN fell 35.2% in a month (rel -36.9) to 69.5% off its high, KTOS -13.5 to 62.8% off, AVAV +0.7 but still 60.7% off, MRCY -9.1 to 32.1% off, ESE -7.8 to 26.0% off. The primes are no refuge: NOC -6.2 at 30.8% off, LHX -8.3 at 33.9% off, GD -8.3 at 10.6% off, HII -8.0 at 38.6% off. Every template FALSE. Whatever rearmament thesis exists here, the market has spent three consecutive scans marking it down and we have no informational edge that contradicts it. AVOID | BROKEN, SECTOR-WIDE, THIRD SCAN — KRMN -69.5% off high (mom1 -35.2); KTOS -62.8% (-13.5); AVAV -60.7% (+0.7); HII -38.6% (-8.0); LHX -33.9% (-8.3); MRCY -32.1% (-9.1); NOC -30.8% (-6.2); ESE -26.0% (-7.8); GD -10.6% (-8.3). All templates FALSE |
Crowded consensus (beta, not alpha): Nothing in this theme is crowded and nothing in it is working. Across nineteen names screened, exactly two hold a full trend template (ATI and DCO) and both print double-digit negative one-month relative numbers; the median name is roughly 27% off its 52-week high. The money that left this complex went to the places named in theme 1: the semiconductor tool chain, memory and compute. The only position the fund holds here is ATI, and our own trim rule fired on it on 9/18.
Wave map: STOOD DOWN BY OUR OWN TRIM RULE: titanium, nickel and superalloys. ATI, held by the fund at $32.5k and -18.0%, template TRUE but mom1 -9.6 (rel -11.3) after three consecutive negative relative months. TRIM PROPOSED AND STANDING →→ BROKEN: forgings, castings and aerostructures. HWM -16.9 relative and 22.3% off its high, DCO -11.1 and 18.8% off, AIR -12.2 and 24.4% off, CW -12.4 and 28.3% off →→ BROKEN AND WORSE: subsystems, drones and munitions. KRMN -35.2 relative to 69.5% off its high, KTOS -13.5 to 62.8% off, AVAV 60.7% off, MRCY -9.1, ESE -7.8. Every template FALSE →→ BROKEN: the primes. NOC -6.2 at 30.8% off, LHX -8.3 at 33.9% off, GD -8.3, HII -8.0 at 38.6% off →→ LEAST DAMAGED, STILL NOT A TREND: defence IT and services. CACI 6.9% off its high with mom3 +40.7 but mom1 -2.1 and template FALSE; BAH and LDOS show the same strong-mom3 / weak-mom1 shape that the June base effect produced elsewhere and that we correctly declined on 9/8
- 10/2 FRIDAY NOTE – THE SCORING SCAN (10/1 close = 10 sessions after the 9/17 print, so the 9/18 proposals and the ETN/NVT/APH prints are legally scoreable for the first time). Tape since the 9/28 close: SPY -0.2, QQQ +0.7. The 10-year touched 5.344% on 10/1 – highest since 2002 – then reversed to ~5.21%; S&P +0.19% to 7,666. Brent ~$97-104 intraday, Hormuz still disrupted. A stopgap CR averted an October shutdown. MICRON (9/30): Q4 revenue $54.2B (guided $50.0B), EPS $33.42, Q1 FY27 guide $61.5B at ~86% GM; FY27 capex RAISED above prior plans (Q1 ~$11.5B, H1 ~$25B, heavier construction in H2) on ~$150B of long-term agreements. MU faded after hours on the capex; the tool chain did not. CENSUS C30 AUGUST (10/1): data-center construction $85B SAAR, +7.5% m/m, +73% y/y, ~149% annualized since March; power construction only +8.5% y/y; total construction unexpectedly up.
- DEFENCE 10/2: a stopgap CR funds programmes at FY26 levels and averts the October shutdown; the Pentagon is pushing a $1.5T multi-year munitions refill and $47B is in reconciliation, and production velocity – not authorization – is named as the binding constraint (interceptor deliveries slipping up to five years). The tape still disagrees: NOC -4.5 rel since 9/28, -8.4 since 9/17; LDOS -38.2% off high, BAH -30.3%, HII -39.0% (all FALSE). HII +6.9 rel since 9/28 is a bounce off -39%. Stand-down continues; no proposals.
- 9/29 TUESDAY NOTE (one session, nothing scoreable): the 10-year made a fresh 19-year high at 5.24-5.27% intraday on 9/28, 30-year 5.55%; SPY -0.7, QQQ -1.1, VIX 16.1; WTI back to ~$93 (from >$103). Fed governor Cook: AI and oil keep inflation up, further hikes data-dependent. The rate-shock sort continued in one day: borrowers and long-duration fell (ORCL -3.3, BE -8.9, VRT -3.6, CRDO -8.7, AMD -3.6) while cash-rich names rose (NVDA +1.7 on a new $150B buyback). Scoring calendar unchanged: NDSN/TECK/ETN/NVT/APH at the 10/1 close (Fri 10/2 scan); ENTG at the 10/5 close (Tue 10/6 scan). MICRON reports WED 9/30 after the close.
- No new catalyst worth a position: a $111M Sentinel award to NOC and AVAV’s earlier $465M LOCUST laser award did not move the group. Every drone and prime name except DCO fails the template; AVAV’s mom1 turned positive (+3.5) from 62.9% off its high, which is a bounce, not a trend.
- THE OVERLAPPING-WINDOW BUG: A TUESDAY SCAN CANNOT CONFIRM OR REFUTE A FRIDAY SCAN, AND THIS FILE ALMOST ACTED ON THE ARTIFACT. Last scan measured to the 2026-09-17 close; this one measures to 2026-09-21. That is TWO TRADING SESSIONS. A 21-bar mom1 window therefore shares 19 OF ITS 21 BARS with the print we already recorded, so the two numbers are not two observations, they are one observation read twice. The proof is NDSN. Nordson ROSE 0.8% over the two sessions and its mom1 relative print FELL from +3.4 to -7.5 — an 8.6-point collapse produced entirely by the two bars that dropped off the BACK of the window, which were worth +10.0%. Had we read this naively we would have recorded ‘NDSN failed on its third print’ and killed a proposal because the stock went UP. The same artifact runs the other way and is just as fake: ETN’s mom1 went -5.1 to +4.8, but 3.7 of those 9.9 points are dropped bars; AVT went +0.1 to +10.4 with 7.0 points of it dropped bars; ANET went +3.3 to +11.8 with 4.9 points dropped. THIS IS THE SAME DISEASE AS THE 9/11 mom3 FINDING (a fixed 64-bar window faking +3 to +13 points of momentum market-wide), in a new place, and we caught it on a name we had already proposed. THE FIX, EFFECTIVE NOW AND BINDING: (1) A CONFIRMATION OR A FAILURE MAY ONLY BE SCORED ON A SCAN AT LEAST 10 TRADING SESSIONS AFTER THE PRINT IT IS TESTING. Tuesday scans four calendar days after a Friday scan score NOTHING — they cannot satisfy a trigger and they cannot reset one. (2) The genuinely NEW information in a short-gap scan is the RELATIVE RETURN SINCE THE PRIOR SCAN, which shares no bars with anything. That number, and the news, are what a Tuesday scan is for. (3) Going forward a two-scan confirmation requires BOTH a positive relative mom1 AND a positive relative return since the previous scan, so that every confirmation contains at least some non-overlapping evidence. CONSEQUENCE FOR THIS SCAN: NDSN and TECK, proposed on 9/18, are NEITHER confirmed NOR failed today. Their triggers are CARRIED UNCHANGED to the Friday scan. Nothing else may be scored either.
- THE ONLY ACTION THIS THEME OWES THE REVIEW TASK IS THE ONE ALREADY WRITTEN: TRIM ATI. The rule fired on 9/18 on a third consecutive negative relative month. The position is still held at $32.5k and -18.0%, and it bounced 3.0% over the two sessions since. The bounce is not scoreable under this scan’s measurement fix and a fired rule is not un-fired by four calendar days. The fund is carrying MINUS $87,145 of cash, so every new position must be funded by a sale, and ATI is the cleanest sale we have identified: a fully-paid +160% twelve-month base with a -11.3 one-month relative print, inside a sector where seventeen of nineteen screened names have already lost their trend templates.
Energy Chokepoint / Strait of Hormuz — OIL SPIKED ABOVE $103 AND THE THEME STILL FELL: energy was the second-worst sector (-3.8%), tankers gave back (FRO -11.7% since 9/17, STNG -7.2%). The oil move is now feeding the bond sell-off rather than the equities. GAP: last reading 9/21 close, this one 9/25 close = FOUR sessions; the 9/18 proposals’ print (9/17) is SIX sessions old. Neither reaches the 10-session scoring floor. The Friday 9/25 scan never wrote to this file, so this Sunday catch-up is the only reading of the week. — NO NEW CHOKEPOINT DATA IN A FOUR-DAY GAP; THE STRUCTURE IS AS RECORDED AND THE TAPE IS THE ONLY NEW INFORMATION. The theme’s whole point is that the cash flows sit in a physical chokepoint rather than in a commodity forecast, and the equities continue to carry the strongest collective trend in the fund’s coverage: measured on the 21-day window, FRO +22.0 with a template TRUE at 3.1% off its high, INSW +15.4 at 1.8% off, VLO +15.2 at 4.8% off, MPC +12.3 at 5.3% off, STNG +10.0 at 1.9% off, PSX +9.1 at 4.5% off — six names, six full trend templates, every one within 6% of its 52-week high, with three-month prints of +63.2 (MPC), +61.9 (VLO), +56.2 (PSX), +29.7 (INSW) and +27.8 (FRO). THAT IS THE STRONGEST GROUP IN THE ENTIRE FILE ON EVERY TREND MEASURE. What is new is that all six gave back ground over the two sessions since the last scan while the AI capital-equipment complex rose 6-12%, which is a clean rotation out of real-asset cash flow. CF fell 7.9% (rel -9.6) with the fertiliser and steel complex and is 11.5% off its high with a template FALSE. GLNG is 9.6% off with a template FALSE. The fund holds MPC (130 shares, $51.2k, +26.9%), which is its best-performing equity position. Scan: 2026-10-02
| Ticker | Tier | Evidence | Momentum |
|---|---|---|---|
| MPC / VLO / PSX | refining — the crack spread as the toll on a constrained molecule (HELD, gave back, no action) | The fund’s MPC position is its best equity holding at +26.9% and this is the first scan in which the refiners have gone backwards: MPC -4.6% (rel -6.3), VLO -4.7% (rel -6.4), PSX -4.5% (rel -6.2) over two sessions. Set against what they did before it, this is small: all three still carry full trend templates, all three are within 5.3% of their 52-week highs, and the three-month prints are +63.2, +61.9 and +56.2. A give-back of five points after a sixty-point quarter is not a signal and our measurement fix explicitly forbids treating two sessions as one. NO ACTION, and specifically no trim on a two-day move — the exit condition for this position remains a broken template or a negative one-month relative print scored on a proper gap, neither of which exists | PASSES, GAVE BACK, TEMPLATES TRUE — MPC -5.3% off high, mom1 +12.3, since-scan -4.6 (rel -6.3), mom3 +63.2, mom6 +74.2, mom12-1 +96.8, HELD. VLO -4.8%, +15.2, since-scan -4.7. PSX -4.5%, +9.1, since-scan -4.5. All templates TRUE |
| FRO / INSW / STNG | crude and product tankers — the ton-mile toll (PASSES, extended, no adds) | The strongest sub-tier in the file on trend and the one with the least room: FRO is 3.1% off its high with mom1 +22.0 and mom6 +72.4, INSW 1.8% off with +15.4 and mom6 +79.1, STNG 1.9% off with +10.0 and mom6 +28.3, all three templates TRUE. All three gave back modestly over the two sessions (FRO -1.7, STNG -1.9, INSW flat to lower). The modest-momentum leg of our signature is long gone here — these are fully-paid positions in an uptrend, which makes them holds rather than entries, and the fund does not own them. The mechanism is unchanged: ton-miles lengthen when a chokepoint is contested, and the fleet cannot respond inside a scan cycle. No adds at these levels | PASSES, EXTENDED, TEMPLATES TRUE — FRO -3.1% off high, mom1 +22.0, mom6 +72.4, since-scan -1.7. INSW -1.8%, +15.4, mom6 +79.1. STNG -1.9%, +10.0, mom6 +28.3, since-scan -1.9. Not entries |
| ET / OKE / WMB / KMI / LNG | US midstream and LNG — the molecule’s path to both the export dock and the data centre (HELD, stalled) | The fund owns ET (1,956 shares, $41.0k, +2.3%) and it is stalled rather than broken: mom1 -1.5 (rel -3.2) at 3.8% off its high with a full template and annualized volatility of just 16, the lowest of any name the fund holds. OKE prints -3.0 at 5.9% off with a template TRUE. WMB (+0.7, 8.4% off), KMI (+0.3, 6.7% off) and LNG (-1.5, 7.1% off) all carry FALSE templates. This tier is the intersection of two themes — the export chokepoint and the data-centre gas build (US gas-fired generation development +50% to 378 GW in H1 2026, data-centre-tied gas nearly doubled to 189 GW) — and it is doing nothing in either. Hold ET, no adds, nothing scoreable | STALLED, TEMPLATES SPLIT — ET -3.8% off high, mom1 -1.5 (rel -3.2), vol 16, HELD, TRUE. OKE -5.9%, -3.0, TRUE. WMB -8.4%, +0.7, FALSE. KMI -6.7%, +0.3, FALSE. LNG -7.1%, -1.5, FALSE |
| CTRA / EQT / AR / RRC / EXE | gas producers — stood down on 9/18 and got worse (STAND DOWN, confirmed by deterioration) | CTRA was stood down on 9/18 on a third consecutive negative relative month and the tier around it deteriorated further: EQT mom1 -7.0 (rel -8.7) at 25.8% off its high, AR -9.8 (rel -11.5) at 24.8% off, RRC -6.8 at 20.2% off, EXE -8.8 at 27.4% off, every template FALSE. CTRA itself still holds a template at 10.3% off its high but prints -6.7 (rel -8.4). This is the tier that should be the direct beneficiary of 189 GW of data-centre-tied gas development and it is being sold across the board, which is the same pattern as the grid-equipment contradiction in theme 3: documented forward demand, no equity response. Stood down, no adds | BROKEN — CTRA -10.3% off high, mom1 -6.7 (rel -8.4), template TRUE, stood down. EQT -25.8%, -7.0, FALSE. AR -24.8%, -9.8, FALSE. RRC -20.2%, -6.8, FALSE. EXE -27.4%, -8.8, FALSE |
| CF / GLNG | nitrogen and floating LNG — the gas-price derivatives (BROKEN, no action) | CF fell 7.9% (rel -9.6) over the two sessions with the fertiliser and steel complex and is 11.5% off its high with mom1 -1.9 and a template FALSE, despite mom3 +19.6. GLNG is 9.6% off its high with mom1 -1.1 and template FALSE. Neither is held and neither is a candidate; they are recorded because they are the cleanest read-through from the gas price into equities and both are saying the same thing the producers are saying | BROKEN — CF -11.5% off high, mom1 -1.9, since-scan -7.9 (rel -9.6), mom3 +19.6, FALSE. GLNG -9.6%, -1.1, FALSE |
Crowded consensus (beta, not alpha): This theme WAS the crowd two sessions ago and stopped being it. On 9/18 refining and tankers were the file’s second destination for capital; since then VLO -4.7, MPC -4.6, PSX -4.5, STNG -1.9 and FRO -1.7 while the semiconductor tool chain rose 8-12%. That is a rotation, not a breakdown: six of the theme’s names still hold full trend templates within 6% of their 52-week highs with three-month prints of +28 to +63. The thing to watch is that the modest-momentum leg of our signature is entirely gone here — every working name in this theme is fully paid — which makes the whole complex a hold-and-trim set rather than a source of new positions, exactly as EXPD is in theme 1.
Wave map: PAID AND FULLY PRICED: the ton-mile toll. FRO 3.1% off its high with mom6 +72.4, INSW 1.8% off with +79.1, STNG 1.9% off with +28.3, all templates TRUE. Holds, not entries →→ PAID AND HELD, AND THE PART THAT JUST GAVE BACK: the crack spread. MPC (held, +26.9%), VLO and PSX, all templates TRUE and within 5.3% of their highs on three-month prints of +56 to +63, all down 4.5-4.7% over the two sessions since the last scan →→ STALLED BUT STRUCTURALLY RIGHT: the fee-on-volume midstream. ET (held, vol 16, template TRUE), OKE, TRGP and WES all hold templates with small off-high damage and negative one-month prints. This is the only tier in the chain paid a fee rather than a price, which is why it is the only one still holding trend →→ BROKEN AND STOOD DOWN DESPITE THE BEST FORWARD DEMAND NUMBER IN THE CHAIN: the gas producers. 189 GW of data-centre-tied gas development against ~2 GW operating, and EQT, AR, RRC, EXE and CTRA are all being marked down. Same shape as the grid-equipment contradiction in theme 3 →→ NOT HERE THIS WEEK. The money rotated out of refining, tankers, steel and fertiliser and into the semiconductor tool chain and memory in forty-eight hours
- 10/2 FRIDAY NOTE – THE SCORING SCAN (10/1 close = 10 sessions after the 9/17 print, so the 9/18 proposals and the ETN/NVT/APH prints are legally scoreable for the first time). Tape since the 9/28 close: SPY -0.2, QQQ +0.7. The 10-year touched 5.344% on 10/1 – highest since 2002 – then reversed to ~5.21%; S&P +0.19% to 7,666. Brent ~$97-104 intraday, Hormuz still disrupted. A stopgap CR averted an October shutdown. MICRON (9/30): Q4 revenue $54.2B (guided $50.0B), EPS $33.42, Q1 FY27 guide $61.5B at ~86% GM; FY27 capex RAISED above prior plans (Q1 ~$11.5B, H1 ~$25B, heavier construction in H2) on ~$150B of long-term agreements. MU faded after hours on the capex; the tool chain did not. CENSUS C30 AUGUST (10/1): data-center construction $85B SAAR, +7.5% m/m, +73% y/y, ~149% annualized since March; power construction only +8.5% y/y; total construction unexpectedly up.
- ENERGY 10/2: THE THEME CAME BACK. Since 9/28 rel: MPC +8.1, VLO +5.1, PSX +4.4, FRO +7.4, INSW +6.0, STNG +3.3. Refiners -1 to -4% off highs with mom3 +50-58 (extended, held); FRO and INSW AT their highs (mom1 +24/+21, extended, no adds). Analysts raising oil forecasts as Hormuz normalization hopes fade; Brent swung ~$97-104 on 10/1. Midstream steadied (TRGP +0.9 rel since 9/28, ET +0.2) but OKE and WES templates FALSE. ET held, mom1 -6.2 – watch it. Nothing new to propose: the names that pass are extended.
- 9/29 TUESDAY NOTE (one session, nothing scoreable): the 10-year made a fresh 19-year high at 5.24-5.27% intraday on 9/28, 30-year 5.55%; SPY -0.7, QQQ -1.1, VIX 16.1; WTI back to ~$93 (from >$103). Fed governor Cook: AI and oil keep inflation up, further hikes data-dependent. The rate-shock sort continued in one day: borrowers and long-duration fell (ORCL -3.3, BE -8.9, VRT -3.6, CRDO -8.7, AMD -3.6) while cash-rich names rose (NVDA +1.7 on a new $150B buyback). Scoring calendar unchanged: NDSN/TECK/ETN/NVT/APH at the 10/1 close (Fri 10/2 scan); ENTG at the 10/5 close (Tue 10/6 scan). MICRON reports WED 9/30 after the close.
- Refiners and tankers remain template TRUE but have given back two scans running (MPC -2.2, VLO -1.5, PSX -2.3, FRO -4.2, STNG -5.5 in four sessions). Held refiner position: no action until a scoreable print. Gas producers and midstream sold with rates.
- THE OVERLAPPING-WINDOW BUG: A TUESDAY SCAN CANNOT CONFIRM OR REFUTE A FRIDAY SCAN, AND THIS FILE ALMOST ACTED ON THE ARTIFACT. Last scan measured to the 2026-09-17 close; this one measures to 2026-09-21. That is TWO TRADING SESSIONS. A 21-bar mom1 window therefore shares 19 OF ITS 21 BARS with the print we already recorded, so the two numbers are not two observations, they are one observation read twice. The proof is NDSN. Nordson ROSE 0.8% over the two sessions and its mom1 relative print FELL from +3.4 to -7.5 — an 8.6-point collapse produced entirely by the two bars that dropped off the BACK of the window, which were worth +10.0%. Had we read this naively we would have recorded ‘NDSN failed on its third print’ and killed a proposal because the stock went UP. The same artifact runs the other way and is just as fake: ETN’s mom1 went -5.1 to +4.8, but 3.7 of those 9.9 points are dropped bars; AVT went +0.1 to +10.4 with 7.0 points of it dropped bars; ANET went +3.3 to +11.8 with 4.9 points dropped. THIS IS THE SAME DISEASE AS THE 9/11 mom3 FINDING (a fixed 64-bar window faking +3 to +13 points of momentum market-wide), in a new place, and we caught it on a name we had already proposed. THE FIX, EFFECTIVE NOW AND BINDING: (1) A CONFIRMATION OR A FAILURE MAY ONLY BE SCORED ON A SCAN AT LEAST 10 TRADING SESSIONS AFTER THE PRINT IT IS TESTING. Tuesday scans four calendar days after a Friday scan score NOTHING — they cannot satisfy a trigger and they cannot reset one. (2) The genuinely NEW information in a short-gap scan is the RELATIVE RETURN SINCE THE PRIOR SCAN, which shares no bars with anything. That number, and the news, are what a Tuesday scan is for. (3) Going forward a two-scan confirmation requires BOTH a positive relative mom1 AND a positive relative return since the previous scan, so that every confirmation contains at least some non-overlapping evidence. CONSEQUENCE FOR THIS SCAN: NDSN and TECK, proposed on 9/18, are NEITHER confirmed NOR failed today. Their triggers are CARRIED UNCHANGED to the Friday scan. Nothing else may be scored either.
- THE ROTATION OUT OF THIS THEME IS THE MIRROR IMAGE OF THE ROTATION INTO THE TOOL CHAIN AND BOTH HAPPENED IN FORTY-EIGHT HOURS. Refining, tankers, steel and fertiliser were sold (VLO -4.7, MPC -4.6, PSX -4.5, NUE -8.6, CF -7.9, STLD -6.3, CLF -5.4) while semiconductor equipment, memory and compute were bought (LRCX +12.2, AMAT +11.2, AMD +12.9, SNDK +9.4, MU +6.8). Two sessions cannot be scored and this file will not score them. What they are worth is a warning about position concentration: the fund’s single best equity position is MPC at +26.9%, and it sits in the group that was sold. No action is proposed — the exit condition for MPC is a broken template or a properly-scored negative relative print, and neither exists — but the review task should know that the theme leading the book’s P&L is the one the market just rotated out of.
AI Infrastructure Financing — THE RATE SHOCK ARRIVED AND THE SORT HELD: the asset side fell again (OWL -7.0, BX -6.3, APO -4.9, KKR -4.1 in four sessions) while the liability-side insurers mostly held near highs (MET +1.1, RGA +1.4, PFG -1.5). CNBC’s ‘something always breaks when rates rise this fast’ is the tail this theme exists to watch. GAP: last reading 9/21 close, this one 9/25 close = FOUR sessions; the 9/18 proposals’ print (9/17) is SIX sessions old. Neither reaches the 10-session scoring floor. The Friday 9/25 scan never wrote to this file, so this Sunday catch-up is the only reading of the week. — THE SORT IS THE STATISTIC AND IT HAS NOT BROKEN ONCE. Measured on the 21-day window against SPY +1.7: on the liability side, EQH mom1 +10.5 (rel +8.8) at 0.3% off its 52-week high, CRBG +10.2 (rel +8.5) at 0.2% off, UNM +7.6 (rel +5.9) at 3.1% off with a template TRUE, PFG +6.2 (rel +4.5) at 1.7% off with a template TRUE, VOYA +3.6 at 4.1% off TRUE, MET +3.4 at 3.3% off TRUE, RGA +1.1 at 4.2% off TRUE, AIZ -1.3 at 7.7% off TRUE. On the asset side, every alternative manager is impaired: OWL 42.8% off its high with mom1 -12.1 and mom12-1 -35.5, BX 29.9% off with -10.6 and -22.0, KKR 31.8% off with -5.8 and -27.8, ARES 27.9% off with -8.7 and -21.2, BAM 22.3% off with -9.5 and -11.8, APO 15.2% off with -1.6 and -7.9, every template FALSE. MCO is 12.8% off with mom1 -6.3 and a FALSE template, and its two-scan trigger reset to zero on 9/18 and has not restarted. THE UNDERLYING FACT HAS NOT CHANGED AND IS GETTING LARGER: hyperscaler capex of roughly $1 TRILLION in 2027 on the Moody’s, JP Morgan and Goldman numbers ($1.3T on S&P’s six-name basis) is outrunning operating cash flow, consensus has cut expected free cash flow while raising capex, and the gap is being closed with EXTERNAL FINANCING — Oracle alone raised $43B of debt in FY26 with ~$40B more of debt and equity planned against $90-95B of gross capex. Scan: 2026-10-02
| Ticker | Tier | Evidence | Momentum |
|---|---|---|---|
| PFG / UNM / MET / RGA / VOYA / AIZ | the LIABILITY side — insurers and annuity writers that are paid MORE as rates rise (PASSES, the file’s most durable sort) | Six consecutive scans with the same result and no exception has yet appeared. These businesses earn a spread on a float that reprices upward with rates, which is the exact opposite of the alternative managers’ position, and the Fed hike of 16 September with 16 of 18 dots expecting another has not been unwound by the two-session rotation. PFG holds a full trend template 1.7% off its high with mom1 +6.2 (rel +4.5), mom6 +39.1 and vol 22; UNM template TRUE 3.1% off with +7.6 (rel +5.9); MET, RGA, VOYA and AIZ all template TRUE within 8% of their highs. EQH (+10.5, 0.3% off) and CRBG (+10.2, 0.2% off) print the strongest numbers in the group but carry FALSE templates on the stacked-average leg. THE HONEST LIMIT ON ALL OF THIS: these are not thin-coverage discoveries, they are large liquid financials that any rate model would have found, so our edge is the SORT rather than the security. The fund holds none of them | PASSES, TEMPLATES TRUE — PFG -1.7% off high, mom1 +6.2 (rel +4.5), mom6 +39.1, vol 22, $161M/day. UNM -3.1%, +7.6 (rel +5.9). MET -3.3%, +3.4. RGA -4.2%, +1.1. VOYA -4.1%, +3.6. AIZ -7.7%, -1.3. EQH -0.3%, +10.5, FALSE. CRBG -0.2%, +10.2, FALSE |
| BX / OWL / ARES / KKR / BAM / APO | the ASSET side — the managers warehousing AI infrastructure credit (BROKEN, sixth consecutive scan) | The other half of the sort and it keeps getting worse. OWL is 42.8% off its 52-week high with a twelve-month base of -35.5 and a one-month relative print of -13.8; BX 29.9% off with -22.0 and -12.3; KKR 31.8% off with -27.8 and -7.5; ARES 27.9% off with -21.2 and -10.4; BAM 22.3% off with -11.8 and -11.2; APO 15.2% off with -7.9 and -3.3. Every template FALSE. These are the entities that hold the private credit and asset-backed paper funding the build, and their twelve-month bases are NEGATIVE in a year when the assets they finance have been the best-performing equities on the exchange. That is not a valuation opinion, it is the market saying the risk is on the lender. AVOID, and note that this is the single most consistent signal in six weeks of scanning | BROKEN, SIXTH SCAN, NO EXCEPTIONS — OWL -42.8% off high (mom1 -12.1, mom12-1 -35.5); BX -29.9% (-10.6, -22.0); KKR -31.8% (-5.8, -27.8); ARES -27.9% (-8.7, -21.2); BAM -22.3% (-9.5, -11.8); APO -15.2% (-1.6, -7.9). All templates FALSE |
| MCO | ratings — paid per issuance on a debt-funded build (TRIGGER RESET, not restarted) | The two-scan trigger we wrote for Moody’s failed its second leg on 9/18 at rel -4.3 and reset to zero. It has not restarted: mom1 -6.3 (rel -8.0) at 12.8% off its high with a template that is still FALSE. The mechanism is intact and is the cleanest fee-on-flow expression of this theme — a build funded with $43B here and $40B there has to be rated, and the issuer pays per deal regardless of whether the deal performs — but a fee-on-flow business only works while the flow lasts, and the asset-side de-rating above is the market’s opinion on how long that is. Trigger unchanged and now non-overlapping: two positive relative prints scored at least 10 trading sessions apart, with a template reclaim | BROKEN, TRIGGER AT ZERO — -12.8% off high, mom1 -6.3 (rel -8.0), mom3 +4.7, mom6 +7.9, mom12-1 +4.3. Vol 26, $342M/day. Template FALSE |
| THE CREDIT-EVENT TAIL | what breaks if the financing stops — a diagnostic, not a position | Held as a standing diagnostic and unchanged in substance this scan, with one new data point that cuts against the alarm and one that supports it. AGAINST: CoreWeave, the most debt-financed entity in the complex, rose 7.0% (rel +5.3) with the compute rotation, and NBIS rose 6.8%. FOR: Oracle FELL 1.4% in the same two sessions and is now 54.2% off its high with a twelve-month base of -51.5%, carrying $664B of RPO funded by $43B of FY26 debt with ~$40B more planned, roughly half of it owed by a single counterparty. The asset-side managers who would hold the paper are 15-43% off their highs. The diagnostic reading is unchanged: nobody disputes AI demand, and what is being marked down is the capital structure carrying it. No position, no proposal — this exists so that if a credit event does arrive we have already written down where it would show up first | n/a — diagnostic. ORCL -54.2% off high (mom12-1 -51.5), since-scan -1.4. CRWV -40.3% (mom12-1 -26.1), since-scan +7.0. NBIS -18.8%, since-scan +6.8, template TRUE. OWL -42.8%, BX -29.9%, KKR -31.8% |
Crowded consensus (beta, not alpha): This theme is the opposite of crowded and that is precisely why it keeps working as a sort. Nobody is writing about annuity writers as an AI trade, and the names that ARE written about as AI credit — BX, OWL, ARES, KKR — are 22-43% off their highs with negative twelve-month bases. The two-session rotation into the semiconductor tool chain did not disturb either half: the liability side held its trend templates and the asset side stayed broken. Six consecutive scans, one sort, no exceptions — which makes this the most reliable observation in the file and, uncomfortably, the one the fund has acted on least.
Wave map: PAID, AND THE FILE’S MOST DURABLE SORT: the liability side. Insurers and annuity writers earning a wider spread on a float that reprices up with rates. PFG, UNM, MET, RGA, VOYA and AIZ all template TRUE within 8% of their highs; EQH and CRBG at 0.3% and 0.2% off their highs on the strongest prints in the group →→ FEE ON FLOW, TRIGGER AT ZERO: ratings. MCO is paid per issuance on a build funded with $43B here and $40B there, and is 12.8% off its high with a FALSE template and a reset trigger. A fee-on-flow business works only while the flow lasts →→ BROKEN AND GETTING WORSE, SIXTH SCAN: the asset side. The managers warehousing AI infrastructure credit — OWL 42.8% off its high, KKR 31.8%, BX 29.9%, ARES 27.9%, BAM 22.3%, APO 15.2% — with NEGATIVE twelve-month bases in a year when the assets they finance led the market →→ THE TELL: the borrower itself. Oracle 54.2% off its high on a -51.5% twelve-month base carrying $664B of RPO, roughly half owed by one counterparty, funded by $43B of FY26 debt and ~$40B more planned against $90-95B of gross capex — and it FELL while CoreWeave and NBIS rose 7% in the same two sessions →→ THEN: whatever absorbs the loss if the flow stops. Held as a diagnostic with no position, so that if a credit event arrives we have already written down where it shows up first
- 10/2 FRIDAY NOTE – THE SCORING SCAN (10/1 close = 10 sessions after the 9/17 print, so the 9/18 proposals and the ETN/NVT/APH prints are legally scoreable for the first time). Tape since the 9/28 close: SPY -0.2, QQQ +0.7. The 10-year touched 5.344% on 10/1 – highest since 2002 – then reversed to ~5.21%; S&P +0.19% to 7,666. Brent ~$97-104 intraday, Hormuz still disrupted. A stopgap CR averted an October shutdown. MICRON (9/30): Q4 revenue $54.2B (guided $50.0B), EPS $33.42, Q1 FY27 guide $61.5B at ~86% GM; FY27 capex RAISED above prior plans (Q1 ~$11.5B, H1 ~$25B, heavier construction in H2) on ~$150B of long-term agreements. MU faded after hours on the capex; the tool chain did not. CENSUS C30 AUGUST (10/1): data-center construction $85B SAAR, +7.5% m/m, +73% y/y, ~149% annualized since March; power construction only +8.5% y/y; total construction unexpectedly up.
- FINANCING 10/2: the asset side kept falling (BX -10.9 rel since 9/17, OWL -10.6, MCO -3.6) and – new – the liability side cracked too: UNM -8.3 rel since 9/17, MET -3.5, PFG -3.6, RGA -3.1 since 9/28. A 10y at 5.34% intraday is now hurting the insurers’ unrealized-loss books faster than it helps reinvestment yields. Templates still TRUE for PFG/MET/RGA; watch, no adds.
- 9/29 TUESDAY NOTE (one session, nothing scoreable): the 10-year made a fresh 19-year high at 5.24-5.27% intraday on 9/28, 30-year 5.55%; SPY -0.7, QQQ -1.1, VIX 16.1; WTI back to ~$93 (from >$103). Fed governor Cook: AI and oil keep inflation up, further hikes data-dependent. The rate-shock sort continued in one day: borrowers and long-duration fell (ORCL -3.3, BE -8.9, VRT -3.6, CRDO -8.7, AMD -3.6) while cash-rich names rose (NVDA +1.7 on a new $150B buyback). Scoring calendar unchanged: NDSN/TECK/ETN/NVT/APH at the 10/1 close (Fri 10/2 scan); ENTG at the 10/5 close (Tue 10/6 scan). MICRON reports WED 9/30 after the close.
- Seventh scan of the same sort and the first one run under a genuine rate shock (10y 5.13%, 30y 5.53%, bond-vol gauge +19% on the week). The warehouse (private credit) is -20 to -43% off highs with mom1 -9 to -21; the annuity writers are within 2-5% of highs. ORCL (-55.8% off high) is the credit-event tail’s lead indicator and it worsened by 7.7%. Hold the sort; add nothing capital-hungry while long yields are above 5%.
- THE OVERLAPPING-WINDOW BUG: A TUESDAY SCAN CANNOT CONFIRM OR REFUTE A FRIDAY SCAN, AND THIS FILE ALMOST ACTED ON THE ARTIFACT. Last scan measured to the 2026-09-17 close; this one measures to 2026-09-21. That is TWO TRADING SESSIONS. A 21-bar mom1 window therefore shares 19 OF ITS 21 BARS with the print we already recorded, so the two numbers are not two observations, they are one observation read twice. The proof is NDSN. Nordson ROSE 0.8% over the two sessions and its mom1 relative print FELL from +3.4 to -7.5 — an 8.6-point collapse produced entirely by the two bars that dropped off the BACK of the window, which were worth +10.0%. Had we read this naively we would have recorded ‘NDSN failed on its third print’ and killed a proposal because the stock went UP. The same artifact runs the other way and is just as fake: ETN’s mom1 went -5.1 to +4.8, but 3.7 of those 9.9 points are dropped bars; AVT went +0.1 to +10.4 with 7.0 points of it dropped bars; ANET went +3.3 to +11.8 with 4.9 points dropped. THIS IS THE SAME DISEASE AS THE 9/11 mom3 FINDING (a fixed 64-bar window faking +3 to +13 points of momentum market-wide), in a new place, and we caught it on a name we had already proposed. THE FIX, EFFECTIVE NOW AND BINDING: (1) A CONFIRMATION OR A FAILURE MAY ONLY BE SCORED ON A SCAN AT LEAST 10 TRADING SESSIONS AFTER THE PRINT IT IS TESTING. Tuesday scans four calendar days after a Friday scan score NOTHING — they cannot satisfy a trigger and they cannot reset one. (2) The genuinely NEW information in a short-gap scan is the RELATIVE RETURN SINCE THE PRIOR SCAN, which shares no bars with anything. That number, and the news, are what a Tuesday scan is for. (3) Going forward a two-scan confirmation requires BOTH a positive relative mom1 AND a positive relative return since the previous scan, so that every confirmation contains at least some non-overlapping evidence. CONSEQUENCE FOR THIS SCAN: NDSN and TECK, proposed on 9/18, are NEITHER confirmed NOR failed today. Their triggers are CARRIED UNCHANGED to the Friday scan. Nothing else may be scored either.
- THE SORT HAS NOW SURVIVED SIX SCANS, A RATE HIKE AND A VIOLENT SECTOR ROTATION, AND THE FUND STILL OWNS NONE OF IT. Liability-side compounders that are paid more as rates rise hold trend templates with positive one-month relative prints (PFG +4.5 rel at 1.7% off its high with vol 22, UNM +5.9 at 3.1% off, MET, RGA, VOYA, AIZ all TRUE); asset-side managers who warehouse the credit funding the AI build are 15-43% off their highs with NEGATIVE twelve-month bases and FALSE templates. That is a clean, repeatable, mechanism-backed sort that has not produced a single exception since we started recording it. The reason it has never become a proposal is that it fails the THIN-COVERAGE leg of our signature — these are large liquid financials where we have no informational edge — and we have been right to say so. But the file should at least be honest that we have applied the modest-momentum and thin-coverage legs strictly enough to exclude the only thing that has worked repeatedly, while spending eight weeks hunting down a supply chain that has not worked at all. That tension does not resolve itself today and it is not resolved by a two-session scan; it is written down so the next scoreable scan has to confront it.
📈 Momentum Leaders
Liquid names in uptrends within 20% of highs, ranked by blended 3/6/12-month momentum (screen of 200+ tickers)
| Ticker | Price | 3-mo | 6-mo | 12-1 | Off high |
|---|---|---|---|---|---|
| MRNA | $188.94 | +136.9% | +277.7% | +497.3% | -7.1% |
| MU | $1,097.39 | +12.5% | +198.4% | +458.8% | -9.6% |
| LITE | $1,045.78 | +43.6% | +36.8% | +434.0% | -0.7% |
| DELL | $541.74 | +37.6% | +221.3% | +203.6% | -7.9% |
| AMD | $615.73 | +18.9% | +192.9% | +184.1% | -2.4% |
| STX | $945.57 | +15.4% | +123.8% | +247.7% | -13.4% |
| VICR | $308.59 | +9.1% | +95.1% | +254.0% | -18.7% |
| INTC | $120.00 | -0.3% | +149.8% | +165.2% | -14.9% |
| HPE | $64.58 | +57.0% | +170.8% | +110.3% | +0.0% |
| MRVL | $268.08 | +9.3% | +151.4% | +150.8% | -15.3% |
| CRWD | $266.09 | +37.2% | +170.6% | +75.4% | +0.0% |
| BE | $277.58 | +2.5% | +109.6% | +152.6% | -19.7% |
🎯 Dip Radar
Quality names in uptrends hitting our proprietary oversold trigger (backtested Sharpe 1.5+, PF 2-4)
| Ticker | Price | Oversold score | Off 20d high | Status |
|---|---|---|---|---|
| UNH | $403.97 | 20.0 | -7.4% | watch |
| ABBV | $243.87 | 24.0 | -7.4% | watch |
🔍 Filing-Checked Verdicts
Does the fundamental picture (10-K/10-Q/8-K, Form 4s, 13Fs) support the price momentum?
| Ticker | Theme | Verdict | Why |
|---|---|---|---|
| DELL | AI servers | SUPPORTED | Q +88% y/y, AI servers +757%; FY guide $165-169B; fwd ~24x; buybacks |
| AMD | AI chips | SUPPORTED | +50% y/y accelerating, DC +107%; net cash $9.9B; China policy = main risk |
| SNOW | Data/AI software | SUPPORTED | Re-accelerating 30→34%, NRR 126%, guide raised; ~20x EV/S |
| GH | Liquid biopsy | SUPPORTED | +44% accelerating, Shield +253%, guide raised; 16x sales |
| RVMD | RAS oncology | SUPPORTED* | Ph3 OS 13.2 vs 6.7mo (HR 0.40), NDA accepted, $3.9B cash; binary, 7.2% SI rising |
| MU | Memory/HBM | PEAK-CYCLE | +346% y/y, GM 85%, $50B qtr guided — fwd P/E 6 = market calls the top; Burry short |
| STX | HDD/HAMR | PEAK-CYCLE | +48%, record margins — fwd 24x vs 8-15x norm; shorts rising |
| ATI | Aero metals | MIXED | EBITDA +37%, big raise, $4.4B backlog — 40x EPS for a cyclical |
| PANW | Cybersecurity | MIXED | Organic ~14% vs headline 31% (CyberArk); ~90x fwd; ~10% dilution |
| ALAB | AI connectivity | MIXED | +104% accelerating, 39% op margin — one customer >70% of revenue |
| NTAP | Storage | MIXED | 5-8% growth carrying an AI multiple; watchlist |
🌎 Macro Read
- Fed: 3.50-3.75%, Warsh chair; 3 FOMC dissents FOR a hike, but July payrolls -23k → Sept ~60% hold
- Inflation: CPI 3.5% / core PCE 3.3% — oil-shock driven (Hormuz closed since early July)
- Growth: Split-screen: ISM mfg 55.6 (4-yr high) + AI capex ~$700B (+77%) vs housing recession, card delinquencies 13.1%
- Positioning: Insiders selling 11:1; semis most-crowded trade ever (82%); BofA cash 3.6% = sell signal; MMF record $7.9T
- Regime: Late-cycle + supply shock (2006-07 x 1990 rhyme). July AI-hardware unwind: chips -$1T, memory -20-47%
⚖️ Strategies That Survived Backtesting
3 years of daily data, honest holdout validation, real CBOE benchmark anchoring — full write-ups in the blog
| Strategy | Rule | Evidence |
|---|---|---|
| Strategy 1 | Proprietary — rules private | 28-46% CAGR in 3y test; decades of published evidence behind the mechanism |
| Strategy 2 | Proprietary — rules private | Sharpe 1.49 (2.53 in holdout), PF 3.8, max DD -3.4% |
| Strategy 3 | Proprietary — rules private | Anchored to a real benchmark index: 16.4% CAGR over the window |
| Strategy 4 | Proprietary — rules private | Income variant: 1-3%/mo while waiting for better entries |
| Strategy 5 | Proprietary — rules private | Raw-return leader of the lab; conservatively flagged |
| Strategy 6 | Proprietary — rules private | 38-40% CAGR, Sharpe 1.3-1.5 in 3y test |
| Strategy 7 | Proprietary — rules private | Best income variant tested; expect low-teens real after friction |
💬 Comments
Public thread — ideas, critiques, requests. Share the page and bring people in.
No comments yet — be the first. What would you trade differently?
Nothing here is financial advice. Backtested performance does not guarantee future results.
Synthetic option backtests overstate returns vs real traded indices (we measured it: ~6-8 CAGR pts) — anchor expectations to real data.
See the risk disclaimer.