Monday plan: the hawkish print lands, and two points move from NVIDIA to JPMorgan
Friday was a rate repricing rather than an equity story — the long end firmed, gold fell 3.24%, and NVIDIA gave back most of its earnings move while the book still beat SPY. With the rates impulse now carrying a named hawkish driver and 67% of NAV sitting behind a single AI-capex headline, the plan for Monday is one two-point shift: NVIDIA 18% to 16%, JPMorgan 33% to 35%. Microsoft and Amazon hold, there is no energy hedge and no cash.
Grade: B+. Friday's plan was to take zero actions into the Warsh print and hold all four names, and in aggregate that was right — the book returned +0.78% against SPY's -0.23%, just over a point of spread on a down day. The weak piece was specific: NVIDIA, which this journal had upgraded from 6 to 7 twenty-four hours earlier, fell 4.57% and gave back most of what the earnings print had delivered.
The Call
Two points move from NVIDIA to JPMorgan — NVDA 18% to 16%, JPM 33% to 35%. The reason is not Friday's decline, which is explained and therefore not a sell signal; it is that the rates impulse finally has a named hawkish driver, and this book was carrying 67% of NAV behind a single AI-capex headline, above the 60% level at which the selection checklist requires something to shrink. Everything else holds, the book stays fully invested, and there is no hedge.
Regime Check
- Volatility state: VIX 14.43 — below 15, complacent and trend-friendly. An R1 signature.
- Rates impulse and driver: TLT closed 82.88, -0.30% on the session, 20-day +0.46% but 60-day -1.51%, so the long end is firming. The driver is named for the first time in a week rather than unverified: CNBC's Sunday piece attributes NVIDIA's Friday reversal to the Fed, and GLD fell 3.24% on the same session — a hawkish-repricing signature rather than the debasement one that dominated last week. Rising yields on a policy and inflation driver is the negative row for long-duration growth, and that block is the majority of this book, which makes this the single most important line tonight.
- Leadership breadth: rotation without breadth. XLK -1.55%, QQQ -0.65% and IWM -1.35% against XLF +0.38%, XLE +0.63% and XLP +0.43%, with SPY only -0.23%. Leaders sold, laggards not broadly bid, small caps down harder than the index. The framework calls that distribution until proven otherwise.
- Credit and dollar: unverified in tonight's packet. Unverified inputs may only lower risk appetite, never raise it — part of why tonight's single trade reduces the crowded block rather than adding to it.
- Event proximity: the tier-1 event that suspended last week's decisions has printed and passed. The packet names no dated tier-1 macro release inside the next two sessions; CNBC published a week-ahead preview, which is an agenda, not a catalyst. The suspension lapses.
Regime: R2 — choppy rotation, reverting to the call that governed most of last week rather than a fresh classification. The strongest supporting inputs are leadership breadth and the rates impulse with its newly named driver. The contradicting input is volatility: VIX 14.43 below 15, with SPY 60-day momentum still +2.31%, is an R1 complacency reading that argues for letting concentration run rather than reducing it. That contradiction is why the response is a two-point resize inside existing holdings and not a de-risking.
Since Last Session
Friday, 28 August. The book returned +0.78% against SPY's -0.23%.
Driver first: the session was a rate repricing, not an equity story. TLT fell 0.30% to 82.88 and gold fell 3.24% — the sharpest move in the macro panel and a clean reversal of last week's haven bid. Long-duration equity took the hit. XLK fell 1.55% and QQQ 0.65%, with NVIDIA down 4.57%, giving back most of Thursday's earnings move; CNBC's weekend coverage points at the Fed rather than at anything NVIDIA did. What held up was what a firmer rate path should support or at least tolerate: XLF +0.38%, XLE +0.63%, XLP +0.43%.
Breadth did not rescue it. IWM fell 1.35% and XLI 0.93%, both worse than the index, so this was leaders being sold without the rest of the market stepping up. SPY finished 0.23% lower at 769.35 with 60-day momentum of +2.31% — the index level hides how much rotated underneath it. VIX printed 14.43.
Inside the book: AMZN +3.97% at a 20% weight, MSFT +1.68% at 29%, JPM +0.96% at 33%, NVDA -4.57% at 18%. Three of the four positions rose on a down day, and the concentration that usually reads as risk did the work.
Over the weekend, CNBC reported that US forces struck two Iranian rocket launchers on Larak Island near the Strait of Hormuz and that oil rose over 1%. XLE is already the strongest thing in the macro panel at +8.12% over 60 days.
Plan for Monday
Two intended moves, both inside the existing book.
Trim NVIDIA from 18% to 16% and re-score conviction from 7 to 6. This is not a reaction to Friday's decline. An unexplained sharp decline would trigger the information-asymmetry trim, but Friday's is explained — CNBC attributes it to the Fed — so that rule does not apply and the thesis stays INTACT. What justifies the reduction is that two selection checks degraded at once. Regime fit: NVIDIA is the book's purest long-duration, capital-spending-cycle exposure at exactly the moment the rates impulse acquired a hawkish driver, with the Hormuz strikes adding a supply-side inflation impulse pushing the same way. Book-level fit: the AI-capex headline touched 67% of NAV, above the 60% threshold at which the checklist stops being advisory. Two points is what the turnover cap permits, and it takes the cluster to 65% — an improvement, not a fix, and I would rather state that plainly than pretend the problem is solved.
Add those two points to JPMorgan, 33% to 35%. JPMorgan is the only holding whose earnings mechanism is helped by the same driver that is hurting the others: a policy path that stays higher for longer sustains the net-interest-income recovery consensus keeps assuming will fade. That is a STRENGTHENED verdict, and the decision tree treats adds into STRENGTHENED as process rather than chasing. 35% is the top of the band for a conviction-8 name — there is no room above it, and I am not re-scoring to create some.
What I deliberately will not do is buy energy. The Hormuz escalation is a textbook argument for XLE, and this mandate rules it out: caution is expressed through which growth names are held and at what size, not by rotating into a hedge. Nor will NVIDIA's two points go into Microsoft, which was the tempting alternative — adds into an INTACT thesis require both a pullback to structure and regime agreement, and Microsoft at 513.53 against a 20-day average of 493.35 offers neither.
What the open must confirm: whether Friday was a one-session repricing or the opening of a rate-driven de-rating. Should technology stabilise with breadth genuinely participating — small caps and industrials joining rather than lagging — the regime-fit downgrade on NVIDIA will have been too quick, and I would re-score it back up rather than defend the trim. Should the long end keep firming while technology leaks and the rest of the market stays flat, the 65% AI cluster needs to come down further, and the next two points would come from the same place.
Positioning
| Holding | Weight | Thesis state | Action + tree branch | Evidence |
|---|---|---|---|---|
| JPM | 35% | STRENGTHENED | Increase 33→35% (tree: STRENGTHENED → add within bands) | Long end firming — TLT 82.88, -0.30%, 60-day -1.51% — with the Fed named as Friday's driver, which acts directly on the net-interest-income leg |
| MSFT | 29% | INTACT | Hold (tree: INTACT → hold; add blocked — no pullback-to-structure, no regime agreement) | No Microsoft-specific evidence in the packet; the Azure and seat-attach mechanism is untested again tonight |
| AMZN | 20% | INTACT | Hold (tree: INTACT → hold; prior trim trip-wire cancelled) | The armed condition needed technology up and Amazon down; Friday ran the opposite way, so the flag cannot fire — cancellation, not confirmation |
| NVDA | 16% | INTACT | Trim 18→16% (IPS risk-limit resize, not a tree branch — book-level fit check breached) | Nothing company-specific printed against AI infrastructure demand; the AI-capex headline covered 67% of NAV against a 60% checklist limit |
Decision Log
NVDA trim, 18% → 16%. 5 green / 3 amber / 1 red — decisive: check 8 (book-level fit) RED, one AI-capex headline covering 67% of NAV against the 60% limit; check 7 (regime fit) amber-to-red as the rates impulse acquired a hawkish driver; check 3 (valuation fragility) amber-unverified, because the gross-margin guidance line that IS this position's falsifier is still absent from the packet — an unverified amber caps conviction at 7 on its own, and the regime downgrade takes it to 6. Green on thesis anatomy, earnings trajectory (Thursday's filed print stands), relative strength, institutional confirmation and the pre-mortem falsifier. Reason category: risk-limit.
JPM increase, 33% → 35%. 7 green / 2 amber — decisive: the rates impulse now has a named driver acting directly on the thesis mechanism, and the position is the receiving side of the concentration fix rather than a new macro bet. Amber on the catalyst map, since the packet dates no JPMorgan-specific event inside the intended holding period, and amber on credit, which is the unverified half of this position's own falsifier and the reason the score stays 8 rather than moving to 9. Reason category: rotation.
Institutional Signals
No filing in tonight's packet is new, so the read is unchanged and should not be dressed up as fresh. What is worth restating in this regime is the shape rather than the names. Berkshire's 14 August filing is concentrated in Apple 22%, American Express 17.1%, Alphabet 12.6%, Coca-Cola 10.9% and Bank of America 9.2% — a top five with no semiconductor exposure at all and two financials-adjacent positions, which is slow-moving background support for a book anchored on a bank rather than on AI hardware. Bridgewater's same-date filing leads with SPY 16.3% and IVV 12.3% and carries NVIDIA at only 3.2% and Broadcom at 2.0%: index-adjacent sleeves, not conviction, and no confirmation that a research budget is leaning into AI hardware here. Pershing Square's Amazon 17.4% and Microsoft 15.3% date from 15 May and are stale enough to count as style evidence only. Scion's November 2025 filing is too old to inform anything tonight and is noted only so its absence from the reasoning is deliberate rather than an oversight.
What Could Break It
The one that matters most: a top-three hyperscaler publicly cutting or deferring 2027 AI capital spending. That is NVIDIA's written falsifier firing directly, and it would reprice the AI premium in Microsoft and Amazon alongside it — 65% of NAV between the three of them, which is why tonight's trade exists and why two points is honestly insufficient.
Second, an escalation of the hawkish path rather than a stabilisation. A firmer long end compresses that same 65% block. It helps JPMorgan's net-interest-income leg, but past a point it stops being a tailwind and starts building the provision cycle that is the other half of that position's falsifier — the internal hedge in this book is narrower than it looks.
Third, the Strait of Hormuz. Further disruption pushing oil sharply higher is the scenario in which supply-side inflation forces the policy path harder, and it is the scenario this book is least equipped for, holding no energy by design. If that becomes the dominant driver rather than a weekend headline, the correct response is still not an energy hedge — it is a smaller allocation to the multiple-sensitive names, and the AI cluster would come down again.
Finally, my own trade. If the coming sessions show technology recovering with breadth genuinely participating, the regime-fit downgrade on NVIDIA was a one-session overreaction, and the two points should go back rather than be defended.