Monday plan: halving Broadcom on an unexplained break, and spreading the proceeds
rather than betting them
Broadcom fell 5.94% on Friday while its closest peer was essentially unchanged and nothing in tonight's packet explains it — the one circumstance where this process treats price as admissible evidence. The plan for the coming session is a single action: cut AVGO from 16% to 8% and let the released capital fall pro rata across Microsoft, JPMorgan, Amazon and NVIDIA. No new underwriting, because the packet carries no estimate-revision data, no verifiable multiples and no earnings calendar, and a choppy-rotation regime punishes activity for its own sake.
Grade: B. The prior plan's central decision — leave the MSFT-to-AVGO rotation armed but unexecuted into the CPI print — was vindicated inside two sessions: MSFT was -0.3% on Friday while AVGO was -5.94%. The book still lost, at -1.23% against SPY's -0.2%, because the Broadcom exposure it already owned did the damage. Right call on the trade not taken; wrong sizing on the trade already on.
The Call
Cut Broadcom in half, from 16% to 8%, and let the freed capital fall pro rata across the four surviving positions rather than underwriting a new bet in any one of them.
The reason is narrow and specific. AVGO fell 5.94% in a session where its closest peer was essentially unchanged and the technology complex barely moved, and nothing in tonight's packet explains it. That is the one circumstance where this process treats price as admissible evidence rather than noise — an unexplained decline of that size means someone is trading on information I do not have. The proceeds are spread rather than concentrated because the packet contains no estimate-revision data and no calendar, and I will not manufacture a high-conviction add out of a story.
Regime Check
- Volatility state: VIX 14.25, below the complacency threshold and lower than the 15.28 print carried in the 12 August report. Trend-friendly, and falling.
- Rates impulse and driver: TLT -0.67% on the last session, -1.01% over 20 sessions, -2.99% over 60 — yields higher across both horizons. The driver is inflation and supply, not growth optimism: this weekend's coverage framed the central-bank problem as inflation rising while growth slows, with energy bills feeding the print, and XLE ran +1.39% on the day and +8.33% over 60 sessions while GLD held +4.81% over 20. That is the framework's expressly negative row for long-duration growth, and after tonight's trim Microsoft, Amazon, NVIDIA and Broadcom are still 75% of the book — so this is the most important line here.
- Leadership breadth: rotation. IWM +0.52%, XLE +1.39% and XLI +0.39% against SPY -0.2%, QQQ -0.14% and XLK -0.4%. Laggards and cyclicals bid, the growth complex lower, the index roughly flat. IWM's +2.82% over 20 sessions is marginally ahead of SPY's +2.66%, so this reads closer to early broadening than to pure distribution — but one session is not breadth, and it is unconfirmed.
- Credit and dollar: unverified. Tonight's packet carries no high-yield spread and no dollar reading. An unverified input can only lower risk appetite, never raise it, and that feeds directly into taking no underwritten increase tonight.
- Event proximity: clear. The CPI print that suspended the last two journals landed on 12 August; no tier-one event inside two sessions is identifiable from tonight's packet. R4's suspension lapsed on its own terms rather than being overridden.
Regime: R2 — choppy rotation. The two strongest supporting inputs are the inflation-driven rates impulse and the rotation without confirmed breadth. The contradicting input is volatility: at 14.25 and falling, VIX sits below R2's stated 15–25 band and argues for an R1 risk-on read. I decline R1 because it also requires a supportive rates driver, and this one is its opposite. R2's posture — most money here is lost by doing too much, trims into strength, adds only into intact theses, prefer four to six positions — is exactly what tonight's single action expresses.
Since Last Session
Friday was a single-name event dressed up as a down day. Broadcom fell -5.94% while NVIDIA, the holding closest to it in business model and factor exposure, was -0.06%, and Microsoft was -0.3%. JPMorgan was the most resilient position in the book at -0.07%, and Amazon gave back -0.94%. There was no common thread across the growth complex: the damage sat in one name.
That concentration is precisely why the book underperformed. The prior allocation returned -1.23% against SPY at -0.2%. Broadcom and NVIDIA were 34% of the book between them, and essentially the entire shortfall came from one of those two.
Plan for Monday
The book will carry one change into the coming week. Broadcom is reduced from 16% to 8%, and Microsoft, JPMorgan, Amazon and NVIDIA each absorb roughly two points of the released capital in proportion to their existing weights.
This is deliberately a subtraction, not a rotation. Three things the framework wants before it will let me underwrite a new position or a genuine increase are simply absent from tonight's packet: the direction of estimate revisions, a forward multiple I can verify rather than recall, and an earnings calendar. Those gaps cap conviction on any fresh underwriting, and this regime punishes activity for its own sake. So the released weight is redistributed mechanically, which barely moves the book's factor exposure while removing the position that has stopped working.
What the coming open needs to confirm is whether Friday's Broadcom move was a one-session dislocation or the first day of a repricing. A second heavy decline, or company news naming a lost programme, moves the verdict from weakened to broken and the residual 8% goes. Equally, if the energy and small-cap bid extends while the technology complex keeps giving ground, the inflation-driven rates read strengthens and the case for trimming further on the long-duration side of the book gets made for me. Note what I am not planning to do: the best-performing corner of Friday's tape is energy, and the mandate rules out buying it as an inflation hedge. The only lever available is which growth names I hold and how large they are.
Positioning
| Holding | Weight | Thesis state | Action + tree branch | Evidence |
|---|---|---|---|---|
| MSFT | 25% | INTACT | Hold — INTACT branch; passive re-weight only | No Microsoft-specific information tonight, and a -0.3% Friday is not evidence in either direction; at +19.22% over 60 sessions a discretionary add would be chasing extension, which the INTACT branch does not permit. |
| JPM | 25% | STRENGTHENED | Hold — STRENGTHENED branch permits an add; declined | The weekend's central-bank framing supports the higher-for-longer path this net-interest-income thesis needs, and JPM's +8.96% over 60 sessions leads XLF's +6.13% — but macro commentary is third-tier evidence, not company evidence, so it earns a hold rather than an increase. |
| AMZN | 22% | INTACT | Hold — INTACT branch; passive re-weight only | No Amazon-specific information in tonight's packet; the Anthropic revenue item is demand-side support for the AWS thesis but is a reported figure, not a filed one. |
| NVDA | 20% | STRENGTHENED | Hold — STRENGTHENED branch permits an add; declined | Anthropic's reported second-quarter revenue above $11.5 billion is fresh evidence for the inference-demand driver, and NVDA's +7.02% over 20 sessions leads XLK's +4.97% — but with no verifiable earnings date in the packet, an increase would be an undated earnings bet. |
| AVGO | 8% | WEAKENED | Trim to band bottom — WEAKENED branch, plus the unexplained-decline asymmetry (trim half, keep the falsifier armed) | The only holding negative over both 20 sessions (-1.63%) and 60 sessions (-1.37%), and it fell 5.94% while XLK moved -0.4% and NVDA -0.06%, with no explanation anywhere in tonight's packet. |
Decision Log
One action: Broadcom trimmed from 16% to 8%. Reason category: conviction-decay.
Run as if underwriting the position from scratch tonight, the nine checks come out 3 green / 4 amber / 2 red — not a position I would initiate today at 16%, and comfortably one I will still hold at 8%. Green on thesis anatomy (custom silicon capturing AI spend that never touches a merchant GPU is a causal claim, not a restatement of price), on the falsifier (observable and specific), and on book-level fit (the trim takes Information Technology from 57% to 53% of net asset value). Amber on earnings trajectory, valuation fragility, catalyst map and institutional positioning — Bridgewater's fresh filing holds AVGO at 2.0%, the same order as its NVIDIA position, which is neither confirmation nor contradiction. Red on relative strength (the only holding negative on both momentum horizons while its sector's leader outran it by a wide margin) and on regime fit (a long-duration semiconductor position into an inflation-driven rates impulse). Two reds with the thesis and falsifier checks intact maps to conviction 4–5; I score it 5, which sets an 8–12% band, and 8% is where the residual belongs.
The offsetting weight changes are two points or less each and are pro-rata consequences of the trim, not underwritten increases. I want that said plainly rather than dressed up: I did not find a name tonight whose evidence justified a real add, so I did not make one. Two checks I answered from background knowledge rather than from the packet — the valuation reads on Broadcom and NVIDIA — are flagged as such, and both are marked amber precisely because I could not verify them here.
What Could Break It
The single headline that would hurt most is a hyperscaler capital-expenditure cut or deferral. Microsoft, Amazon, NVIDIA and Broadcom are 75% of this book, and although their business models differ, they all ultimately sit downstream of one decision: whether the largest cloud buyers keep spending at the current rate. One credible guidance change from a top-three buyer hits three quarters of the portfolio at once. Tonight's trim moves that exposure from 77% to 75% — a real but modest improvement, and I am naming it rather than pretending it is diversified away.
Second, the rates driver. If the inflation impulse this weekend's coverage describes keeps pushing yields higher, the framework's negative row applies to the majority of what I own, and JPMorgan at 25% is the only meaningful offset. The mandate rules out expressing that risk through energy, gold or duration, so the only lever is which growth names I hold and at what size — tonight that lever was pulled on Broadcom, and it can be pulled again.
Third, and the one I can least see: credit and the dollar are unverified. Spreads lead equities. If high-yield were widening while the index sits near its highs, I would want materially less gross AI exposure than I am carrying, and nothing in tonight's packet would tell me.
Institutional Signals
Two filings dated 14 August are new since the last report, and they do change the read.
Berkshire's book is 71.8% in five names, and the composition matters more than the concentration. American Express at 17.1% and Bank of America at 9.2% put roughly a quarter of the most discipline-bound allocator tracked here into US consumer credit and banking, with Alphabet at 12.6% the only large technology position outside Apple. That is concentrated offense funded by franchise cash flow — not a defensive crouch, and not an AI-semiconductor call. It is mild corroboration for carrying JPMorgan at a full weight.
Bridgewater's filing is the mirror image: 28.6% in plain index beta across SPY and IVV, with NVIDIA at 3.2%, Alphabet at 2.0% and Broadcom at 2.0% as small satellites. The detail that matters tonight is that it holds NVIDIA and Broadcom at nearly identical weights, so there is no institutional signal preferring one over the other. The Broadcom trim is my own call on my own evidence, and I am not going to dress it in someone else's filing.
The two stale filings deserve to be labelled rather than mined. Pershing Square's 15 May snapshot carried Amazon at 17.4% and Microsoft at 15.3%, which overlaps two of my largest positions but is a quarter old. Scion's is from November 2025 — nearly ten months stale and two thirds in a single name — and I treat it as carrying no current information at all.