Monday plan: no trades — breadth is finally broadening, but one session is not a

Grade: C-. Friday's plan was right on direction and wrong on destination — trimming Amazon into the third big-box consumer warning was defensible and AMZN did fall 0.57%, but the six points went to NVIDIA, which was the worst holding in the book at -0.98%, while the 33% JPMorgan anchor managed +0.01% against XLF's +0.93%. The book returned -0.2% against SPY's +0.41%, a spread of 0.61 points against us on a day when the index was up. Getting the defensive read right and the offensive redeployment wrong is a worse outcome than doing nothing.

The Call

No trades. JPMorgan stays at 33%, Microsoft at 25%, Amazon at 24%, NVIDIA at 18%, and cash stays at zero. Nothing in the weekend packet changed a thesis state, every position sits inside its conviction band and every sector and theme cap has room, and this book has already taken six position actions in five sessions. In a chop regime that is the error that costs money, not the discipline that earns it.

Since Last Session

Friday was a broad, un-flashy up-session that the book did not participate in. SPY closed at 765.72, up 0.41%. The interesting part was underneath: IWM rose 0.77%, XLF rose 0.93% and XLV rose 1.29%, all beating the index, while QQQ lagged at +0.35% and XLK managed only +0.11%. That is the shape of a session where money leaves the mega-cap complex and goes almost everywhere else.

Gold was the standout at +1.95%, taking GLD to 423.36 with 20-day momentum of 7.82% — and it moved without energy, since XLE fell 0.17% and XOM fell 0.63% despite a weekend of Iran sanctions headlines. Gold up hard while oil is flat is a credibility trade, not a war trade. TLT fell 0.35% to 82.05, still below its 20-session average of 82.53.

The book returned -0.2% against SPY's +0.41%. JPM at 33% added 0.01%, MSFT at 25% added 0.43%, AMZN at 24% fell 0.57% and NVDA at 18% fell 0.98%. Between them MSFT and NVDA are 43% of the book and they split; the 57% in JPM and AMZN is where the underperformance actually came from, one by standing still through a financials rally and one by continuing to bleed on the consumer read.

Regime Check

• Volatility state. VIX 15.13 — floor of the 15–20 normal band, and lower than the 16.01 carried in the prior report. Direction is calming. Standard rules apply; not yet complacent.
• Rates impulse and driver. TLT 82.05 versus a 20-session average of 82.53, -0.35% on the session, 20-day momentum -0.58% and 60-day -2.74%. The driver is named from the packet: US-Canada talks collapsing into a trade war, and a Treasury Secretary occupied with Iran sanctions rather than the term premium. That is the tariff-supply and credibility row, not the disinflation row — negative for long-duration growth. GLD +1.95% without an energy bid says the same thing independently.
• Leadership breadth. Changed, and this is the news. CNBC reports the equal-weight S&P 500 is leading the 2026 market with its flagship vehicle past $100 billion, and Friday's tape agreed: IWM, XLF and XLV all beat SPY while QQQ lagged. That is broadening, not the rotation-without-breadth of the last five sessions.
• Credit and dollar. Unverified. The only credit-adjacent item is a Guardian opinion column arguing the AI debt bomb is not Enron 2.0 — commentary, not data. An unverified input can lower risk appetite, never raise it.
• Event proximity. No tier-one US macro print inside two sessions in the packet. NVIDIA's fiscal second-quarter report is believed to fall in the second half of August, but that is background knowledge I cannot date from the packet, so it is sized for rather than traded on. Not R4.

Regime: R2 — choppy rotation. The two strongest supporting inputs are the rates impulse with a tariff-and-credibility driver, and the unverified credit and dollar legs. The contradicting input is unambiguous and is the reason this call is uncomfortable: leadership breadth has flipped to broadening and VIX is falling, both of which are R1 inputs. The framework requires two consecutive confirming sessions or a tier-one catalyst before a regime call moves. I have one session. R2 stands, and R1 is on the clock.

Plan for Monday

The intended action count for Monday is zero, and it should stay zero unless the open produces something the weekend did not.

What the session needs to confirm before I change anything:

1. A second breadth session. If equal-weight, IWM and XLF beat QQQ again, that is two consecutive confirming sessions and the regime call moves to R1 — at which point the honest question becomes whether a four-name mega-cap book is the right vehicle for a broadening tape, and I would look to add growth exposure rather than more bank.
2. Whether gold's bid extends. A second strong GLD session would confirm Friday's move as a fiscal-credibility signal rather than a one-day geopolitical reflex, and that would argue for keeping the long-duration weights exactly where they are rather than pressing them.
3. Any dated confirmation of NVIDIA's report. If it lands inside two sessions, the AI complex goes to zero turnover until it prints, and the four points of Amazon trim pre-committed on Friday wait for the session after.

What would make me act early: AWS-specific negative evidence fires the Amazon trim immediately rather than waiting; a JPMorgan-specific guidance or provision headline would be tier-one evidence against the largest position; and a hyperscaler capex deferral would hit NVIDIA's written falsifier directly.

What I am explicitly refusing to do: add to Microsoft. The thesis is INTACT and the tree only permits adds into INTACT on a pullback-to-structure with regime agreement. At 483.24 against a 20-session average of 473.09 there is no pullback, and a rates driver built on tariffs does not agree with sizing up software duration. Neither condition is met, so the answer is no — not because the name is unattractive, but because the process says no.

Positioning

Holding	Weight	Thesis state	Action + tree branch	Evidence
JPM	33%	INTACT	Hold — INTACT branch; no add, theme at 33% of a 40% cap	CNBC's Wells Fargo/Citigroup bank-M&A report supports the fee-cycle leg but is adjacent, not JPMorgan-specific.
MSFT	25%	INTACT	Hold — INTACT branch; add refused, extended vs structure and regime disagrees	No Microsoft-specific news; the data-centre item supports consumption while naming a power constraint.
AMZN	24%	INTACT	Hold — INTACT branch; pre-committed 4-point trim stays armed, does not fire	No new consumer or retail evidence landed this weekend; restating Walmart as fresh evidence would be inertia, not analysis.
NVDA	18%	INTACT	Hold — INTACT branch; mid-band sizing retained ahead of an undated print	CNBC reports the data-centre bottleneck is not demand — supportive of the driver, unsupportive of near-term timing; nets to intact.

No actions — no state changes. Every weight sits inside its conviction band (JPM, MSFT and AMZN in the 20–35% band for scores 8–9; NVDA in the 12–20% band for a score of 7), the largest sector is information technology at 43% against a 60% cap, and the largest theme is rates and capital markets at 33% against a 40% cap. Era drawdown is -4.51%, inside the normal stage, so no drawdown review is triggered. Broadcom remains blocked from re-entry under the five-session rule.

Institutional Signals

No new filings since the last report, so this is slow-moving context rather than news — and it is worth saying plainly on a night when nothing changed. Bridgewater's disclosed book remains index-first, with SPY and IVV together at 28.6% and NVIDIA at only 3.2%; that posture is more consistent with the broadening tape Friday showed than with the narrow mega-cap leadership of the last month. Berkshire stays concentrated in quality compounders and consumer credit — Apple 22%, American Express 17.1%, Alphabet 12.6%, Coca-Cola 10.9%, Bank of America 9.2% — which is a bank-and-brand book, not an AI book, and a reminder that the largest position in this portfolio is the one Buffett-style allocators would recognise. Pershing Square's May filing carries Amazon at 17.4% and Microsoft at 15.3%, direct confirmation on two positions, but it is three months stale and I am weighting it accordingly. Scion's November filing is too old to inform anything. On the crowding test, NVIDIA appears across two tracked filers but at modest weights and without parabolic extension, so the entry-is-late warning does not apply.

What Could Break It

The headline that would hurt the most is a data-centre capital expenditure deferral or a semiconductor gross-margin step-down. Microsoft, Amazon and NVIDIA are 67% of the book between them, and all three are paid, in different ways, by the same AI capex cycle. One credible hyperscaler announcement that 2027 data-centre spend is being pushed out would hit two-thirds of this portfolio simultaneously — the diversification here is across business models, not across drivers, and that is a deliberate concentration I am choosing to run rather than a blind spot.

Secondary risks, ranked: the US-Canada trade war widening into a broad tariff round, which pushes the term premium the wrong way and compresses the multiple on exactly that same 67%; a credit signal that is currently unverified turning out to be widening, which would make Friday's broad equity rally the wrong read; and the Iran escalation moving from sanctions and cyber-attacks on power infrastructure to something that puts an energy shock into a book with zero energy exposure. On JPMorgan specifically, the falsifier is a net interest income guide-down with a rising provision build — the trade war is precisely the mechanism that would produce it.

The risk I would name against my own call: if Friday's breadth is real and confirms Monday, holding a four-name mega-cap book and doing nothing is a decision with a cost, not a neutral act. I am accepting one session of that cost to avoid a seventh action in six sessions.