Viska macro book · monthly report · draft, data through July 24 · prepared by the intelligence platform

July 2026: the month the book rotated

Positioning, performance, and a decision-by-decision account of the largest theme rotation of the year. Every figure is computed from the broker statement and the platform's own records; nothing in this report is estimated by hand.

35% → 3%

AI weight, Jul 1 to Jul 24

43% → 71%

Minerals weight

$5.9M

Gross rotated between themes

+$385,750

Net new money (199 trades)

−$181,344

Realized in July (FIFO)

1 · What moved: allocation by the tape

Weights on invested capital. Line thickness scales with dollars traded.

July was a rotation, not drift. Net new money was under $0.4M while roughly $2.9M left AI and $3.6M entered Minerals and Energy Infrastructure. The price effect was negative in every theme, meaning the market paid the book nothing this month; every point of the reallocation was bought and sold deliberately.

Theme	Weight Jul 1 → 24	Net traded	Price effect	Realized (July)	Trades
Minerals & Metals	43.24% → 71.13%	+$2,756,843	−$83,333	+$50,984	78
Energy Infrastructure	12.72% → 19.64%	+$840,884	−$184,855	$0	24
Power Generation	8.81% → 6.11%	−$265,812	−$47,986	+$5,661	71
AI & Datacenters	35.24% → 3.12%	−$2,946,165	−$490,988	−$237,988	26
Total		+$385,750	−$807,162	−$181,344	199

2 · Decision accountability

Every material decision of the episode, graded against what was knowable at the time. This is the accountability view the CAO asked for; it is designed to become a standing section of every month report.

Decision	Grade	What was in hand	Outcome and lesson
Jun 17–18 · Sell $3.15M of the AI sleeve	GOOD	The delivered signal read bullish. The desk sold anyway.	Banked +$533k, eight sessions before the top. The desk's exit instinct beat the signal and deserves the credit.
Jun 29 · Re-enter $3.15M at the top, 4–6% above own exits	COSTLY	Delivered picture: 0.31 on 4 brokers, uncontested. The engine had already scored heavy contested deterioration on Jun 25, but that reading was never rendered in a brief.	The one decision that made July red. With the engine record consumed directly, a stated entry gate blocks this trade. Lesson: entries need a rule, and the rule needs the engine, not the email.
Jul 1 · Add $517k on the first break day	BLOCKABLE	The morning brief rendered 0.17 on 9 brokers, CONTESTED, the second heavy warning in two mornings.	Lot sold five sessions later at a $31k loss. This add was blockable on delivered information alone.
Jul 1–8 · Build Minerals +$2.76M, Infrastructure +$0.84M	ON THESIS	Doctrine allocation into the book's core theses.	Already +$51k realized inside July on Minerals trading. The rotation's receiving side executed cleanly.
Jul 6–8 · Unwind the AI re-entry within 2–4 sessions of the break	GOOD	Delivered read 0.28 contested; the full corroboration collapse printed a day later.	Realized −$246k on the day, but every exit was vindicated: by Jul 24 the sold names sat 0.4–20% below exit prices, roughly $222k of further loss avoided.
Jul 13 · Trim GOOGL/TSLA remnants	NEUTRAL	Signal recovered on thin breadth.	+$8k. Kept GOOGL subsequently marked down about 10%.

The pattern the month teaches: the desk's exits were consistently better than any signal we produced. Its entries were not. The July loss traces to a single re-entry that a stated, machine-checked entry gate would have stopped, worth $31k on delivered data and $244k on the engine's full record. Accountability going forward means grading both the fund's decisions and our signal, side by side, every month, exactly as this table does.

3 · The platform's own report card

The same accountability, applied to us.

What the signal got right

The theme conviction engine scored the AI deterioration by Jun 25, four days before the re-entry, with the heaviest evidence weight of the month.
The two decisive warnings (Jun 30: 0.12 on 10 brokers, CONTESTED; Jul 1: 0.17 on 9 brokers) were rendered and delivered on the exact mornings that mattered.
Scored broker calls behind them read, verbatim: "SHORTS, AI Vulnerable" (JPM), "take profits in secular growth" (BofA).

What we got wrong, stated plainly

The signal read bullish through the desk's profitable Jun 17–18 exit. It is an entry-discipline instrument, not an exit oracle.
The rendering layer dropped conviction blocks on 8 briefs including Jun 22–26, exactly the days that would have armed an entry gate before Jun 29. Measured cost of that gap: the difference between $31k and $244k of avoidable loss.
The scored series only exists since Jun 18. Our backtest is one episode; we will validate the gate forward, in paper, daily, before asking anyone to trust it with capital.

4 · What an extension buys: closing the loop

Three months built the instruments. The next phase connects them to decisions.

Daily (running now, hardened)

Ledger report with 8 deterministic integrity gates over the broker statement.
Theme universe re-derived every day with confidence-scored classification.
Morning brief with conviction blocks on every covered theme, rendering gap fixed, plus entry-gate NOTICE checks: concentration at peak, re-entry above own exit, contested-breadth warnings.

Weekly (new)

Positioning review: theme weights vs doctrine, drift, concentration.
Signal scorecard: what the engine said, what the book did, divergences named while they are actionable.
Actionable plan: entries the gate currently permits, holdings under warning.

Monthly (this document, automated)

Performance and allocation by the tape.
The decision accountability table, every material trade graded.
Our own report card beside it, both improving from a written record.

The decision layer (the pitch)

Entry gates computed from the engine record, not the rendered email.
Research feeds (beyond broker material) scored into the same signal spine.
Forward paper-validation with published rules, so by the next quarterly review the gate has a live, honest track record.

Method and provenance. Tape: IBKR Flex statement as of 2026-07-24, sha-verified; trades deduped by tradeID; weights on invested capital from daily SUMMARY position snapshots; net traded per line equals proceeds times FX, signed. Realized figures are the statement's own FIFO. Signal: production conviction pipeline export (2026-07-27) as the engine record; delivered-brief coverage cross-checked against the full 204-brief archive parse. Companion documents: the four-themes memorandum, July by the tape, the conviction-versus-tape study (all in this gallery). Draft for the fund; figures refresh automatically when the month-end statement lands.