ViskaStrat × ViskaRes · three-layer study · 2026-07-27 · operator-directed

July, three layers: what brokers signalled, what markets did, what the fund chose

Layer 1 is the ViskaRes broker-signal ledger: 36 dated signals compiled from broker-authored documents, desk notes and chart images only. Layer 2 is market reality as the fund's own book experienced it: flow-neutral price indices per theme, moved only by marks, never by the fund's trading. Layer 3 is the tape: every net dollar traded, by theme, by day. No conviction-engine output appears anywhere on this page.

One clock, five lanes

▲ bull signal▼ bear signal○ neutral glyph size = editorial weight w1–w3 bars = fund net traded (green buy / red sell) line = flow-neutral index, 100 = Jun 30 copper dotted verticals = dated market anchors (eye-read, alignment only) glyphs sit on the AVAILABILITY morning (first readable in the broker flow); dateline in tooltip where it differs (17 of 36)
Jul 06 Jul 13 Jul 20 Jul 27 tape ends Jul 24 Brent low ~$71 gold $4,027 · 10Y 4.60% SPX −1% · Asia AI −5..8% Brent $100 · 30Y 5.17% token index −26% AI & Datacenters 80 90 100 79 2026-07-01: net +517,500 USD 2026-07-06: net -2,715,420 USD 2026-07-13: net -748,245 USD 2026-07-01 [neutral w1] Rampart: the small-model thread opens — Goldman's 1-delta desk flags Rampart, a 14.7MB in-browser model doing privacy redaction with no server at all, as 'probably the most interesting release overnight'. The efficiency threat to hyperscaler compute demand is on a desk note by 1 Jul. (GS 1-delta via Zero Hedge, 30 Jun-1 Jul) avail 2026-07-05 (dated 2026-07-03) [bear w3] Blackstone exits data centers — Blackstone sells its Northern Virginia data-center stakes for $3.5bn, then QTS walks away from Prince William Digital Gateway, the 2,100-acre project that would have been the world's largest campus, after years of resident lawsuits. Smart-money exit plus community resistance, ten days before NY's permit pause. (Blackstone/QTS coverage, 3-4 Jul) avail 2026-07-05 (dated 2026-07-04) [bear w2] Token economics warning goes public — The Silicon Data token-expenditure index is down almost 20% from its May high and falling 11 of 12 days. Allianz puts the AI investment-vs-sales gap at 46%, worse than the 32% at the 2001 telecom bust. The honest caveat is printed too: the index blends price and usage, so the bull read is 'digestion'. (Bloomberg via Zero Hedge, 4 Jul) avail 2026-07-10 (dated 2026-07-07) [bear w3] Goldman: start hedging AI debt — Ten days before the 'carnage' headline: Goldman urges clients to hedge the AI debt bubble as ORCL falls 50% from its 2 Jun high, CDS widens across ORCL/CoreWeave/SpaceX, and Amazon scrambles a $25bn bond sale to prefund capex before the window shuts. The credit thread of Act 2 starts here. (Zero Hedge / GS, dated 7 Jul; flowed 10 Jul) 2026-07-10 [bear w3] 'Token unmaxxing': price cuts — AI model makers cut token prices; Goldman's 1-delta desk warns of a 'very rotational market'. The revenue side of the buildout starts deflating in public. GPT-5.6 and Grok 4.5 launch into it. (Zero Hedge / GS 1-delta, 10 Jul) 2026-07-10 [bear w2] Momo unwind goes systemic — 'Samsung, semis, the strait and sanctions spark market maelstrom, momo meltdown.' Samsung fades on the GS TMT desk two days earlier; Korea becomes the stress transmission channel. (Zero Hedge, GS TMT, 8-10 Jul) 2026-07-08 [bear w3] The financing question is named — MUFG asks 'Can future cashflows justify the AI buildout' the same day Goldman's desk covers datacenter ABS and data sovereignty. The month ends with the answer trading at 95 cents on the dollar. (MUFG Asia FX talk, GS Basics, 8 Jul) avail 2026-07-14 (dated 2026-07-13) [neutral w1] Semis vol at dotcom frequency — Semiconductors moved 3%+ fifteen times in thirty sessions, last seen in the dotcom era; SOX had been ~80% above its 200dma in early June, exceeded only by 2000, now back to a normal band. The same note insists most fundamental data still points the other way -- two-sided by its own author. (The Market Ear, 13 Jul; flowed 14 Jul) 2026-07-14 [neutral w2] 'Useful output per watt' — Goldman's Basics leads with 'useful output per watt'. The debate moves from how much compute gets built to what each watt returns -- the framing the deliverability critique needs. (GS Basics, 14 Jul) avail 2026-07-20 (dated 2026-07-14) [bear w3] NY pauses 50MW+ permits — Hochul signs a one-year environmental-permitting pause on large data centers. 155 local moratoria already active across 40 states; TX links grid-upgrade costs to large loads. Deliverability risk becomes policy, not opinion. (NY EO 14 Jul, moratorium note in 20 Jul flow) NOTE: The EO itself was public news on 14 Jul; `available` measures the broker-flow channel only. Anyone watching general news had this six days before the flow carried it -- grade accordingly if the fund trades on news. avail 2026-07-20 (dated 2026-07-17) [bear w3] 'Age of scaling is over' — Goldman's desk calls the hardware liquidation a de-grossing event and frames CXMT (200x oversubscribed) as China creating a fourth scaled DRAM competitor; 'China erases America's AI lead' runs the same evening, and the Kimi K3 alarm pieces follow through the weekend. The open-weights shock has its broker moment here, not on the 26th. (Zero Hedge / GS 1-delta, dated 17 Jul; flowed 20 Jul) 2026-07-15 [neutral w2] BofA splits the semis complex — 'Core Wars: insatiable demand for servers vs. not-so-much others.' The complex stops trading as one block: server/HBM demand intact, everything else softening. Capex to $60-64bn at TSMC while its ADRs fall on the print. (BofA semis Q2 preview, 15 Jul; JPM tape 16 Jul) 2026-07-17 [bear w3] 'Did Goldman pop the AI debt bubble?' — Hyperscaler bond 'carnage' becomes the headline. For the first time in the unwind, Mag 7 and semis are sold together -- capex funder and capex receiver repriced at once. Moonshot's model release knocks Asia AI baskets 5-8% the same session. (Zero Hedge 17 Jul; JPM morning briefing 17 Jul) 2026-07-24 [neutral w2] UBS: 'not breaking, maturing' — The strongest counter-view of the month, dated: amid 'token optimization and open-model invasion', UBS argues the AI trade is maturing, not breaking. The bull case shifts from growth to discipline. (UBS via 24 Jul flow) avail 2026-07-27 (dated 2026-07-26) [bear w3] The evidence lands in one payload — Academy: 'The Upcoming AI Spend Slowdown'. Token spend -26% from its 31 May peak. SOXX shares outstanding at a record while price falls. Meta's Hyperion SPV bond at 94.9. Goldman: $31tn of market cap added vs $9tn baseline AI revenue PDV -- and the PDV swings 4x on the discount rate alone. (26 Jul payload: Academy, Zero Hedge, Goldman) avail 2026-07-27 (dated 2026-07-26) [bear w2] Open-weights frontier — Kimi-K3 ranks #1 on Arena's code leaderboard, ahead of every closed US model; forecast medians put a Chinese frontier-class model at Feb 2027. Scarcity value of closed models is the collateral under the capex. (Jensen Huang open-weights note, 26 Jul; Arena chart) Minerals & Metals 80 90 100 98 2026-07-01: net +1,725,528 USD 2026-07-07: net +869,002 USD 2026-07-08: net +243,600 USD 2026-07-17: net +306,398 USD 2026-07-22: net -387,685 USD 2026-07-06 [bull w2] 'Chipflation': memory crisis named — MS names the memory shortage a crisis to navigate; IBM later confirms clients front-running 'supply-constrained infrastructure ahead of expected price increases' in its Q2 pre-announcement. (Morgan Stanley 6 Jul; IBM letter via JPM 14 Jul) avail 2026-07-13 (dated 2026-07-11) [bull w3] Memory prices reach the Fed — FOMC minutes warn inflation 'would remain elevated due to strong AI-related demand'; the Fed's semi-annual report repeats it. Apple, Microsoft and Dell hike device prices up to 25%. Chipflation stops being a sector story and becomes a monetary-policy input -- the bridge from the minerals theme to the rates channel. (Zero Hedge, dated 11 Jul; flowed 13 Jul) avail 2026-07-17 (dated 2026-07-16) [bear w2] China prices a $10bn DRAM challenger — CXMT prices its IPO -- 'a formidable DRAM competitor'. The memory-scarcity thesis gets its first structural supply answer, from China, at the top of the pricing cycle. (CXMT IPO coverage in 17 Jul flow) 2026-07-17 [bear w1] BofA: 'Gold's lost year' — With gold above $4,000, BofA titles the year lost -- gold up but flat-to-down real, lagging silver badly. A week later silver leads the complex at +4.3% w/w with gold at +0.9%. (BofA 17 Jul; tape w/e 24 Jul) avail 2026-07-20 (dated 2026-07-17) [neutral w2] Central banks flood back into gold — China buys 48 tons. UBS reads gold as consolidating and finding a base, holding $4,000 despite renewed hike repricing and rising real yields -- 'underlying support remains in place'. The official-sector bid against the 'lost year' label three days earlier. (Zero Hedge / UBS, dated 17 Jul; flowed 20 Jul) 2026-07-23 [bull w2] Copper prints validate the supply thesis — SCCO 2Q 'in line with strong expectations'; Goldman metals desk constructive on BHP/S32 copper mid-month on funded brownfield growth. Silver leads the metals complex into the week's close. (GS LatAm copper 23 Jul; metals tape) Power Generation 80 90 100 100 2026-07-08: net +886,808 USD 2026-07-20: net -1,152,620 USD 2026-07-01 [bull w2] Buy-the-dip on crude at $72.5 — Import-demand research argues crude buying picks up quickly with Brent at $72.5; JPM follows with 'The Return of Oil' and 'Saudi's potential' within the week. (broker research in the 1 Jul flow; JPM 9-10 Jul) 2026-07-09 [bear w2] HSBC cuts oil: 'mini glut now, surplus later' — Sell-side capitulation printed within days of the bottom. Brent had based at ~$71 on 30 Jun; it was above $100 within four weeks of the note. (HSBC oil markets, 9 Jul) avail 2026-07-21 (dated 2026-07-20) [bull w3] Oil jumps most since the war started — 'Kimi, Kuwait and Korean chaos cap ugly week for tech; oil jumps most since start of war.' Houthis extend strikes to Saudi tankers; Brent takes out $100 by the 23rd, +41% off the 30 Jun low. (Zero Hedge w/e 20 Jul; Brent chart) 2026-07-23 [bull w3] Power equipment decouples from the AI selloff — GE Vernova: 'record orders, outlook encouraging, buy the dip.' Siemens Energy read-across: 'stronger gas service outlook.' The picks-and-shovels leg holds its bid while announcement-driven AI exposure is repriced. (Goldman on GEV / Siemens Energy, 23 Jul) avail 2026-07-27 (dated 2026-07-26) [bull w2] Supply shock with no Western buffer — OECD commercial oil stocks at the 8th percentile since 2017, diesel 4th, global SPR 1st -- while China's stocks rose to the 72nd. Decomposition: supply +$45/bbl, demand -$18. The shock is structural, not positional. (oil-stocks and decomposition charts, 26 Jul payload) Energy Infrastructure 80 90 100 88 2026-07-07: net +666,876 USD 2026-07-13: net +174,008 USD 2026-07-13 [bull w2] Goldman defends energy against pushback — 'Energy, Utilities & Mining Pulse: responding to investor pushback, 10 ideas.' The desk is defending an out-of-consensus overweight days before the oil leg validates it. (Goldman EU&M pulse, 13 Jul) 2026-07-22 [bull w1] Goldman raises TTF forecast — European gas forecast raised amid 'European energy concerns' running since mid-month. UK NatGas up ~4% on single sessions repeatedly in the JPM tape. (GS morning 22 Jul; JPM tape) Macro (signals only, no fund sleeve) 2026-07-01 [neutral w1] Fed independence ruling absorbed — Supreme Court ruling leads the morning notes; MUFG frames it 'Fed independence maintained, for now'. Gasoline below $4.00/gal for the first time since March supports the disinflation read. (GS morning, MUFG Asia FX, Raymond James) 2026-07-02 [neutral w1] 'The hawkish Fed trade is nearing its end' — A dated contrarian rates call, printed 2 Jul. Three weeks later September hike odds stood at 70.9%. The flow carried both sides of the rates debate all month; this one lost. (2 Jul flow) 2026-07-06 [bear w2] September-hike debate opens — Morgan Stanley publishes 'Assessing the Case for a September Hike' and 'Hike to Wait and See' the same day. The month's dominant macro thread starts here, three weeks before hike odds peak. (Morgan Stanley, 6 Jul) avail 2026-07-18 (dated 2026-07-17) [bear w2] TS Lombard: 'We are no longer bullish equities' — A dated, unhedged capitulation from a house that had been constructive -- printed into the week the S&P closes below its 50DMA. BofA vol desk flags single-stock vs index vol at 'dotcom extremes' the same week. (TS Lombard 17 Jul; BofA vol insights) avail 2026-07-20 (dated 2026-07-18) [bear w2] Retail buys what the pros sell — Tech funds took $71bn of inflows in six weeks, $50bn of it into semiconductors (DB), while hedge funds sold tech in six of eight weeks -- the largest cumulative selling in 10+ years of Goldman records -- and corporate insiders sold at a near-record pace. The month's positioning divergence, quantified. (weekend chart packs, 18-19 Jul; flowed 20 Jul) avail 2026-07-27 (dated 2026-07-20) [bear w3] Bear flattener: 2Y +18bp in four sessions — September hike odds ramp to 70.9% -- the chart labels oil as the driver. The 30Y nominal touches 5.17%, level with the 2007 high. Rate risk becomes the AI trade's discount-rate problem. (rate-odds and curve charts, 20-23 Jul) avail 2026-07-27 (dated 2026-07-26) [bear w2] The political feedback loop closes — Gasoline back above $4; Trump approval at 40.9 and tracking pump prices inversely. Food-crisis warnings return as the Iran conflict runs. The reflexivity argument: crude forces policy, policy reprices duration, duration reprices AI. (approval/gasoline chart; food-crisis note, 26 Jul)

Signal vs action: the alignment scorecard

Each signal cluster against what the fund did next, graded on outcome through Jul 24 marks. ALIGNED = traded with the signal; AGAINST = traded opposite it.

Signal cluster (layer 1)Fund action (layer 3)VerdictOutcome (layer 2)
Jul 3–10: six bear-AI signals, four at w3 (spend slowdown, debt-hedge call, investor fatigue, exploit-window) — all available in the flow by Jul 5–10, timestamp-verified Sold −$2.72M Jul 6, trimmed −$0.75M Jul 13; AI weight to 3% ALIGNED AI index fell from ~91 at exit to 79.2 by Jul 24. Exit was with the flow and right.
Jul 9: sell-side cuts power/oil at the bottom (HSBC class) Had bought +$0.89M Jul 8; held through the cut AGAINST, RIGHT Power index bottomed 90.4 on Jul 9, the exact day of the cut, then +11.8% to 101.1 by Jul 22. Brent +41% off the Jun 30 low.
Jul 20 / 23 / 26: strongest bull cluster of the month on Power Generation (two w3, one w2) — which flowed Jul 21–26, after the trade: the Jul 20-dated entry is a week-end wrap that led the Jul 21 morning brief Sold −$1.15M on Jul 20 at index 93.0. Available ledger state at execution was mixed: Goldman energy defense (bull w2, Jul 13) vs the Jul 9 sell-side cut (bear w2) AHEAD OF SIGNAL, WRONG Index +8.7% in the next two sessions (93.0 → 101.1). Wrong on outcome, but not a trade into a visible bull cluster: the cluster arrived starting the next morning.
Jul 6: minerals bull (supply tightness) Added +$0.87M Jul 7, +$0.24M Jul 8 ALIGNED Minerals index 101.2 → 95.3 near-term, recovered to 100.4 by Jul 22. Neutral-to-fine.
Jul 17 minerals, mixed available state: a w3 BULL four days old (memory pricing reaches the FOMC minutes, available Jul 13) against two fresh lighter bears (w2 + w1, readable that morning) Added +$0.31M on Jul 17 — siding with the heaviest available signal WITH HEAVIER SIGNAL, RIGHT Minerals index +5.2% Jul 17 → 22 with silver leading; HL +9.8% Jul 20 → 22. (Gap-week ledger rows reclassified this from "against two bears" — fairness cuts both ways.)
Jul 13 / 22: energy-infrastructure bull Added +$0.17M Jul 13 (after +$0.67M Jul 7) ALIGNED Infra index range-bound 88–93; the sleeve's story in July was resilience, not rally.

The four frames, answered from the book

1 · Did picks-and-shovels insulate the fund from the AI repricing?

Partly, but the real insulation was the exit. The AI sleeve index fell to 79.2 (−21% on the month); the fund watched almost all of it from a 3% weight, having sold at index ~88–91. The picks-and-shovels expression that remained (VRT, ETN, NVT classed as infrastructure) held flat through the Jul 23 break (ETN +1.0%, VRT +0.3% Jul 17→24) while kept GOOGL fell −7.8%. The stance helped; the timing decision did the heavy lifting.

2 · The oil miss: with the flow or with the desks?

With the flow, against the desks, and right. The fund bought power generation +$0.89M on Jul 8; the sell-side cut came Jul 9, the exact bottom of the sleeve index (90.4); Brent then ran +41% off the Jun 30 low to $100 by Jul 23. The blemish is the other side: the Jul 20 trim of −$1.15M, executed on mixed available signals a day before the month's strongest bull cluster flowed, two sessions before the index topped at 101.1.

3 · The Jul 23 decoupling: infrastructure vs AI

Confirmed in the fund's own marks. On Jul 23 the AI sleeve fell −7.2% while the infrastructure sleeve rose +1.3%. Name-level nuance: GEV itself had already given back its run (−4.1% Jul 17→24) despite the record-orders print, so the decoupling is a sleeve-level fact, carried by the electrical-construction and equipment names, not by any single stock.

4 · Silver leadership vs the metals sleeve

The fund owns the lever. In the leadership week the minerals sleeve rose +5.2% (Jul 17→22) with the silver-heavy names in front: HL +9.8% and FCX +10.6% Jul 20→22 against NEM +7.3%. The Jul 17 add of +$0.31M sided with the heaviest available signal (the w3 memory-to-FOMC bull) over two fresh lighter bears, and caught that move.

Month verdict, three layers together: the fund traded WITH the broker flow on AI and was right; it held power through a sell-side bottom-tick cut and was right; its minerals add sided with the heaviest available signal and was right; and its one losing trade, the Jul 20 power trim, preceded the strongest bull cluster of the month by a day. On the full 36-row ledger, every graded decision made on an available signal was right — the only loss came where the decisive signals had not yet flowed. Which is the argument for the decision layer: not signal-follows-fund or fund-follows-signal, but both graded on one availability clock, every month, with divergences named while they are still actionable.

Caveats

Method and provenance. Layer 1: ViskaRes july-broker-signals.json, 36 signals (30 original + 6 from the 171-document gap-week conversion), broker-authored material only, each row carrying its binding availability timestamp; chart readings traceable to cropped figures. Layers 2–3: IBKR Flex statement as-of 2026-07-24, sha-verified; SUMMARY snapshots deduped by (reportDate, conid); trades deduped by tradeID; theme mapping from the universe store (full held book, zero unmapped names); flow-neutral indices chained on names held across consecutive days, rebased 100 = Jun 30. Theme keys join 1:1 to the four-major-theses names via the vocabulary crosswalk. No conviction-engine data on this page. Data file: three-layer.json alongside this page. Authors: ViskaStrat (layers 2–3, join, verdicts) and ViskaRes (layer 1). Draft for the operator.