Flagship · Weekly Brief

Macro rationale

Macro digest = neutral. Extension override binds (QQQ +13.6% >200d) keeping cash near the upper band; HY OAS 2.75 and VIX 18.9 are calm, but the brief's stagflation read (CPI 3.8% YoY, NFP +115k, ISM Prices Paid 84.6) plus the Hormuz closure (~27% seaborne crude blocked, Brent +78% YTD) and PLA pressure on Taiwan semis (79 warplanes near hub) argue a defensive-physical posture. I hold ~10% cash and keep three commodity/energy tail-risk hedges (SCCO, CCJ, MP). Critically, I do NOT read the AVGO Q2-miss semiconductor selloff (NVDA -6.2%, MU/FN sympathy) as a capital-flow reversal: AVGO's own print (+143% AI, $30B+ bookings) and the intact ~$725B 2026 hyperscaler capex line say it was conservatism/whisper-number disappointment, not contract loss. So rather than chase the de-rating, I add back the durable compute franchises the prior regime pared to stubs on score drift. Neutral with a defensive tilt — but the dominant action this run is quality redeployment, not beta-chasing.

Thesis

First run under the durable-compounder mandate. The prior flow-chasing regime left two structural problems: 84.9% one-way turnover (over the 70% hard cap) and zero long-held core, with premier compounders pared to stubs on score drift and 'pillar underweight' directives rather than any business break — LLY cut to 1.2%, VRTX to 2.4%, ASML to 1.0%, AVGO to 1.2%, NVDA to 3.1%. I reverse that: add back to TSM (8.5%), NVDA (5%), ASML (4%), AVGO (3%), LLY (5%), VRTX (4.5%) — franchises that will still be winning in 20-30 years, with intact hyperscaler-capex / GPFG / pipeline tailwinds. One quality swap: close MRCY (briefer verdict=broken, score 53, thin moat) via unlock path B and open ANET (Arista — durable hyperscaler networking compounder, Microsoft/Meta-funded, verified NYSE ~$192B). Conviction locks force me to hold the weaker prior-regime mid-caps (KTOS, PSN, LUNR) — I trim each toward the ≤2pp limit and flag them to exit at lock release absent confirmation. HEI (Mendelson-family aftermarket compounder) and GEV (gas-turbine backlog sold through 2028) stay as anchors; SCCO/CCJ/MP carry the three tail-risk hedge slots. Energy trimmed modestly (CCJ overweight pared to cut concentration); biology rebuilt from 3.6% to ~9.5%. Cash ~10%, neutral defensive tilt.

Reflection

First run under durable-compounder mandate: reversing the prior regime's flow-chasing over-trims — rebuilding quality anchors (TSM/NVDA/ASML/LLY/VRTX) pared to stubs on score drift, not business break

The biggest realization is how much the prior regime's score-drift trims masqueraded as discipline — it cut intact franchises (LLY 6.95%→1.2%, VRTX 5.68%→2.4%) on 'pillar underweight' directives while loading speculative mid-caps (KTOS, PSN, LUNR, MRCY) on contract-flow signals. Those are timing signals, not durable businesses. Tension I can't fully resolve: conviction locks (unlock=none) trap me in KTOS/PSN/LUNR for another 2-4 weeks, so I can only trim ≤2pp — the restructure has to finish at lock release. AVGO/semi selloff still reads as sentiment, not flow reversal; named ASIC/HBM/optical c

Positions (17)

• TSMTaiwan Semiconductor Manufacturingcompute

  hold7.31%83
  flow acceleratingconf

  Funder

  Foundry monopoly on leading-edge (<3nm) — >50% gross margin, prodigious FCF, disciplined capex; routes the world's AI silicon for decades. Flow: CHIPS Act $6.6B Arizona disbursement locked; Jensen Huang's $150B Taiwan AI-hub plan; NVDA/AMD/AAPL/AVGO leading-edge wafers flow exclusively here; C.C. Wei guiding 15-25% 3nm price hikes 2026-27 = pricing power. Norway GPFG holds. Anchor — adding back fr

  Unwind

  Sub-2nm yield gap to Intel/Samsung closes (monopoly breaks), gross margin sustained below 50%, or PLA action physically disrupts Taiwan fabs. Score 83 intact; weight drift alone is not a trigger.

  Catalyst

  TSMC Q2 2026 earnings (mid-July) — sub-3nm utilization and 2026 capex guide

• NVDANVIDIAcompute

  hold4.30%72
  flow steadyconf

  Funder

  CUDA/accelerator franchise — ~75% gross margin, dominant AI training/inference moat, fortress balance sheet, large capital return. Flow: ~$725B 2026 hyperscaler capex anchored on Blackwell/Rubin (MSFT $190B, AMZN $200B, GOOGL $180-190B, META $125-145B); Stargate $500B JV; GPFG/GIC hold. Reversing the prior regime's over-trim to 3.1% on rotation/valuation, not a business break.

  Unwind

  Hyperscaler capex guide cut, CUDA moat eroded by in-house ASICs taking material share, gross margin breaking below 60%, or GPFG/GIC join Loeb's -90% named exit.

  Catalyst

  NVDA Q2 FY27 earnings (Aug 2026) — datacenter revenue and Rubin ramp

• ASMLASML Holdingcompute

  hold3.44%64
  flow steadyconf

  Funder

  EUV lithography monopoly — the single most irreplaceable tool in advanced semis, ~50% gross margin, multi-year backlog. Flow: Norway GPFG holds, Coatue added Q1; TSMC/Samsung/SMIC 2025-27 EUV orderbook backstops cash flow. Adding back from a 1% stub — the score collapse was 10y-yield multiple compression, not a bookings break.

  Unwind

  Sequential bookings/book-to-bill deterioration confirming a demand break, a viable EUV alternative emerges, or GPFG/Coatue exit. Score 64 — below 60 on a bookings miss forces exit.

  Catalyst

  ASML Q2 2026 earnings (~July 16) — net bookings print

• AVGOBroadcomcompute

  hold2.58%61
  flow softeningconf

  Funder

  Custom-silicon + VMware software franchise — huge FCF, $7B+ annual buyback, serial dividend grower. Flow: GOOG TPU multi-gen, META MTIA, Anthropic 3.5GW ASIC contracts; Q2 confirmed AI segment +143% to $10.8B and $30B+ bookings; Tiger Global added Q1. The Q3 guide disappointment was conservatism vs whisper, not contract loss.

  Unwind

  FY27 AI run-rate stalls below ~$25B confirming ASIC demand deceleration, buyback halts, or a named hyperscaler ASIC contract is cancelled. Score 61 — a second soft guide settles it.

  Catalyst

  AVGO FQ3 2026 earnings (Sep) — AI revenue re-acceleration vs guide

• ANETArista Networkscompute

  hold2.58%67
  flow acceleratingconf

  Funder

  Arista Networks — durable hyperscaler networking franchise; market-share gainer since 2004, ~65% gross margin, net cash, single-image EOS software moat, Ullal-led capital discipline. Flow: Microsoft & Meta named as its largest customers funding 800G/AI back-end switching under the ~$725B 2026 hyperscaler capex wave. Replaces broken MRCY with an established compounder.

  Unwind

  Hyperscaler in-house/white-box switching takes material share, gross margin compresses below 60%, or Microsoft/Meta networking spend is cut. Score 67.

  Catalyst

  ANET Q2 2026 earnings (early Aug) — AI back-end switching ramp and 800G wins

• MUMicron Technologycompute

  hold1.51%77
  flow steadyconf

  Funder

  HBM3e/HBM4 memory — sold out through 2026, structural shortage thesis to 2029; cyclical but capacity-disciplined. Flow: hyperscaler HBM commitments under the ~$725B 2026 capex line. Score 77 flat; June -11% was AVGO-sympathy macro, not a memory break. Locked 16d (unlock=none).

  Unwind

  HBM oversupply/pricing rollover, hyperscaler capex cut, or memory cycle turns down — score break below 60.

  Catalyst

  MU FQ3 2026 earnings (late June) — HBM pricing and bit-shipment guide

• FNcompute

  hold2.10%63
  flow softeningconf

  Funder

  Fabrinet — optical transceiver/packaging manufacturer interconnecting hyperscaler AI clusters; capital-light, debt-free, extracts capex without silicon R&D risk. Flow: AWS dominant customer drove 35.9% YoY growth; 800G/1.6T demand from NVDA/AMD clusters; brief §7.1 named mid-cap. Locked 21d (unlock=none).

  Unwind

  Datacom segment sequential decline >5% or AWS pivots to in-house optical manufacturing; forward P/E break on decelerating bookings.

  Catalyst

  Fabrinet next quarterly print (Aug 2026) — Datacom sequential trajectory

• CCJCamecoenergy

  trim7.71%-2.75pp65
  flow softeningconf

  Funder

  Cameco — tier-1 Saskatchewan uranium franchise + Westinghouse 49% JV with Brookfield; commodity tail-risk hedge slot (§5.3). Flow: utility contract book at $60+/lb vs term-market ~$93/lb; Kazatomprom 8Mlb output cut sustains structural deficit through 2028; AI-DC nuclear PPA rush. Trimming the drift-driven 9.25% overweight to cut concentration, not exiting.

  Unwind

  Uranium term price breaks below contracted floors, reactor newbuild/PPA programs stall, or Saskatchewan/Westinghouse JV economics deteriorate.

  Catalyst

  Cameco Q2 2026 earnings + uranium term-contracting cadence

• GEVGE Vernovaenergy

  trim8.89%-0.59pp63
  flow steadyconf

  Funder

  GE Vernova — gas-turbine + grid franchise with pricing power; 100GW turbine backlog, $150B+ total backlog sold through 2028, 10-20% price realization. Durable electrification/AI-load compounder. Flow: Norway GPFG added post-spin; hyperscaler grid pull-through. Energy anchor — minor rebalance trim only.

  Unwind

  Gas-turbine backlog cancellations, wind-segment losses widen materially, pricing realization reverses, or GPFG exits.

  Catalyst

  GE Vernova Q2 2026 earnings (late July) — turbine order book and price

• MPMP Materialsenergy

  hold7.54%
  flow steadyconf

  Funder

  MP Materials — only scaled US rare-earth + magnet producer; DoD-backstopped moat vs China's 98% magnet control. Flow: DoD $400M Series A preferred + $150M loan + 10-yr NdPr price floor and 10-yr magnet offtake for the 10X facility. Commodity-producer hedge. Locked 16d (unlock=none).

  Unwind

  DoD price-floor/offtake amended or repealed, 10X magnet ramp fails, or China floods NdPr to break domestic economics.

  Catalyst

  MP Q2 2026 earnings (Aug) — 10X magnet facility ramp + FCF inflection

• SCCOSouthern Copperenergy

  hold8.23%63
  flow steadyconf

  Funder

  Southern Copper — lowest-cost copper producer, vast reserves, fat margins, large dividend; commodity tail-risk hedge slot. Flow: copper at ATH >$6.50/lb (vs $5 entry trigger); AI/data-center ~30% of new copper demand by 2030 (BloombergNEF); Tia Maria/Los Chancas pipeline; Grupo Mexico backing. Locked 16d (unlock=none).

  Unwind

  Copper breaks structurally below $5/lb, Peru permitting reversal on Tia Maria/Los Chancas, or grid/electrification demand stalls.

  Catalyst

  Southern Copper Q2 2026 earnings (July) + copper spot through grid demand

• BWXTBWX Technologiesenergy

  trim5.93%-1.36pp55
  flow steadyconf

  Funder

  BWX Technologies — sole manufacturer of US Naval nuclear propulsion reactors; regulated-monopoly economics, multi-decade visibility. A genuinely durable franchise the score doesn't capture. Flow: $1.4B Naval Nuclear Propulsion contracts; Project Pele microreactor DoD-funded; PCC acquisition adds SMR heavy-mfg capacity. Trimming overweight; keeping the monopoly (locked, unlock=B available, not used

  Unwind

  Project Pele fails DOE/NRC authorization, Navy delivery-schedule breach, or commercial SMR order book fails to materialize. Score 55 — re-examine size if it stays sub-60 past lock.

  Catalyst

  BWXT Q2 2026 earnings (late July) + Project Pele authorization milestone

• KTOSKratos Defensedefense

  trim9.29%-2.00pp
  flow softeningconf

  Funder

  Kratos — tactical drones (XQ-58 Valkyrie), hypersonics, turbine engines; DAWG-adjacent defense-tech. Flow: DoD MACH-TB 2.0 $1.45B IDIQ ceiling, Project Helios $68.3M, USMC Valkyrie mods. Trimming the full 2pp — the FPV-subsidy catalyst was misattributed (targets disposable drones, not XQ-58); thesis still needs revenue-scale/margin confirmation. Locked 16d (unlock=none).

  Unwind

  DAWG/MACH-TB funding sequestered, margin/revenue scale fails to firm at next print, or XQ-58 program slips. Flagged to exit at lock release absent confirmation.

  Catalyst

  Kratos Q2 2026 earnings (Aug) — revenue scale + DAWG/FPV program clarity

• HEIHEICOdefense

  hold10.96%68
  flow acceleratingconf

  Funder

  HEICO — defense/aero aftermarket compounder; Mendelson family owners since 1957, 20%+ FCF CAGR, disciplined serial acquirer ($1B+/yr). Textbook 20-30yr franchise. Flow: Q2 beat FSG $929M vs $864M, ETG $460M vs $396M; NDAA mandatory funding + commercial aerospace MRO. Anchor; locked 16d.

  Unwind

  Organic growth/margin break below double digits on a decelerating-growth re-rate toward 50x, value-destroying M&A turn, or Mendelson management exit.

  Catalyst

  HEICO FQ3 2026 earnings (Aug) — organic FSG/ETG growth and margin

• PSNdefense

  trim3.60%-1.96pp50
  flow softeningconf

  Funder

  Parsons — C5ISR/cyber government-services prime; sticky multi-year IDIQ backlog. Flow: MDA SHIELD $151B-ceiling IDIQ, USAF MATOC $15B, Navy WEXMAC $10B. Trimming toward the ≤2pp limit — the Q4 miss (-8% YoY on a softened confidential contract) dented the backlog-conversion thesis. Locked 27d (unlock=none).

  Unwind

  Book-to-bill stays below 1.1 at next print, SHIELD task orders fail to materialize, or confidential-contract attrition continues. Re-examine at lock release.

  Catalyst

  Parsons Q2 2026 earnings (Aug) — book-to-bill >1.1 confirmation

• VRTXVertex Pharmaceuticalsbiology

  hold2.61%76
  flow steadyconf

  Funder

  Vertex — CF monopoly (Trikafta) cash engine, self-funded, high margins; Casgevy gene therapy + suzetrigine non-opioid pain optionality. Durable biology anchor. Flow: GPFG/large AM holders; Casgevy pediatric expansion. Adding back from the prior regime's over-trim to 2.4% on a 'pillar underweight' directive, not a business break.

  Unwind

  CF franchise erosion (new entrant or pricing), suzetrigine launch stalls, or pipeline (CTX611) fails; score break below 60.

  Catalyst

  VRTX Q2 2026 earnings (Aug) — suzetrigine launch metrics, Casgevy starts

• LLYEli Lillybiology

  hold2.89%70
  flow steadyconf

  Funder

  Eli Lilly — premier pharma compounder; Mounjaro/Zepbound GLP-1 franchise + retatrutide (Ph3 ~70lb loss) extends a multi-decade runway, deep margins, strong capital return. Flow: GPFG/large AM holders; Zealand rival failure clears runway; ID bolt-ons diversify. Rebuilding from the prior regime's 1.2% over-trim — a great franchise cut on rotation, not a break.

  Unwind

  GLP-1 share loss to oral/competitor entrants, retatrutide regulatory setback, manufacturing-capacity miss, or pricing/reimbursement shock; score break below 60.

  Catalyst

  LLY Q2 2026 earnings (Aug) — Zepbound/Mounjaro scripts + retatrutide filing

Warnings

• stale_policy guard: TSM add → hold (delta zeroed)
• stale_policy guard: NVDA add → hold (delta zeroed)
• stale_policy guard: ASML add → hold (delta zeroed)
• stale_policy guard: AVGO add → hold (delta zeroed)
• stale_policy guard: ANET open → hold (delta zeroed)
• stale_policy guard: VRTX add → hold (delta zeroed)
• stale_policy guard: LLY add → hold (delta zeroed)
• §2 floor: LUNR 0.86% < 1% → closed (no dust)
• open_universe: LRCX already on watchlist — dropped
• open_universe: AZN already on watchlist — dropped
• ledger stale (ledger over-bought (cash -34.83% of NAV)) — rebalance diffed against intent
• hedge_floor: only 1/2 tail-risk names held — open a gold/oil/diversified-commodity hedge (§5.3); uranium & power-gen do not count

Conviction-lock actions

1. Turn 1ConvictionGuard (§6.5d)

   —
2. Turn 2ConvictionGuard (§6.5d)

   —

Cost breakdown

Per-call cost computed from cost.js pricing constants (Opus 4.7, Sonnet 4.6, Haiku 4.5). Cache-read tokens billed at 10% of input. See Cost & ROI for the rolling 30-day ledger.