Flagship · Bulletin

Macro rationale

Neutral stands. The only material macro change since 9/08 is mechanical: the extension escalator CLEARED as QQQ fell from +9.5% to +7.59% above its 200d, so the §5.4 NEUTRAL ceiling snaps back from 25% to 15% and my carried 21.77% intent cash is now above-band. That is a named cause to deploy — not a new risk view. The defensive substance of the paper's stagflation read is intact and argues against leaning harder into beta: 25% of sectors above their 50d (narrow), 10y real yield 2.43% (genuinely restrictive), EIA lifting 2026 Brent to $91 with WTI +16% in a month on the Hormuz exclusion zone, SPR below 290M bbl, and copper at a record above $6.80/lb. Credit and vol remain complacent (VIX 16.46, -9% vs 200d; HY OAS ~2.65-3.0) — the same complacency gap that keeps me at the top of the base band rather than the bottom, and keeps both §5.3 hedges (SCCO, NEM) at full size against $18B of August gold-ETF inflows. On the trigger itself: AI-load look-through at 31.76% is DRIFT above the 30% early-warning line, not a cap breach — the §2 hard cap is 35% and my intent-weight AI-load is 29.3% (compute intent 24.38% + GEV 4.92%). I will not trim a score-87 anchor (TSM, gross margin at an all-time high) to cure a warning line. The correct cure is exactly what I am doing: deploy the escalator-cleared cash entirely OUTSIDE the AI-capex cluster and add no compute weight. Rotation context (XLE +8.54pp, ITA -11.79pp rel SPY) is noted as when/how-hard only; it does not fire a trade.

Thesis

Two tickets, both funded by the escalator-cleared cash: add LLY +3.4pp to 8.04% and VRTX +3.4pp to 7.26%. Every other name is a verbatim hold at entry weight, delta 0. No exits. Why these two: the deployment had to land outside the AI-capex cluster (31.76% live, 35% hard cap) and outside info-tech (24.38% vs the 30% sector cap). Foundation remains mechanically un-openable — zero dossiers on file, fourth run running. Both §5.3 hedges sit at record commodity prices (copper all-time high, gold post-$18B-inflow), so adding there is a poor entry, and NEM's own briefing is weakening on a 13% production decline with AISC +24%. Defense is bound by KTOS's negative FCF margin (-31.9%) and HEI's stretched multiple. What remains is the pillar with the book's two best score drifts and its thinnest weight: VRTX 82 (+6, raised FY guide to $13.10-13.20B, Casgevy +151%) and LLY 74 (+4, GLP-1 script lead retaken from Novo). Both are 84-86% gross-margin cash generators, not the rate-vise cash-burn biotech the paper underweights. Mid-cap sleeve: considered TDY (RSI 30, defense ISR), CACI (SkyValor military-wide), DRS (RSI 25, $8.7B backlog), LEU (Oklo HALEU LOI) and the §3b setups (HON at 8.6x with Elliott live, KBR, FDXF). None open: this fire's scope lists no entry candidates, so opens are barred, and I will not spend the wake manufacturing one.

Reflection

No funder broke. The only real change is mechanical: the extension escalator cleared, so the 15% ceiling forces 6.8pp of parked cash into the two intact biology compounders.

The auto-fire called AI-load 31.8% a §2 breach; it is not — the hard cap is 35%, 30% is the early warning, and my intent-basis load is 29.3%. The 31.8% is price drift in TSM and AEIS. I refused to trim a score-87 anchor to cure a warning line, which feels right but I want it on record in case the cluster keeps climbing. Second, honest discomfort: I deployed by elimination. Compute is sector/AI-load bound, foundation is dossier-locked for a fifth run, both hedges are at record commodity prices, defense is bound by KTOS's -31.9% FCF margin. Biology won on score drift and cash generation, but it

Positions (16)

• MUMicron Technologycompute

  hold3.75%87
  flow acceleratingconf

  Funder

  Only US-domiciled leading-edge DRAM/HBM franchise: 72.5% gross / 65.4% operating margin, 48.8% ROIC, net cash (-0.3x), FCF+ — and after two decades of cycle abuse a three-player HBM oligopoly that is finally capacity-disciplined. Named flow: 16 Strategic Customer Agreements (~$100B contracted), HBM4 capacity pre-sold, $250B capex program, and the ~$600B 2026 hyperscaler capex line (CreditSights).

  Unwind

  Erosion of the SCA contract book or customer deposits returned; HBM oversupply as 2027 capacity lands; CXMT converting its ~11% DRAM share into leading-edge nodes; or gross margin reverting toward the 37% trough level that historically marks the cycle top.

  Catalyst

  FY Q4 earnings ~2026-09-24 (est): HBM4 allocation commentary and whether the SCA book extends past 2027.

  Scenarios · 12mo targets

  $1,150.00 base

  $1,500.00 bull — HBM4 pricing holds through the 2027 capacity add and the SCA book extends; memory keeps the scarcity rent it earns from NVDA/AVGO's cost line.

  $600.00 bear — 2027 capacity lands into softer AI order flow, HBM spot cracks, and the 72%+ gross margin mean-reverts toward the historical 37% trough.

• TSMTaiwan Semiconductor Manufacturingcompute

  hold5.83%87
  flow steadyconf

  Funder

  Effective foundry monopoly at <=3nm — every meaningful AI accelerator routes through it. >50% gross margin (a Sep-8 print confirmed all-time-high gross profit margin), prodigious FCF, famously disciplined capex, ~73% foundry and ~90% advanced-node share. Named flow: CHIPS $6.6B Arizona disbursement, Goldman's $920B-$1.4T 2027 AI-capex forecast naming TSMC explicitly, and multi-billionaire 13F rota

  Unwind

  Gross margin sustained <50% for two consecutive quarters; a rival closing the leading-edge yield gap; a Taiwan-strait event Arizona capacity cannot offset; or a customer-funded capex reversal that breaks the sold-out advanced-node book.

  Catalyst

  Monthly revenue prints and the October Q3 call — advanced-node mix and 2027 capex guide.

  Scenarios · 12mo targets

  $520.00 base

  $650.00 bull — N2 ramps sold out at premium pricing, CoWoS oversubscription persists, and margin holds at record while Arizona de-risks the geographic tail.

  $330.00 bear — A China/Taiwan escalation or an abrupt hyperscaler capex digestion phase compresses the advanced-node book and the multiple simultaneously.

• NVDANVIDIAcompute

  hold3.51%74
  flow steadyconf

  Funder

  AI-accelerator franchise with the CUDA software moat — 74.2% gross margin, 77% ROIC, fortress net-cash balance sheet, share count -1.3% YoY, and developer lock-in no ASIC rival has replicated in a decade. Named flow: ~$279B of disclosed supply commitments (doubled), Blackwell/Rubin anchoring the MSFT/AMZN/GOOGL/META capex lines, and Stargate.

  Unwind

  Data-center revenue YoY <20% for two consecutive quarters; gross margin breaking below the guided 71-72% floor because memory cost cannot be passed through; a named hyperscaler shifting a majority of training to in-house silicon; or a compute export-control regime removing a material share of demand.

  Catalyst

  FY Q3 earnings ~2026-11-25 (est): data-center growth rate and the 71-72% gross-margin floor.

  Scenarios · 12mo targets

  $250.00 base

  $320.00 bull — Rubin ramps on schedule, inference demand broadens beyond the top five buyers, and NVDA passes the HBM cost increase through without breaking the margin floor.

  $150.00 bear — Memory cost squeezes the guided margin floor while custom ASIC in-sourcing at two named hyperscalers slows data-center growth below 20%; multiple compresses on 2.43% real yields.

• ASMLASML Holdingcompute

  hold2.96%64
  flow steadyconf

  Funder

  EUV / High-NA monopoly — the single most irreplaceable tool in advanced semis, with no commercial-scale alternative at 2nm, a multi-year backlog, ~50% gross margin and an installed-base service annuity that earns through the cycle. Named flow: expanding High-NA collaborations with Samsung and Intel, the TSMC/Samsung/Intel 2026-28 EUV orderbook, and Norway GPFG holding with Coatue/Third Point added

  Unwind

  Book-to-bill <0.7 for three consecutive quarters; a Dutch/US export-control tightening removing a material share of the orderbook; or a credible non-EUV patterning path emerging at 2nm.

  Catalyst

  Q3 bookings print — the book-to-bill number is the whole tell on the score's export-control discount.

  Scenarios · 12mo targets

  $1,950.00 base

  $2,400.00 bull — High-NA orders convert from collaboration to volume purchase orders at Intel/Samsung/TSMC and bookings re-accelerate, closing the 21-point score discount.

  $1,250.00 bear — Further Dutch/US China licensing restrictions plus a leading-edge capex pause push book-to-bill under 0.7 for three quarters.

• AVGOBroadcomcompute

  hold2.84%63
  flow steadyconf

  Funder

  Custom-silicon ASIC leader (~60-70% share) plus VMware infrastructure-software lock-in — 68.3% gross margin, 44.3% FCF margin, $7B+ annual buyback, serial dividend grower, and Hock Tan's capital-allocation record. Named flow: GOOG TPU multi-generation, META MTIA and Anthropic ASIC programs; management guides AI chip revenue to double in FY27 and again in FY28.

  Unwind

  AI custom-chip revenue stalling below a ~$25B annual run-rate; full hyperscaler in-sourcing displacing the ASIC design franchise across multiple named accounts; or the VMware software base churning enough to break the 68%+ blended gross margin.

  Catalyst

  FY Q4 earnings ~2026-12-08 (est): AI semiconductor run-rate versus the doubling guide.

  Scenarios · 12mo targets

  $430.00 base

  $540.00 bull — AI silicon doubles as guided on the Google/Meta/Anthropic programs while VMware ARR compounds — the software leg carries the multiple through any capex wobble.

  $260.00 bear — A named hyperscaler pulls its ASIC program in-house, AI revenue stalls under a $25B run-rate, and the memory cost line eats the blended margin.

• ANETArista Networkscompute

  hold2.53%65
  flow steadyconf

  Funder

  Durable hyperscaler-networking franchise and a consistent share-gainer since 2004 — 63.5% gross margin, 43.9% operating margin, net cash (-2.9x), and the single-image EOS software moat that makes rip-and-replace prohibitive. Named flow: Microsoft and Meta named as largest customers funding ~45% YoY revenue growth; record $3B quarter with 45% margins and a 12.7% consensus estimate raise.

  Unwind

  Microsoft+Meta combined concentration >50% with revenue decelerating; Spectrum-X or whitebox displacing EOS in a NAMED hyperscaler footprint; or gross margin breaking below ~60% on Ethernet price competition.

  Catalyst

  Q3 earnings ~2026-11-04 (est): hyperscaler concentration disclosure and 800G/1.6T AI-cluster order commentary.

  Scenarios · 12mo targets

  $215.00 base

  $270.00 bull — Ethernet keeps winning the AI back-end against InfiniBand, EOS extends into campus, and the two named hyperscalers broaden into a third and fourth large buyer.

  $135.00 bear — Spectrum-X displaces EOS in a named hyperscaler footprint while customer concentration passes 50% into a decelerating capex quarter.

• GEVGE Vernovaenergy

  hold6.06%61
  flow acceleratingconf

  Funder

  Gas-turbine and grid franchise with genuine pricing power — heavy-duty turbines effectively sold out through 2030, 10-20% price realization on new orders, 48.4% ROIC, net cash, and an installed-base service annuity. Named flow: $176B backlog, Norway GPFG added post-spin, Oracle/Alphabet capex pull-through, the DOE bulk-power-security emergency order favouring domestic grid hardening, and a $400M S

  Unwind

  Gas-turbine backlog cancellation or a named hyperscaler power-capex cut; wind-segment losses widening enough to consume gas-segment profit; or the order book ceasing to price at 10-20% realization.

  Catalyst

  Q3 earnings ~2026-10-21 (est): backlog print and gas-segment margin, plus any DOE bulk-power procurement language.

  Scenarios · 12mo targets

  $1,100.00 base

  $1,350.00 bull — Brent near $99 and the DOE grid-security order accelerate gas + T&D orders; price realization holds while the wind drag finally rolls off.

  $650.00 bear — A hyperscaler power-capex digestion cancels turbine slots and widening wind losses swallow gas profit; the 61 score breaks the 60 floor.

• SCCOSouthern Copperenergy

  hold10.52%67
  flow acceleratingconf

  Funder

  Tail-risk hedge slot (§5.3): lowest-cost major copper producer with the industry's largest reserve base, 67.6% gross margin, FCF+ with a large dividend, and Grupo Mexico control enforcing capital discipline through the cycle. Named flow: copper at an all-time record above $6.80/lb on the tariff squeeze and AI power demand, H1 operating cash flow +117% YoY to $3.68B, and the $20.5B capex plan unblo

  Unwind

  Copper price collapse breaking the structural deficit thesis; Tia Maria / Los Chancas permitting reversal; a dividend cut signalling balance-sheet stress; or a Peruvian/Mexican fiscal or nationalization shock.

  Catalyst

  Tia Maria construction milestones and the US copper import-tariff decision that sets the domestic premium.

  Scenarios · 12mo targets

  $220.00 base

  $280.00 bull — The ~320kt structural deficit persists as AI/datacenter demand takes ~30% of new copper by 2030; Tia Maria volumes land into record pricing.

  $145.00 bear — A Chinese demand air-pocket or tariff reversal unwinds the speculative premium; today's -7.2% day shows how fast the record price can give back.

• NEMNewmont Corporationenergy

  hold6.74%83
  flow steadyconf

  Funder

  Tail-risk equity hedge (§5.3): largest gold producer with tier-1 long-life assets, 54.4% operating and 45.4% FCF margin, share count -5.3% YoY, and the Nevada JV dispute with Barrick resolved for $1.95B. Named flow: $18B of August gold-ETF inflows plus continued central-bank accumulation against a pinned Fed and 2.43% real yields.

  Unwind

  Cost overruns or mine disappointments breaking FCF and the dividend — the 13% YoY production decline and AISC to ~$1,680/oz are the live version of this; durable disinflation plus a real-rate surge removing the stagflation-hedge rationale; or a value-destroying acquisition.

  Catalyst

  Q3 earnings ~2026-10-22 (est): whether AISC guidance stabilizes and 2026 production holds the cut 5.26Moz line.

  Scenarios · 12mo targets

  $145.00 base

  $185.00 bull — Gold keeps bidding on ETF inflows and central-bank buying while cost guidance stabilizes — the FCF/dividend leg is restored on higher realized price.

  $95.00 bear — AISC keeps climbing past $1,680/oz on falling volumes, so a flat gold price no longer converts to FCF and the hedge degenerates into a cost-inflation short.

• KTOSKratos Defensedefense

  hold11.41%64
  flow steadyconf

  Funder

  Scarce non-prime supplier of attritable airframes (XQ-58A Valkyrie), hypersonics and turbine engines, with a decade of embedded program positions the primes cannot easily displace and vertical propulsion integration. Named flow: DoD MACH-TB 2.0 $1.45B IDIQ ceiling, Project Helios, USMC Valkyrie mods, $55M of fresh awards, and the FY2027 $1.5T national-defense request with C-UAS now a formal procur

  Unwind

  Backlog or contract cancellations; margin failing to inflect through the drone ramp with FCF staying negative through FY2027 (FCF margin is -31.9% today and share count +21.7% YoY — this is the live watch item); or a DoD unmanned-systems funding cut in the FY2027 appropriation.

  Catalyst

  Q3 earnings ~2026-11-03 (est) and the CR/FY2027 defense appropriation path — margin inflection is the whole question.

  Scenarios · 12mo targets

  $60.00 base

  $80.00 bull — Drone and hypersonics volume finally converts 30%+ revenue growth into positive FCF as the FY2027 topline lands; -37% YTD and RSI 34 leave a lot of room.

  $32.00 bear — FCF stays negative through FY2027 on supply-chain slippage while dilution continues at +21.7% share growth — that is the named unwind, not the price.

• HEIHEICOdefense

  hold10.23%68
  flow steadyconf

  Funder

  Defense/aero aftermarket compounder and a textbook 20-30 year franchise — the Mendelson family has owned and run it since 1957, compounding FCF above 20% through disciplined tuck-in acquisitions funded from cash flow; 40.3% gross margin, 24.1% operating margin, 17.9% ROIC, 1.6x net debt. Named flow: record Q3 with revenue +36.5% YoY and net income +49%, FSG and ETG both beating; NDAA mandatory fun

  Unwind

  Organic growth decelerating below 6% WITH margin compression; a value-destroying large acquisition breaking the tuck-in discipline; a Mendelson management succession failure; or the forward multiple holding above ~50x while growth normalizes.

  Catalyst

  Q4 earnings ~2026-11-26 (est): organic growth split and PMA parts approvals cadence.

  Scenarios · 12mo targets

  $360.00 base

  $430.00 bull — Aftermarket demand stays tight, tuck-ins keep compounding at 20%+ FCF growth, and the multiple re-rates back up from an RSI-22 washout.

  $240.00 bear — A commercial-aero downcycle drops organic growth below 6% with margin compression while a 50x-plus multiple has no cushion left.

• LLYEli Lillybiology

  add6.66%+3.40pp74
  flow acceleratingconf

  Funder

  Premier pharma compounder: 84.1% gross margin, 38.8% operating margin, 62.3% ROIC, share count -1.2% YoY, and a $27B manufacturing capacity moat rivals cannot replicate quickly on top of the tirzepatide franchise. Named flow: Sep-8 confirmation that Lilly retook the US GLP-1 weight-loss script lead from Novo, neuroscience revenue +32%, orforglipron oral approved Apr-2026, and revenue +67.9% YoY.

  Unwind

  Tirzepatide US script share falling >5pts to Novo over two quarters; retatrutide Phase 3 disappointing on safety or durability; IRA/MFP negotiation reaching the obesity franchise on price; or a capital-allocation turn away from the manufacturing build into large dilutive M&A.

  Catalyst

  Q3 earnings ~2026-11-04 (est): Zepbound/Mounjaro script share and orforglipron launch curve.

  Scenarios · 12mo targets

  $1,350.00 base

  $1,650.00 bull — Orforglipron converts the oral market Novo cannot serve at scale while retatrutide Phase 3 confirms ~28% weight loss — the $27B capacity build becomes the binding advantage.

  $850.00 bear — Novo's comeback plus Amgen entry take >5pts of script share, or MFP/IRA pricing reaches the obesity line and the premium multiple unwinds on 2.43% real yields.

• VRTXVertex Pharmaceuticalsbiology

  add6.02%+3.40pp82
  flow acceleratingconf

  Funder

  Durable biology compounder: the cystic-fibrosis monopoly (Trikafta/Alyftrek) is a self-funding cash engine on 86.2% gross margin and 30.6% FCF margin with patent protection into the late 2030s and a net-cash balance sheet (-1.2x). Named flow: FY revenue guide raised to $13.10-13.20B, Casgevy sales +151% YoY to $76M with an FDA label expansion (confirmed in partner CRISPR Tx's print), and score dri

  Unwind

  The CF franchise eroding faster than non-CF revenue can replace it; the Crinetics integration destroying value; a suzetrigine commercial failure removing the diversification leg; or R&D spend breaking the 30%+ FCF margin.

  Catalyst

  Q3 earnings ~2026-11-03 (est): Journavx/suzetrigine launch traction and Casgevy patient-start cadence.

  Scenarios · 12mo targets

  $620.00 base

  $750.00 bull — Suzetrigine converts the non-opioid pain market while Casgevy compounds off a +151% base — CF cash funds the diversification without dilution.

  $420.00 bear — Suzetrigine launch stalls, Casgevy patient starts stay boutique, and CF erosion arrives before non-CF revenue scales; Crinetics integration consumes the FCF margin.

• ARGXargenxbiology

  hold3.23%71
  flow steadyconf

  Funder

  The Vyvgart / Vyvgart Hytrulo (efgartigimod) FcRn autoimmune franchise: first-mover platform with a decade of indication runway, commercial-stage, high-margin and net cash, so it self-funds its own pipeline. Named flow: FDA gMG approval expanded to all serotypes including seronegative, the Forte Biosciences acquisition closed (FB102 anti-CD122), and Sanofi's riliprubart CIDP discontinuation removi

  Unwind

  Vyvgart revenue growth stalling; a pipeline indication failure in myositis or Sjogren's; a competitor FcRn entrant (J&J nipocalimab, UCB rozanolixizumab) taking measurable share; or Forte integration spend breaking the self-funding profile.

  Catalyst

  Next quarterly Vyvgart revenue print and the myositis/Sjogren's readouts; Forte FB102 proof-of-concept data.

  Scenarios · 12mo targets

  $1,150.00 base

  $1,400.00 bull — Hytrulo subcutaneous conversion accelerates while CIDP share consolidates post-Sanofi exit and a second indication clears — the platform re-rates on breadth, not one drug.

  $720.00 bear — Nipocalimab takes measurable gMG/CIDP share, a pipeline indication fails, and Forte integration spend turns the self-funding story into a cash burn.

• AEISAdvanced Energy Industriescompute

  hold2.53%69
  flow steadyconf

  Funder

  Starter (§6.6). Advanced Energy: 45-year incumbent in precision RF/DC plasma power delivery, designed into etch and deposition process recipes at the major WFE OEMs — requalifying a power supply means requalifying the recipe, which is the switching cost. 38.9% gross margin, 13.2% ROIC, net cash, FCF+. Named flow: revenue +35.3% YoY with EPS +499%, raised FY guidance, and the DOE bulk-power order f

  Unwind

  Gross margin sustained below 36% on price competition from MKS/Comet/Delta; semiconductor-segment revenue declining two consecutive quarters as the WFE recovery stalls; or design-out at a named WFE OEM. Scale-in gate: two consecutive quarters of gross margin >=40% with semi-segment growth intact. Kill: margin <36% or a named design loss.

  Catalyst

  Q3 earnings ~2026-11-02 (est): gross-margin trajectory toward the 40% scale-in gate.

  Scenarios · 12mo targets

  $320.00 base

  $400.00 bull — WFE recovery broadens, data-center power products scale, and gross margin crosses 40% — that earns the scale-in to full size.

  $200.00 bear — MKS/Comet/Delta price competition holds margin below 36% while the WFE upcycle slips a year; the -21.8% post-earnings derating extends.

• BEAMBeam Therapeutics Inc. Common Stockbiology

  hold1.02%43
  flow steadyconf

  Funder

  Growth-starter sleeve (§2). Origin platform for base editing (David Liu science, Broad-licensed): single-base correction without double-strand breaks — the most differentiated editing chemistry in the field — with ~$1.2B cash funding BEAM-302 in AATD into 2028, so it self-funds to its own proof point. Named flow: BEAM-302 dosing completed with a late-breaking oral selected; ARK adding; the in-vivo

  Unwind

  Named kill criteria, unchanged: hepatotoxicity or off-target signal in BEAM-302, a discounted equity raise, or the cash runway breaking before the AATD proof point. NOT a price move and NOT a score that entered below 60 by design.

  Catalyst

  BEAM-302 AATD late-breaking oral presentation data — the dated proof point the starter was sized for.

  Scenarios · 12mo targets

  $32.00 base

  $55.00 bull — BEAM-302 shows durable functional AAT correction with clean liver safety, validating base editing in vivo and inviting a pharma partnership on the platform.

  $12.00 bear — Hepatotoxicity or off-target editing appears in the AATD data, or the 2028 runway forces a discounted raise — either is a named kill, and I exit.

Warnings

• §2 cap re-applied after pillar tilt: MU
• §2 cap re-applied after pillar tilt: NVDA
• §2 cap re-applied after pillar tilt: AVGO
• §2 cap re-applied after pillar tilt: ANET
• §2 cap re-applied after pillar tilt: GEV
• §2 cap re-applied after pillar tilt: NEM
• §2 cap re-applied after pillar tilt: KTOS
• §2 cap re-applied after pillar tilt: HEI
• §2 cap re-applied after pillar tilt: LLY
• §2 cap re-applied after pillar tilt: VRTX
• §2 cap re-applied after pillar tilt: AEIS
• §2 cap re-applied after pillar tilt: BEAM
• hedge_floor: only 2/3 tail-risk names held (floor escalated: AI-load 30% > 30%) — open a gold/oil/diversified-commodity hedge (§5.3); uranium & power-gen do not count
• cash_reconcile: executed cash 19.23% below the 14.16% SGOV reserve — sold 2.42pp of over-deployment back to cash across AEIS (§5.4/§10.6)
• live_sizing: 3 ticket(s) re-expressed against the executed ledger; dropped add SCCO (live already at target) (§6.6)
• live_sizing: dropped add SCCO (Δ2.22pp vs intent) — live weight 9.33% already satisfies target 10.52% (§6.6; the ledger, not intent, is what a trade moves)

Conviction-lock actions

1. Turn 1ConvictionGuard (§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.