Stock-market dispatch — 2026-07-28

Headless run (Tue). Research output, not advice — brain never trades. Marks are live 07-28 intraday twelvedata quotes; 43/43 symbols priced, 0 429s (api.twelvedata.com reachable; SSNLF OTC excluded). Step 4d (weekly ledger) not due (Fri cadence). Supabase sync + push are performed by the wrapper, not this run.

Top of mind

A corroboration day, not a chain day — so the edge came from breadth discipline, not the AI tape. The two fresh primary sources (Sam Altman on ILTB, NASA Administrator Isaacman on Moonshots) were both dense but almost entirely reinforcement of the already-over-weight ai-infrastructure/space cluster (breadth 2a: ai-infra 47% ⚠ over), so — correctly — no net-new ai chain was drafted. Altman's two most tradeable disclosures both fold into existing pages: OpenAI's "Jalapeno" custom inference ASIC ("a huge competitive advantage… good at a specific workflow") is the demand-side principal validating inference-asic-wave-to-tsm-demand-broadening (fabbed at TSMC, likely a Broadcom design partner) — a stronger tell than a chip vendor's pitch; and his two-path oversupply framing (demand-ceiling vs scaling-wall; "uncapped demand implies a certain price") sharpens ai-roi-reckoning into a cost-curve question, corroborated by external data that the risk is timing, not a glut (~$725B 2026 capex +77%, DC vacancy record ~1%, but capex outrunning cloud revenue). The day's net-new went to thin verticals: a breadth-steered bucket scan strengthened two absent-cluster hypotheses (consumer trade-down→private label with fresh July hard data; freight capacity→TL/rail with the printed rate inflection finally showing), and prospect-chains drafted one financials/insurance chain (KNSL/PGR off the same trucking insurance-hardening). On the tape: NOW +6.9% (2nd straight gap-closing day post beat-and-raise), NVDA −5.0% / semis soft (MU −2.3% at $900).

Strongest-conviction buys

Ranked shortlist (4c). Prices 07-28 intraday, twelvedata.

⚠ LIN (high, 0.78) — $507.02 (−1.0%), through target → watchlist. LLY (med-high 0.60) — $1197.53 (+0.1%), at 52w hi → priced-for-perfection, watchlist.

Watchlist

Undervalued candidates (today)

• agentic-ai-seat-erosion-to-saas-rerate — NOW $105.56, still −48% off high, base $140–150; +6.9% today on top of +7.4% Friday — a two-day, ~14% move that is the market beginning to close the gap the thesis flagged, two weeks after the beat-and-raise.
• hbm-cowos-as-binding-bottleneck — MU $900.20, −2.3% into a broad semis selloff (NVDA −5.0%) on a thesis whose supply-constraint leg keeps getting first-party corroboration; ~6× fwd, −28% off the 52w high.

New chains to investigate (hypothesis-stage)

• commercial-auto-insurance-hardening-to-specialty-insurer-rerate — KNSL / PGR. Prospect-chains (2b): the trucking chain's un-named insurer beneficiary — the same commercial-auto premium hardening that's a carrier cost (driving capacity out) is the insurer's revenue. Thin cluster (financials/insurance), novel vs. the book's float-income insurance pages. To graduate: a KNSL/PGR commercial-lines rate + combined-ratio read confirming the hardening is still accelerating (not already priced — both near highs; insurer-exposure leg ⚠ unverified). → /explore-chain. priority low.

New theses (now active)

• None. Corroboration day — no new mechanism graduated (net-new was steered to thin-vertical hypotheses, per breadth 2a).

Updated theses (existing active)

• inference-asic-wave-to-tsm-demand-broadening — Jalapeno: sam-altman confirms OpenAI is building its own workflow-specific inference ASIC ("huge competitive advantage") — the biggest inference buyer validating the ASIC-beats-GPU-on-tokens/watt thesis with its own silicon (TSMC-fabbed, ~Broadcom partner). Reinforces the TSM-broadening leg + share-shift-at-inference vs NVDA; no step-status flip (conviction held low-med). as_of→07-28.
• ai-roi-reckoning — sharpened, both ways: Altman's two-path oversupply framing (demand-ceiling / scaling-wall; demand "uncapped… at a certain price") + external corroboration that the near-term risk is timing not glut ($725B capex sustained, vacancy ~1%, but Amazon FCF turning negative) → routes the actual trigger to mega-issuance-peak-to-ai-capex-derate (a financing break, not physical oversupply). as_of→07-28.
• ai-gas-demand-to-appalachian-producer-price-floor — EQT: EIA primary confirms a 2027 Henry Hub lift (~$3.50 2026 → ~$4.60 2027, +33%) but attributes it to LNG-export ramp, not data centers near-term (DC leg end-of-decade) — an attribution risk to the causal claim (2026-27 tailwind may be coincident, not caused). Basis leg still confirmed; floor still forward. EXE Q2 actuals post-close 07-28, call 07-29. Held medium. as_of→07-28.

Contradicted / weakened

• No active thesis falsified. 0 new calibrations. 4e gate: UUUU −10.4% sector re-fires (med-high) but is the same vehicle-selection contradiction calibrated 07-27 (NXE +6.2% / DNN +7.8% legs of the same uranium mechanism beat sector; UUUU improved from −12.6%) → no re-file. PWR (−18.1%) and TJX (−11.3%) are both medium tier → no gate.

Open questions worth a human's eye

• Is NOW's two-day, ~14% pop the durable start of the seat-erosion re-rate the thesis predicted, or a short-covering bounce in a broad SaaS unwind? The falsifier (NRR/ACV deceleration) didn't fire at Q2 — Q3 is the test.
• The gas chain's LNG-vs-DC attribution problem. If EIA is right that the 2027 lift is LNG-led, the AI-gas-demand thesis is riding a coincident tailwind near-term; the DC-specific causal leg only bites end-of-decade. Does EQT's realized-price disclosure (EXE call 07-29 as the read-through) ever isolate a data-center-driven floor?
• Does the developer-tier uranium vehicle (UUUU) systematically under-deliver theme torque vs. NXE/DNN, or is this one drawdown regime? (Third day the same split holds.)

What I looked at

• Run status: headless; steps 0.5–7 in-turn; price fetch paced 8/min — 43/43 priced, 0 429s. Step 4d (weekly ledger) not due (Fri cadence; last mark 07-24). Full per-concept bull/base/bear narrative carried forward (headless budget); load-bearing marks live in the signal feed + tables above.
• Breadth (2a): ai-infrastructure 47% ⚠ over (43 distinct chains) → net-new steered to thin verticals. Macro buckets rotated to #10 consumer + #11 transport/industrials ex-AI.
• Ingestion veins (0.5–0.7): podcast 2 (Sam Altman/ILTB, Isaacman/Moonshots — diarized AssemblyAI, ~$0.40); earnings 0 (no tracked name reported un-ingested — nearest are PWR 07-30, CCJ/LIN 07-31, LLY 08-05); feeds 0 (SemiAnalysis latest already ingested; Fabricated Knowledge new post fully paywalled; Construction Physics off-topic).
• Steps 1–2b: 3 autoresearch (EXE/gas producer-floor gap → floor forward, LNG-led; AI compute-oversupply corroboration; thin-vertical bucket scan consumer+freight). prospect-chains: 1 auto-drafted (commercial-auto-insurance-hardening-to-specialty-insurer-rerate).
• Sources: 5 promoted + 5 ingested → 3 created (2 person entities sam-altman, jared-isaacman + 1 question from 2b); ~6 updated (ai-roi-reckoning, inference-asic-wave, ai-gas-demand mechanisms; consumer-trade-down + trucking questions; broadcom link); 0 new mechanisms (corroboration day). 2 zero-novelty thesis-grade sources surfaced (both podcasts — over-weight-cluster reinforcement).
• Live feedback (4e): 23 scored rows; 3 flags < −10% sector (PWR −18.1% medium, TJX −11.3% medium → no gate; UUUU −10.4% medium-high → gate re-fires but already calibrated 07-27). 0 new auto-calibrations. Best: DNN +7.8%, NXE +6.2%, STVN +5.4% sector.
• Signal feed (7): full active/armed set — 57 rows, validated 57/57 clean (dry-run; not writing to Supabase — sync + push deferred to wrapper). Prices re-marked 07-28 (56/57 priced; SSNLF OTC stale). as_of re-dated only for the EQT gas row (the sole emitted feed mechanism materially changed today; the Jalapeno→inference-asic-wave and ai-roi updates land on pages that aren't emitted feed rows). All other rows carried as_of forward. No net-new rows (no mechanism graduated; the new commercial-auto-insurance chain is a hypothesis, not active/armed).