Stock-market dispatch — 2026-08-07

Headless run (Fri). Research output, not advice — brain never trades. Marks are 2026-08-06 closes, and that label is a correction (see Top of mind). 70/70 symbols priced, 0 429s. Weekly ledger (4d) due and run. Supabase upsert + push are performed by the wrapper, not this run. Podcast vein dark for a 3rd day.

Top of mind

A falsifier this project wrote itself, months ago, fired exactly on schedule — and the honest answer is that half the chain was right and the tradeable half was wrong. spacex-ipo-comp-anchor-to-space-peer-fade has been armed since 2026-06-17 on one dated catalyst: the first post-IPO lockup expiry. It fired 2026-08-06 (911.5M of ~13.6B shares released, float more than doubled). The anchor leg was vindicated emphatically — SPCX closed $114.92, an implied ~$1.56T against ~$2.64T at debut, a ~41% de-rate and below its own $1.77T IPO valuation. And the transmission leg — the part that names a trade — was refuted just as emphatically: in the same week VOYG +54%, RDW +38%, ASTS +29%, RKLB +28% while SPCX fell. The long-SPCX/short-peers pair loses on both legs. The drivers were idiosyncratic and dated: RKLB's $397M Space Force award, ASTS's BlueBirds 11/12/13. Contracted government backlog beats comp-multiple gravity. Status active → weakened (not closed — Step 1 is now better evidenced), /calibrate logged under a new pattern, correct-mechanism-wrong-transmission. The position was never opened, so there is no ledger damage — only a belief corrected. One structural blind spot named: the ASTS launch flew on a Falcon 9, so the "peer" is a SpaceX customer and I had modelled a supplier relationship as purely competitive.

Second finding, and it is a process defect: every mark this routine has published pre-open has been one session stale. Yesterday's dispatch reported "SPCX −13.6% today"; that was 08-05, and SPCX actually rose +6.14% on 08-06. Spot-checks confirm it is systematic — LLY $1,169.86, PWR $682.99 and MU $893.19 in yesterday's "live 08-06 intraday" table are each exactly the 08-05 close. Cause is structural: the routine runs ~05:00 ET, 4.5 hours before the open, so Twelve Data /quote returns the last completed session. No conviction score changes — it is a labelling defect — but it mis-stated the largest single move in the book by a full session. Fixed from today. In mitigation it made yesterday's PWR calibration conservative: PWR fell a further −2.22% on 08-06.

Net: 8 sources promoted + ingested (3 earnings, 2 analyst feeds, 3 autoresearch), 2 net-new hypotheses, 1 new entity (cf-industries — a tracked ticker that had no page), 1 new person entity, 0 new mechanisms (6 touched, 1 falsified, 1 conviction raised, 1 falsifier-set rewritten), 1 calibration.

Strongest-conviction buys

Ranked shortlist (4c). Prices = 2026-08-06 closes, twelvedata.

⚠ MP is ranked #2 with its live paper return at −28.0%, and that needs saying plainly. Its sector-excess is +0.1% — the chain has tracked a de-rated theme, not underperformed it — and today's call confirmed the exact milestone the chain has waited on since May. But a −52.6% drawdown is not a discount until the theme stops de-rating. Near-misses: LIN (0.78) $490.11 — high conviction, priced at −10.6% off hi. TSM (0.70) $418.20 (+1.0%) — −12.7% off hi. STVN (0.58) $20.29 (+0.9%, −27.5% off hi, +2.6% live sector-excess) — a genuine #6. LLY (0.60) $1,191.94 (+1.9%) — only −4.6% off highs.

Watchlist

Undervalued candidates (today)

• drone-warfare-demand-to-mp-hree-rerate — MP −52.6% off high, the deepest discount in the book, on a chain whose Step 2 was confirmed first-party today and whose sector-excess is flat (+0.1%) rather than negative.
• agentic-ai-seat-erosion-to-saas-rerate — NOW −39.7% off high, FCF-positive, and the book's best SaaS live mark (+2.6%).
• pjm-capacity-prices-to-nuclear-premium — CEG −36.7% off high while guidance was raised and the Crane 2031-slip risk this wiki tracked resolved to H2 2027.

New chains to investigate (hypothesis-stage)

• naic-rating-scrutiny-to-pe-insurer-flywheel-derate — APO / KKR. Insurers hold ~$419B of privately-letter-rated credit; since 2026-01-01 the NAIC can challenge any rating differing by three or more notches, with new RBC charges for structured credit and mandatory PIK-interest disclosure from the 2026 reporting year. Athene (Apollo) and Global Atlantic (KKR) both run >15% allocations. Dated catalyst inside the week: American Academy of Actuaries RBC analysis, 2026-08-11/14. To graduate: evidence a capital charge actually changes an allocation. Deliberately framed as a falsifier candidate for insurance-illiquidity-rotation-to-private-credit-managers, which runs the other way.
• siting-gate-plus-stranded-capacity-to-demand-side-beneficiary-rotation — CEG / grid-flexibility long vs PWR, CAT, EPC short. A cross-source chain no source states: state siting gates defer construction (EO 62, end-condition unbounded) while CEG's own management says it will serve datacenter load from existing stranded generation + batteries + demand response. It would explain a live anomaly — PWR −11.5% and CAT −8.1% sector-excess simultaneously while CEG sits at −1.9%. ⚠ Consistent with, not evidence of; three positions is not a sample.

New theses (now active)

• None net-new as mechanisms. 0 new mechanisms is the correct outcome — see "What I looked at."

Updated theses (existing active)

• drone-warfare-demand-to-mp-hree-rerate — conviction medium → medium-high, as_of→08-07. Dy/Tb circuit "actively commissioning," on track to ship Mountain Pass→Independence this year (retires the Step-2 falsifier). First HREE pricing evidence: a nine-figure multi-year gadolinium offtake to a US aerospace/defense manufacturer where ryan-corbett confirms "we've locked in economics on this contract" — non-spot. Project Swarm signs subscription agreements with allied drone makers to aggregate magnet demand. ⚠ Held honest: it's gadolinium, not Tb/Dy, and Swarm's volumes/bindingness are undisclosed — partial, not confirmed.
• pjm-capacity-prices-to-nuclear-premium — as_of→08-07. Capacity-price→earnings transmission now realized: EPS $2.55 (+$0.64), FY raised, ~920 MW of PPAs at 18.5-year average duration. ⚠ The Walmart PPA is "the first transaction of its kind for a major retailer" — the buyer set is widening past hyperscalers, which makes the premium less AI-cycle-dependent and its ai-infrastructure cluster tag less clean.
• insurance-illiquidity-rotation-to-private-credit-managers — regulatory ceiling added; no conviction change (the binding question is unanswered).
• orbital-inference-efficiency-to-tpu-advantage — a partial written-analysis leg: SemiAnalysis reads Google's DeepMind exodus as freeing TPU capacity for external sale (>$250B potential GCP RPO). The inversion is the value — lab consolidation raises merchant compute. Chain not upgraded (paywalled, five days old).
• state-datacenter-siting-moratorium-risk / legislative-divergence-base-rate — both updated; see below.

Contradicted / weakened

• spacex-ipo-comp-anchor-to-space-peer-fade — active → weakened. Falsifier fired. See Top of mind. /calibrate logged.
• ⚠ hormuz-nitrogen-supply-shock-to-cf-risk-premium — CF's own CEO contradicts the attribution while confirming the conclusion. christopher-d-bohn: the market's Middle-East lens "misses a fundamental structural shift… Higher global capital costs have structurally raised the incentive price required for new global nitrogen capacity… This is before we factor in any geopolitical premium," and "our low cost, low risk North American asset base and not geopolitical risk is the foundation of our profitability." Step 4 is strengthened and quantified (mid-cycle EBITDA → ~$2.9B, path to $3.3B, on a $410/short ton mid-cycle price); Step 1's routing through Hormuz is contested. Falsifier set rewritten: "global capital costs deflate" is now load-bearing; "Hormuz reopens" is demoted — freight is $70 vs $35 a year ago and "probably not" snapping back. Independent corroboration for energy-shock-2026-vs-2022's structural-muting read, from the fertilizer side. ⚠ Q2 EPS $4.73 missed ~$5.79 — a raised mid-cycle guide in a missed quarter carries an incentive.
• ⚠ Drift caught: pwr-transformer-moat-to-eps-doubling's page still read medium-high a day after the 08-06 calibration lowered its signal to 0.55/medium. Page and feed now agree.
• 4e gate: 23 live rows; 1 flag <−10% — PWR −11.5% (widened from −10.7%). UUUU healed −12.5%→−9.3%, TJX healed −11.6%→−8.2%. 0 new calibrations from the gate, and that is a pass not a skip — PWR was calibrated to medium on 08-06, so it no longer arms the high/med-high gate, and the widening is that update's predicted consequence. Best live: AMZN +11.8%, NXE +10.6%, DNN +9.1%. Worst cluster: energy-oil −4.7% (4th week).

Open questions worth a human's eye

• Does a NY-style pause destroy load or merely relocate it to PJM/ERCOT/MISO? Now the highest-value open question in the book — it is logically prior to today's new siting-rotation hypothesis, and if load relocates, that chain is geographic, not directional.
• Do Project Swarm subscriptions bind? MP's demand leg graduates to confirmed only if they carry volume commitments. Undisclosed.
• Is PWR's lag multiple-compression or demand-rotation? The two explanations are observationally similar right now. The next PWR print is the clean test — guidance citing project deferral would separate them.
• The hypothetical inception re-mark has been deferred four consecutive weeks. The ledger's only inception-anchored measure has been dark for a month, blocked by three standing data defects (KLAC split factor, CBRS entry_ref, SPCX SPY backfill). Escalating as a human call.
• What are Almonty's offtake terms? Carried from 08-06 — the tungsten hypothesis is one disclosure from confirmed or dead. Not researched today (budget went to the three above); deliberately flagged rather than quietly dropped.

What I looked at

• Run status: headless; steps 0.5–7 in-turn. Price fetch paced 8/min across 9 batches — 70/70 priced, 0 429s, all 08-06 closes (SSNLF 08-05, OTC). Weekly ledger (4d) due and run. Per-concept bull/base/bear narratives carried forward (headless budget); load-bearing marks in the feed + tables.
• Breadth (2a): ai-infrastructure 47% ⚠ over (45 distinct chains, 61 signals, 7 clusters) → net-new steered to thin verticals. Delivered: financials (the NAIC hypothesis — a near-absent vertical) and critical-minerals + materials (MP, CF). The one ai-infra hypothesis opened is a rotation-away chain, not another beneficiary; the SemiAnalysis ingest was filed as evidence, not as a new name.
• Ingestion veins (0.5–0.7): podcast 0 — dark a 3rd day, both providers down. AssemblyAI balance still negative; xAI now returns a different error — "team doesn't have any credits or licenses", i.e. a billing state, not the dead key recorded on 08-06 (correction filed). Helper refuses before download (~4s/episode), so the failed run was cheap. 4 high-value episodes deferred, led by All-In / Saronic (autonomous warships — defence), which is directly adjacent to today's Project Swarm finding. Action for Paul: top up AssemblyAI or add xAI team credits — either restores it. Earnings 3 (MP, CEG, CF — all reported 08-06). Feeds 2 (SemiAnalysis ⚠paywalled-partial, Construction Physics).
• ⚠ PODCASTS.md filter gap, 2nd day running: Odd Lots 12 new, 1 kept, dropping "Why Private Credit Got Entangled With Insurance", Fed independence, Apple, Iranian economy, yen intervention. For the second consecutive day the routed-to-autoresearch replacement produced a headline finding — today's NAIC hypothesis is the dropped private-credit episode's topic, reached the long way. Suggest adding private credit, insurance, Fed, data center, defense, nuclear — or dropping the Odd Lots filter entirely.
• Sources: 8 promoted + 8 ingested → 1 entity + 1 person-entity + 2 hypotheses created; 0 new mechanisms, 6 touched (1 falsified, 1 conviction ↑, 1 falsifier-set rewritten, 1 regulatory-ceiling added, 1 partial-evidence leg, 1 drift-fixed); 2 concepts updated; 1 contradiction recorded; 1 calibration.
• Why 0 new mechanisms is right, per source: MP/CEG/CF map to existing spines; NAIC has open beneficiary+tradeable legs → correctly a hypothesis; SpaceX is an anti-chain finding; NY GEIS is a base rate feeding a risk register; Construction Physics is a deflationary policy read (most of the Act does nothing); SemiAnalysis is a partial leg.
• prospect-chains (2b): 1 cross-source auto-draft (the siting-gate rotation, connecting today's CEG call to yesterday's EO 62 ingest and a live performance anomaly). 1 candidate reported but not filed — CF's capital-cost argument generalizes to every capital-intensive commodity (incumbents with built, depreciated assets earn a structural rent as the incentive price for new capacity rises). Real, but concept-level and names no clean tradeable beyond names already held; noted here rather than padding the hypothesis cap.
• Signal feed (7): full active/armed set re-marked to 08-06 closes; as_of re-dated only for rows whose research moved — MP ×2, CEG, CF, SpaceX comp-anchor ×2, insurance-illiquidity, TPU-advantage. The rest carried forward.