Stock-market dispatch — 2026-08-13

Headless run (Thu, pre-open). Research output, not advice — brain never trades. Marks are 2026-08-12 closes (Wednesday, the latest completed session). 107/108 symbols priced, 3 transient 429s recovered, 1 persistent (ALMTF). Weekly ledger (4d) not due (last 08-07; next 08-14). Supabase upsert + push are performed by the wrapper, not this run.

Top of mind

A chain that had been held at hypothesis for 64 days graduated today — and the same source handed us a second, spine-distinct chain the wiki had never seen.

trucking-regulatory-capacity-removal-to-tl-carrier-rate-recovery has been open since 2026-06-10 and has twice refused to graduate on bad evidence: on 07-16 it declined to cite the record LMI price print because the LMI itself attributed it to the Hormuz fuel shock, and it discounted Werner's 250–400k driver figure as an interested party's number. The gate it set was specific: a carrier Q2 call confirming both capacity exit and rate realization, not guidance. Schneider's Q2 transcript clears it outright. CEO jim-filter: "We would now categorize the market as driver-constrained" — naming the enforcement stack by name — plus "roughly half of the non-compliant capacity is left" to exit through next year, network revenue per truck per week +16% YoY, double-digit renewals, and a raised EPS guide ($0.90–1.10 from $0.70–1.00) on cut capex ($350–400M from $400–450M). A carrier raising earnings while shrinking capex is not adding capacity into its own tightness. Alongside it, KNX: truckload revenue +2.8% while truckload operating income +96.3% — operating leverage on price, volume flat. That is the supply-driven signature the page predicted in June and could not evidence until now. Filed as driver-supply-removal-to-truckload-contract-rate-inflection (5 steps, medium-high).

The second find is Montgomery v. Caribe Transport II — Supreme Court, 9-0, 2026-05-14, Barrett — holding that freight brokers can be sued for negligently selecting unsafe carriers and that the FAAAA does not preempt those claims. c-h-robinson brokered the shipment in the case. This is a forcing function with a property almost nothing else in this book has: a unanimous holding from the court of last resort has no appeal, no agency discretion, and no election that reverses it. Filed separately as broker-negligent-hiring-liability-to-freight-capacity-bifurcation, because it removes carriers from the brokered market while the other chain removes drivers from the licensed pool — ⚠ they compound, but a rate move must not be booked to both.

And the honest counterweight, which is why neither is ranked #1: the entire freight complex ripped on 08-12 — KNX +4.1%, JBHT +4.0%, WERN +3.9%, LSTR +4.6%, SNDR +2.4%. The brokers rallied too, which cuts directly against the broker-derate leg. The episode's own show notes link Bloomberg's "Trucking Stocks Fall on Legal Risk in Worst Month Since Tariffs." The market has already reacted to Montgomery — by selling, then buying back. "Nobody has noticed" is not available here.

Strongest-conviction buys

Ranked shortlist (4c). Prices = 2026-08-12 closes, twelvedata. Fundamentals carried forward except where dated today.

⚠ The new chain is ranked 5th, not 1st, and the reason is valuation not evidence. Chain strength is now among the best in the book (first-party attribution, disinterested corroboration, realized numbers, a raised guide). But KNX is only −13.9% off its high and rose 4.1% on the mark date; SNDR — the name that supplied the evidence — is just −9.0% off its high and is therefore not the expression. WERN at −20.3% off high is the cheaper leg, but Werner's CEO is the source of the 250–400k figure this chain explicitly discounts, so the cheapest name rests on the weakest number. That tension is unresolved and is the honest reason this sits at #5.

Deliberately not ranked, for the second consecutive day: VLO / PSX / MPC. Yesterday they closed at 52-week highs; today VLO −0.2% off high, PSX −0.4%, MPC −0.0% — they printed new highs. The chain got materially stronger today (see below) and there is still no discount. The ranking refusing to move on a strengthening thesis with no valuation gap is the ranking working.

⚠ BWXT (0.55) ranks just outside at $172.56 (+1.7%, −28.6% off hi, ~9.8% above its 52-week low) — the deepest discount in the top tier after MP, on a dated end-2026 margin inflection.

Watchlist

Undervalued candidates (today)

• drone-warfare-demand-to-mp-hree-rerate — MP −46.0% off high, the deepest discount in the book, and live sector-excess +6.5%. Discount and realized selection edge still point the same way.
• agentic-ai-seat-erosion-to-saas-rerate — NOW −35.8% off high, FCF-positive, live sector-excess +3.0%.
• legacy-priced-backlog-rolloff-to-bwxt-margin-inflection — BWXT −28.6% off high, ~9.8% above its 52-week low, with a dated margin inflection and a raised guide.
• ⚠ driver-supply-removal-to-truckload-contract-rate-inflection — WERN −20.3% off high is the only genuinely discounted leg of the newly-graduated chain. Flagged rather than promoted: it is the name whose CEO supplied the figure the chain discounts.

New chains to investigate (hypothesis-stage)

No new hypothesis pages drafted — third consecutive day, and again deliberately. Today's generative budget went into two mechanisms and two hypothesis resolutions instead. prospect-chains mining found the two best candidates were graduation moves on existing pages, not net-new spines, which is the correct outcome on a wiki with 78 mechanisms and 77 open hypotheses.

• tungsten-export-controls-to-allied-primary-producer-rerate — the gating item is answered, and it did NOT graduate. This page named offtake structure as "the killer": if Sangdong's concentrate were committed at a fixed or capped price, "the entire thesis fails while every upstream step stays true." The 2026-07-14 GTP amendment (company release + SEC Form 6-K) is the opposite: floor US$183/MTU with NO ceiling, term 15→21 years, volume +40% to 4.41M MTU, ~90% of Phase I, $490M/yr at current APT. Implied realization ~$2,333/MTU — priced at spot, not at the floor. Beneficiary leg re-tagged ⚠ unverified → partial. Held anyway, because the other half of the same watch-item — a realized $/mtu from a reported production quarter — does not exist yet. The gate worked symmetrically, which is the harder test. Also: a new dated forcing function 14 days out — from 2026-08-27 US sellers must reserve all tungsten waste/scrap and battery black mass for domestic buyers. ⚠ Deliberately not wired as an Almonty step: it acts on secondary supply, Almonty is a primary producer, and the connecting link is unsourced.

New theses (now active)

• driver-supply-removal-to-truckload-contract-rate-inflection (mechanism, 5 steps, medium-high) — graduated from a 64-day-old hypothesis. See Top of mind.
• broker-negligent-hiring-liability-to-freight-capacity-bifurcation (mechanism, 5 steps, medium) — Montgomery, 9-0. ⚠ No signal emitted: steps 3–5 are partial, the market-structure leg rests on one hedged practitioner source, and the sector has already traded on the news in both directions. The long-asset-based / short-broker pair is the right shape and the entry is not established.
• 6 new entities: schneider-national, c-h-robinson, plus person entities jim-filter, darrell-campbell, reed-loustalot, luisa-palacios.

Updated theses (existing active)

• venezuelan-heavy-sour-return-to-usgc-coker-differential-capture — step 1 moves to confirmed on a genuinely disinterested primary. luisa-palacios — Columbia CGEP, and former chairwoman of Citgo — states "Venezuela is now the second most important supplier of oil exports to the US after Canada." Production ~1.2M bpd (June, OPEC); ~800k bpd redirected from opaque Chinese-teapot channels to the US and India; four-week export averages above 2018 pre-sanctions levels. Until today every confirmation came from refiners describing their own margin. She also sharpens the falsifier past anything on the page: "It is completely predicated on a US sanctions policy. So if there's any change in that sanctions policy, this story ends." The chain's horizon is policy-tenured, not structural.
• The flagged "net heavy-sour" weakness narrows. Maya fell 71k→22k bpd (~49k) against a Venezuelan increment sized ~200k bpd — roughly 4:1 in the Venezuelan direction, so the balance is up, not offset. ⚠ Search-extraction only; the 200k is modelled and conditional against an observed 49k. No step re-tagged on it. A sharper question replaces it: USGC coking capacity, not barrel availability, may be the binding ceiling — which would narrow the beneficiary set to refiners with spare coker utilization rather than the whole basket.

Contradicted / weakened

• ⚠ /calibrate #40 — a new pattern tag, recency/lens-overfit. I made "the offtake is probably capped" the dominant risk on Almonty — literally "the killer" — with zero evidence about Almonty's contract. The prior came entirely from a lens I had just learned elsewhere: lta-contract-structure-as-price-insulation, made vivid on 08-12 by legacy-priced-backlog-rolloff-to-bwxt-margin-inflection. The truth was the opposite (floor, no ceiling). The asymmetry is the tell: that lens is two-sided — a structure that strands one supplier can equally let another capture — and I carried only the direction I'd just been burned by. Distinct from #38/#39, which were overconfidence/favored-narrative: this wasn't wanting a conclusion, it was over-applying a correct, freshly-learned rule to a case with no evidence. Rule recorded: a risk named with zero company-specific evidence should be labelled a transferred prior and gate research, not conviction. It was one search away for five days.
• ⚠ Correction to this project's own record. The 2026-07-16 entry called the circulating "97% of ~200,000 non-domiciled CDL holders" figure "an unsourced aggregation… could not be sourced to any primary." Its origin is now located: J.B. Hunt's own published supply model. So it is a named, interested-party estimate rather than an unattributable one — J.B. Hunt's economics improve if truckload tightens. Practically different: an unattributable number gets discarded; an interested model gets discounted and cross-checked. Step 2 (scale) stays partial.
• 4e gate: 23 live rows, 2 flags < −10% — TJX −11.9% (⚠ worsened from −11.8%), PWR −11.4% (improved from −12.5%). CAT −9.2% has cleared the threshold (from −10.6%). 0 automatic calibrations, a pass not a skip — the gate covers high/medium-high and both breaches are medium or below. ⚠ Fifth consecutive dispatch in which every breach sits in the medium-or-below tier and the high/med-high tier has zero. CALIBRATION.md's standing worry is that conviction labels have been a poor guide; on live evidence the tiering keeps sorting correctly. Five is a streak, not proof. Best live: DNN +9.7%, NXE +8.5%, AMZN +8.1%, MP(drone) +6.5%. Worst cluster: energy-oil −6.1%, 8th week — again on a day the refiners printed highs, because the live energy-oil rows are CCJ/CF/NTR, not refiners. Best cluster: critical-minerals +3.9%.

Open questions worth a human's eye

• ⚠ The brokers rallied with the carriers. CHRW +1.1% and LSTR +4.6% on 08-12, alongside KNX +4.1% and WERN +3.9%. If Montgomery genuinely impairs the brokerage model, these should diverge. Either the market is not pricing the ruling, or the chain's step 4 is wrong. This is the cleanest near-term test of the new mechanism and it is currently pointing the wrong way.
• Does Montgomery survive Congress? The Court decided a question of statutory interpretation, so a FAAAA amendment reverses it. Brokerage is a well-funded lobby. Highest-probability reversal path, no date.
• Is USGC coking capacity the real ceiling on the Venezuelan chain, rather than barrel availability? Would narrow VLO/PSX to whoever has spare coker utilization.
• ALMTF cannot be priced on the Twelve Data free plan — and tungsten-export-controls-to-allied-primary-producer-rerate now names ALM as its tradeable. Needs a substitute line or an explicit unpriced marker before that chain could ever carry a signal row.
• Why do PWR and CAT still move together when the corrected mechanism says they shouldn't? PWR +1.1%, CAT +1.5% today; seventh consecutive mark of co-movement. Still unexplained.
• The hypothetical inception re-mark is deferred a seventh week — blocked by three standing data defects (KLAC split factor, CBRS entry_ref, SPCX SPY backfill). Escalating as a human call for the sixth dispatch running.

What I looked at

• Run status: headless; steps 0.5–7 in-turn. Price fetch paced 8/min across 16 batches — 107/108 priced, all 08-12 closes. Two mechanical defects found and fixed in-run, both mine and both silent: (1) a missing trailing newline fused XOP+SNDR into a bogus symbol, dropping both from the request set; (2) curl >> without newlines fused responses into unparseable lines, making 10 fetched symbols look missing — fixed by re-parsing the raw buffer with a streaming decoder rather than re-spending credits. Both caught by the priced-vs-requested reconciliation, which is now earning its keep on the third consecutive day.
• Breadth (2a): ai-infrastructure 45% ⚠ over (down from 47%; 47 distinct chains, 63 live signals, 8 clusters) → net-new steered to thin verticals. Delivered: both new mechanisms are transport/regulatory, an absent target vertical; the bucket-#2 scan landed in critical-minerals (6%). Zero net-new ai-infrastructure chains opened.
• Macro buckets: #2 energy & critical minerals, #9 financials & the rate regime. ⚠ #9 produced nothing filable — generic 2026-outlook material (private credit "true" default rate ~5% vs headline <2%; bank NII outlook) with no dated forcing function and no named beneficiary, so no clipping was written. Negative result, recorded.
• Ingestion veins (0.5–0.7): Podcast 2 — both free via publisher RSS transcripts, no Whisper time, no provider spend, and both aimed at chains yesterday's dispatch named as next actions. Earnings 0 (Q2 watchlist cycle genuinely complete; a KNX ad-hoc pull was attempted and abandoned honestly — no full free transcript exists this quarter). Feeds 0 (three feeds dry; the fourth's only new post is fully paywalled, a 68-char RSS stub — filed nothing rather than a stub).
• ⚠ Podcast diarization unfunded, 7th day — not re-probed, because six consecutive identical failures make a seventh probe worthless. Budget is $1.05 of $50; the accounts are unfunded, not capped. Both of today's transcripts came back speakers_source: none; luisa-palacios attribution is solid only because the Columbia transcript carries inline Name (mm:ss): labels in its body. 23+ episodes still backed up. Action for Paul: top up AssemblyAI, or fund xAI and switch diarize_provider: grok.
• Sources: 5 promoted + 5 ingested → 2 new mechanisms, 6 entities created (2 companies, 4 people), 1 hypothesis graduated, 1 hypothesis substantially resolved, 3 mechanisms/concepts updated, 1 calibration (#40, new pattern tag).
• prospect-chains (2b): 0 auto-drafted. Both top candidates resolved to graduation moves on existing pages rather than net-new spines — the healthy outcome at 78 mechanisms / 77 open hypotheses.
• Not done, and named: EARNINGS.md still has no HWM, ATI or BWXT — fourth consecutive day; today adds SNDR, KNX, CHRW as newly-cited names worth watchlisting. And a new one: FEEDS.md's Active list is 3-of-4 semis/AI-infra, i.e. the ingestion vein is itself feeding the 45%-over cluster, while four Candidate feeds map directly to the absent verticals (Net Interest, The Overshoot, Ground Truths, The Diff). Promoting one or two is the cheapest available breadth fix. Both files are user-curated — flagged, not edited.
• Signal feed (7): full active/armed set re-marked to 08-12 closes; as_of re-dated only for rows whose research moved.