Stock-market dispatch — 2026-08-14

Headless run (Fri, pre-open). Research output, not advice — brain never trades. Marks are 2026-08-13 closes (Thursday, the latest completed session). 118/123 symbols priced, 0 429s, 5 persistent plan-tier failures. Weekly ledger (4d) due and run. Supabase upsert + push are performed by the wrapper, not this run.

Top of mind

The most important thing found today is a defect in this project's own scorecard, and it was hiding in plain sight for six dispatches.

I have reported, six times, that "every live sector-excess breach sits in the medium-or-below tier and the high/med-high tier has zero" — and offered it each time as evidence the conviction labels are calibrated. On 2026-08-13 I even hedged it correctly: "five is a streak, not proof." The streak was partly an artifact. iran-fuel-shock-consumer-bifurcation — which holds three live signal rows including TJX, the worst sector-excess position in the entire book (−12.8%, worsened on three consecutive marks) — has carried empty status: and conviction: frontmatter since 2026-06-08. The step-4e gate fires on high/medium-high. A mechanism with no conviction value at all can never match that predicate. The worst position in the book was structurally invisible to the check built to catch exactly it.

The part that stings: I flagged that empty frontmatter in log on both 08-12 and 08-13, and in the same dispatches cited the clean tier record as evidence the labels work. The disconfirming fact and the reassuring conclusion sat in the same document, twice. Filed as /calibrate #41 under a new pattern tag, process/unverified-denominator — distinct from #38/#39 (favored-narrative, wanting a conclusion) and #40 (lens-overfit, over-applying a fresh rule). Here the reasoning was fine and the conclusion conditionally true; the defect was validating a claim over a filtered population without checking what the filter excluded. Rule recorded: before citing "N breaches, all in the low tier," verify every row was eligible to be counted. A gate that silently skips a record produces a clean report, and a clean report reads as a pass.

Reading that page closely then produced a second finding: its beneficiary step is probably wrong too. It routes a fuel-price squeeze into off-price apparel (TJX/ROST). Today's consumer research says the trade-down is real but landing in grocery own-brand, driven by quality perception rather than price — ~85% of consumers now rate private label at or above national brands — and continuing "even as inflation has moderated." Not the fuel shock at all. Same error class as the Howmet correction: right forcing function, wrong tier. Conviction → low. Deliberately not closed — nobody measured off-price apparel traffic directly, and closing on inference would mirror the error being corrected.

Strongest-conviction buys

Ranked shortlist (4c). Prices = 2026-08-13 closes, twelvedata. Fundamentals carried forward except where dated today.

⚠ KNX drops out of the top five — on valuation, and the move is the reason. KNX ranked #5 yesterday at −13.9% off high; it rose another 2.79% to −11.5% off high. SNDR is now only −7.5% off its high and JBHT −5.8% — the freight complex has rallied for a third consecutive session (KNX +2.8%, JBHT +2.3%, WERN +1.7%, SNDR +1.6%, LSTR +1.5%, CHRW +1.8%). The chain's evidence got stronger today (see below) and its cheapest expression got more expensive. WERN takes the slot only because it is the last leg with a discount — and it is the leg resting on the weakest number. That tension is unresolved and is stated for the second day running.

Deliberately not ranked, for the third consecutive day: VLO / PSX / MPC. All three printed new 52-week highs again — VLO $342.92 (+3.85%, −0.2% off hi), PSX $232.61 (+3.12%, −0.1%), MPC $356.37 (+2.33%, −0.2%). Today the chain's step 3 moved partial → confirmed on a genuinely disinterested quantification (below). A thesis strengthening for the third straight day, with the tape at all-time highs and zero valuation gap, still does not rank. The ranking refusing to chase is the ranking working.

⚠ BWXT (0.55) at $170.35 (−1.3%, −29.6% off hi, only 8% above its 52-week low) remains the deepest discount in the tier after MP.

Watchlist

Undervalued candidates (today)

• drone-warfare-demand-to-mp-hree-rerate — MP −44.5% off high, deepest discount in the book, and live sector-excess +10.0%, now the best live position. Discount and realized selection edge point the same way, harder than ever.
• agentic-ai-seat-erosion-to-saas-rerate — NOW −34.7% off high, FCF-positive, live sector-excess +1.8%.
• legacy-priced-backlog-rolloff-to-bwxt-margin-inflection — BWXT −29.6% off high and just 8% above its 52-week low, on a dated end-2026 margin inflection.
• ⚠ driver-supply-removal-to-truckload-contract-rate-inflection — WERN −19.0% is now the only discounted leg; KNX has rallied to −11.5% and SNDR to −7.5%. The window on this chain's cheap expression is closing while the evidence improves.

New chains to investigate (hypothesis-stage)

• driver-not-equipment-scarcity-to-class-8-oem-derate ⭐ new — PCAR / CMI / WNC. The truckload rate cycle is turning on a driver shortage, not an equipment shortage — so the consensus reflex ("freight rates up → buy the truck cycle") may invert. Schneider raised EPS guidance while cutting capex; KNX posted +96.3% TL operating income on +2.8% revenue. A truck without a qualified driver earns nothing, so the carrier's optimal response is fewer, better-utilized trucks at higher prices — and Class-8 orders can fall into a rate upcycle. What it would take to graduate: ACT/FTR Class-8 net orders falling while TL contract rates rise, with the 2027 emissions pre-buy ruled out as a confound. ⚠ Step 5 is entirely un-cited — no Class-8 order data exists anywhere in this wiki. It is also a falsification instrument for the parent chain: a long-carrier/short-OEM pair only pays if the tightening really is driver-driven.
• ⚠ No hypothesis page drafted for the healthcare find, deliberately. sole-source-concentration-in-generic-sterile-injectables was filed as a concept, not a hypothesis, and no ticker was named as a beneficiary — because the forcing function is quantified and the beneficiary is entirely uncited. Naming GEHC/BAX/ICUI from category structure alone is precisely the transferred prior error logged as /calibrate #40 one day ago. They are listed as research targets and the page says so.

New theses (now active)

• sole-source-concentration-in-generic-sterile-injectables (concept, medium) — ASHP / University of Utah, tracking since 2001: 227 active US drug shortages in Q2 2026, third consecutive quarterly rise; 48% of new 2026 shortages are sole-source; 10% are CT/MRI contrast agents; generic sterile injectables ~67% of shortages overall. The cause is economic, not disruptive — "near-zero margins that discourage redundancy" — so it does not self-correct, and near-term FDA enforcement tightening removes marginal capacity before adding any. The decisive open question is named: do GPO fixed-price contracts absorb this as allocation rather than price? That is the lta-contract-structure-as-price-insulation lens that stranded BWXT.
• 3 new person entities: dave-bozeman, damon-lee, michael-castagnetto (C.H. Robinson CEO/CFO/NAST president).

Updated theses (existing active)

• venezuelan-heavy-sour-return-to-usgc-coker-differential-capture — step 3 partial → confirmed, and yesterday's replacement question is answered NO. Kpler — a commodity-analytics firm with no refining P&L — supplies the quantification the step had been missing since it was filed: sour coking margins vs WCS "around $20/bbl in Q1-26 (QTD), compared with roughly $15/bbl during the same period last year" (~33% widening), attributed by name to Venezuelan crude returning, which "materially improved feedstock economics for complex Gulf Coast refiners." Until today every confirmation came from refiners describing their own margin. And the coking-ceiling question resolves NO: US Gulf coke production fell 4% YoY through September 2025 while weak HSFO is prompting refiners "to use more HSFO in their cokers." You do not top up cokers that are full — barrel availability, not conversion capacity, is binding, so the beneficiary set does not narrow. ⚠ Two qualifications carried: the $20/bbl benchmarks against WCS, not Merey; and Argus's actual thesis is petcoke oversupply weighing on the coker's by-product revenue, an unquantified offset.
• broker-negligent-hiring-liability-to-freight-capacity-bifurcation — the defendant's own quarter cuts BOTH ways, and answers yesterday's #1 open question. Yesterday I flagged "the brokers rallied with the carriers — either the market isn't pricing the ruling, or step 4 is wrong," and called it the cleanest near-term test. CHRW's Q2 FY2026 (2026-07-29) answers it: neither — the cost hasn't landed yet. Supporting step 3, CFO damon-lee: "We are covered through the end of 2026. We are just now starting to have preliminary discussions with the various insurance carriers"; expense "will likely rise." That converts a vague forecast into a dated catalyst: the 2027 renewal cycle. Against steps 4–5, the brokerage P&L is completely unimpaired — NAST operating margin 40.9% (+280bps YoY), truckload AGP per load flat against spot costs +29–30%, 13th consecutive quarter of volume beating the Cass index, guidance held. ⚠ The chain's timing was wrong, not necessarily its direction — and steps 4–5, not step 3, are now its weakest links. A new binary falsifier is set: if CHRW's 2027 renewal completes at ordinary market terms, step 3 is falsified and the chain should be closed.
• ⚠ And a deeper problem for step 4, unresolved. dave-bozeman defends the case by noting the carrier "held the highest rating from FMCSA when we selected it and after a federal review of this accident" and had "safely delivered nearly 270 loads." If the highest available public safety signal confers no protection, brokers cannot sort capacity by safety rating — which cuts the bifurcation mechanism at its root. No source addresses this.
• driver-supply-removal-to-truckload-contract-rate-inflection — corroborated by a party that LOSES from it. CHRW is a broker: truckload inflation is its cost, not its revenue, so it has no incentive to talk rates up. michael-castagnetto: "75%-85% of truckload freight moves on contracts", repricing "continuing into Q3"; damon-lee: repricing the book "for the next six, 12, 18 months." It absorbed spot costs +29–30% while holding AGP/load flat. ⚠ And CHRW cut its market volume assumption from 0–5% growth to −3% — rates up on volumes down is exactly this chain's supply-driven signature, and it rules out any volume-growth version of the thesis.
• commercial-auto-insurance-hardening-to-specialty-insurer-rerate — evaluated for graduation; did NOT graduate. Priority raised low → medium (it now has a dated catalyst it lacked). Three of its four watch-items gained evidence — Montgomery as a specific structural driver, CHRW's dated renewal, and confirmation the line is still hardening. But its own second stated risk is now better evidenced than its thesis: commercial auto liability ran a 113.0% combined ratio in 2024 and S&P projects it above 100% through at least 2029. Rate is rising because the line loses money, which is not the same as the line becoming profitable. The gate is now sharper: a specialty writer showing rate running ahead of loss-cost trend, not merely rising premium.

Contradicted / weakened

• ⚠ /calibrate #41 — new pattern tag process/unverified-denominator. See Top of mind. The six-dispatch "tiering is calibrated" claim was validated over a filtered population without checking what the filter excluded.
• iran-fuel-shock-consumer-bifurcation downgraded to conviction: low (from unlabeled), beneficiary leg in question. Not closed — the competing explanation is strong but inferential as applied to off-price apparel. Dated falsifier set: TJX/ROST next comp print.
• 4e gate: 23 live rows, 2 flags < −10% — TJX −12.8% (⚠ worsened for a third consecutive mark, from −11.9%) and PWR −10.6% (improved from −11.4%). CAT −9.2% has cleared. 0 automatic calibrations — but 1 calibration filed anyway, on the gate's own blind spot rather than on a price divergence. Filing on the instrument rather than the reading is the higher-value entry. Best live: MP(drone) +10.0%, DNN +8.6%, NXE +7.1%, AMZN +6.1%, STVN +5.3%. Worst cluster: energy-oil −6.0%, 9th week — ⚠ again on a day the refiners printed new highs, because the live energy-oil rows are CCJ/CF/NTR, not the refiners. Third consecutive week this label mismatch is noted without being fixed. Best cluster: critical-minerals +4.2%.

Open questions worth a human's eye

• The freight complex has now rallied three sessions running while I have twice declined to chase it. KNX −11.5% off high, SNDR −7.5%, JBHT −5.8%. The evidence keeps improving and the discount keeps closing. At what point is "no valuation gap" the wrong reason to stay out of the book's best-evidenced new chain? This is a genuine human call, not a data gap.
• Does GPO contracting turn the drug shortage into allocation rather than price? Decides whether sole-source-concentration-in-generic-sterile-injectables is tradeable at all. One earnings call away.
• Is the private-label rent captured by the retailer or the co-manufacturer? Unquantified by any source; blocks the consumer chain from graduating.
• Does Montgomery survive Congress? Statutory interpretation, so a FAAAA amendment reverses it. Highest-probability reversal path, no date. Unchanged.
• The energy-oil cluster label is measuring the wrong names — third week flagged. CCJ/CF/NTR sit in energy-oil while the refiners (the cluster's actual best chain) do not. This corrupts the trader's cluster-level risk budgeting.
• Why do PWR and CAT still move together when the corrected mechanism says they shouldn't? PWR −0.8%, CAT −0.1% today; eighth consecutive co-movement mark. Still unexplained.
• The hypothetical inception re-mark is deferred an eighth week — blocked by three standing data defects (KLAC split factor, CBRS entry_ref, SPCX SPY backfill). Escalating as a human call for the seventh dispatch running.

What I looked at

• Run status: headless; steps 0.5–7 in-turn. Price fetch paced 8/min across 16 batches — 118/123 priced, zero 429s, all 08-13 closes. No mechanical defects this run (the two found on 08-13 stayed fixed). ⚠ 5 persistent plan-tier failures: GNKG, HXSCL (known) plus THS, ASE, BESI — all three newly identified as not available on the Twelve Data free plan. ASE and BESI matter: cowos-packaging-capacity-crunch names both as suggested tickers and neither can be priced, so that concept's snapshot is structurally incomplete.
• Breadth (2a): ai-infrastructure 44% ⚠ over (48 distinct chains, 66 live signals, 8 clusters) → net-new steered to thin/absent verticals. Delivered: the one new hypothesis is transport/industrials-ex-AI; the new concept is healthcare; both macro buckets were absent verticals. Zero net-new ai-infrastructure chains opened, third consecutive day.
• Macro buckets: #8 healthcare & demographics and #10 consumer shift — both absent verticals, both produced filable material (unlike 08-13's #9, which produced nothing).
• Ingestion veins (0.5–0.7): Earnings 1 — CHRW Q2 FY2026, pulled ad-hoc; CHRW is not on EARNINGS.md but is the named exposure in a mechanism filed yesterday, and it turned out to be the highest-value source of the day. Watchlist proper was dry (DE moved to 08-20, not 08-14 as estimated; WMT ~08-20; NVDA late Aug). Feeds 0 — every post through 08-10 already ingested; the only new item (Fabricated Knowledge, 08-13) is fully paywalled, a 157-char RSS stub, second consecutive paywalled post from that feed. Podcast 0 filed — see below.
• ⚠ Podcast vein: BOTH diarization providers confirmed unfunded, day 8, and this time both were probed rather than assumed. AssemblyAI returns "Your current account balance is negative"; xAI returns "Your newly created team doesn't have any credits or licenses yet." Budget is $1.05 of $50 — funding, not budget, is the constraint. 22 new episodes discovered across the watchlist, zero had a publisher RSS transcript (the free path that carried 08-13). A local Whisper fallback was started on Biotech Hangout ep.192 (64 min, healthcare — the absent vertical) and did not finish within the run; no clipping was written. Action for Paul (8th day): top up AssemblyAI, or fund xAI and switch diarize_provider: grok.
• Sources: 5 promoted + 5 ingested → 1 new concept, 1 new hypothesis, 3 person entities, 5 mechanisms/questions updated, 1 step promoted to confirmed, 1 mechanism downgraded, 1 calibration (#41, new pattern tag).
• prospect-chains (2b): 1 auto-drafted (driver-not-equipment-scarcity-to-class-8-oem-derate, move: second-order beneficiary — inverted). The other strong candidate was a graduation move on commercial-auto-insurance-hardening-to-specialty-insurer-rerate, which was evaluated and declined.
• Valuation snapshots refreshed: 21 active concept pages carrying tickers, all dated to 08-13 closes.
• Weekly ledger (4d): run (prior 08-07). Hypothetical inception re-mark deferred a 5th consecutive week, not fabricated.
• Not done, and named: EARNINGS.md still has no HWM, ATI, BWXT — fifth consecutive day — and still no SNDR/KNX/CHRW despite CHRW being today's best source, which had to be pulled ad-hoc. FEEDS.md Active list is still 3-of-4 semis/AI-infra while four Candidate feeds map to the absent verticals. Both user-curated — flagged, not edited.
• Signal feed (7): full active/armed set re-marked to 08-13 closes; as_of re-dated only for rows whose research moved.