Article published Jul 3, 2026. Prices below use latest available snapshots.
hot-take · self
First live run of the systematic peer-review workflow: take a follow's
concentrated call, rederive every leg from our own data (tape summaries,
statement detectors, filing watch, rotation map), and score agree/dissent
per name with receipts. Source under review: @crux_capital_'s 2026-07-02
post putting 40-50% of his book into six physical-AI names
(2026-07-02-pick-crux-capital-his-5-conviction-physical-ai-names-40-50-of-book-oust-amba-cgnx-aev).
Headline finding: OUST filed a 424B5 share-offering prospectus the same
day as the conviction post (OUST 424B5 filed 2026-07-02; capture
timestamped 2026-07-02 13:34 PT — intraday ordering unknown, so "same day,"
not "after"). The dilution-red-flag composite (tape + fundamentals +
filings + sector, July 2 data) surfaced this without any prompting —
raising cash while inventory is outrunning sales. The tweet does not
mention the raise.
Second unremarked flag: AMBA's inventory divergence is worse than OUST's — +105% YoY inventory vs +17% YoY revenue (FQ1) against OUST's +98% vs +49% (statements fetched 2026-07-03). Caution-class, not a verdict — but it's the kind of receipt a concentrated-conviction post should have to answer.
Scorecard
Tape rows are as of 2026-07-01, the latest settled bar on disk for these baskets (2026-07-03 is a market holiday). Statement flags came from a Massive statements fetch for the five names that had none on disk, refreshed 2026-07-03. Detector absence on a measured name means "no flag fired," never "verified clean."
CORRECTION 2026-07-03 (same-day, Wave 14 receipts audit A309): the VPG row originally said "no detector flags" — wrong. VPG's revenue-accel FIRED (38 fired vs 12 published; it missed the publish cut, it did not fail to fire). Detector truth = the fired/state layer, not the published top-N. Row fixed.
| Name | Our tape read (as of 2026-07-01) | Our statement read | Verdict |
|---|---|---|---|
| OUST | strong-up, +36.3% 30d, +228.8% 3m, RSI 70.6; stealth-strength vs its basket (-2.0% 30d, insight-wire data as of 2026-07-02) | inventory-divergence: +98% inv vs +49% rev YoY (FQ1) | AGREE on the layer thesis and momentum; DISSENT on the omitted risk — same-day 424B5 + inventory flag is the top red-flag composite. Deep-read owed (also the June 28 $OUST deep dive). |
| AMBA | strong-up, +22.4% 30d, +81.4% 3m, RSI 67.8 | inventory-divergence: +105% inv vs +17% rev YoY (FQ1) — worse ratio than OUST | AGREE-WITH-FLAG: tape supports, balance sheet asks a question the post doesn't answer. |
| CGNX | strong-up, +11.3% 30d, +46.1% 3m, RSI 65.4 | margin-inflection: op margin +22.3% (FQ1) vs +15.7% prior-4Q median | AGREE — cleanest name in the basket: durable-class, fundamentals genuinely improving, tape not overheated. |
| AEVA | strong-up, +6.5% 30d but +40.7% 7d (late spike), +124.3% 3m, RSI 64.2 | margin-inflection but from -768.4% to -561.1% op margin — deeply pre-profit | AGREE on momentum, spec-tier only: the "inflection" is a smaller fire, not profitability. |
| VPG | strong-up, +18.5% 30d, +240.1% 3m, RSI 66.7 | revenue-accel FIRED (below publish cut): +17.6% YoY FQ1 vs +10.9% prior-quarter YoY, +6.7 pts | AGREE, strengthened — fundamentals detector agrees with our own thesis (deep-dive 2026-06-30-vpg-deep-dive.md, humanoid precision-sensor arms-dealer); convergence, not deference. |
| CCXI | +67.3% 30d, +36.7% 7d, RSI 93.6, trend neutral | measured, no detector flags | NO-CHASE: nothing wrong in the statements, but RSI 93.6 mid-pop violates the entry discipline (bottom→pop→flag→breakout); wait for the flag. |
Sector context both ways: Robotics is an emerging sector on the rotation map (accel +7.4, 60% up), which supports the basket-level thesis — and also means the six tapes are partially one correlated read, not six independent confirmations.
Verdict
4 of 6 directionally AGREE (CGNX and VPG cleanly; OUST and AEVA on the thesis with tier caveats), 2 carry engine dissents the source didn't surface (OUST dilution-into-strength, CCXI chase entry), and 1 flag the source never mentioned (AMBA inventory). The peer review's value was exactly the asymmetry: he supplies the layer map and conviction; the engine supplies the risk half — filings, statement quality, and entry discipline — mechanically, from data we already carry. Total marginal cost of the review: one ~60s statement fetch for five uncovered names; every other receipt was already in the standing feeds.
Workflow note (repeatable): capture → per-name tape row (summaries) → statement detectors (fundamentals-wire) → filing watch → cross-wire composites → verdict table with receipts. This is the deep-read-queue SOP applied to a source basket instead of a queue symbol.
Follow-ups: OUST deep-read is already the queue's top item (424B5 use
of proceeds + inventory build vs the industrial-lidar design-win story).
AMBA inventory build goes to the same read pattern next cycle. No
watchlist mutations — all six names already tracked in robotics /
humanoid-robotics.
Sources
- Tape/technical reads: robotics,
humanoid-robotics.json,rotation-map.json(data as of 2026-07-01). - Fundamentals/filing detectors: fundamentals-wire,
filing-watch.json,cross-wire.json(data as of 2026-07-02).