journal 2026-10-22

Journal — 2026-10-22

Promoted the one capture in today's queue: 10-inbox/raw/2026-10-22-bipartite-culture-epistemology-the-vaults-source-tier-apparatus-revives-the.md — a bipartite bridge task pairing observation-vault-source-tiers-revive-preuve-legale-run-in-reverse (the 2026-07-12 reading of the vault's source-tier apparatus as a reversed medieval système de preuve légale) against claim-independent-roots-is-bookkeeping-not-a-categorical-bar (the 2026-10-19 close reading of the one field that apparatus actually gates on). Running headless, so I winnowed and wrote without pausing for a go-ahead.

What I kept. One note:

observation-independent-roots-and-preuve-legale-are-graduated-bookkeeping-not-flat-bars — the actual verdict, and the only genuinely new synthesis in the capture. The two target notes don't cite each other and I found no prior capture pairing them; the earlier note's "two witnesses" framing read the vault's corroboration rule as a flat categorical bar, and reading the field directly shows it's narrower — gated at exactly two checkpoints, otherwise a label, not a refusal. The medieval system it's compared to turns out to have the same shape on close reading: graduated half-/full-proof accounting, with categorical refusal reserved for named, high-stakes chokepoints, not a blanket two-witness rule either. I folded in the eleven-week gap between the two notes' dates as supporting context rather than its own note — it explains why the earlier note overshot (there was no named field yet to check its image against), but it has no independent truth-value on its own, so per the "atomic is not many" ruling it belongs inside the one finding it supports.

What I left behind. Three of the capture's four "Claim:" sections were already fully established by existing notes and would have been pure duplication: the narrower-than-it-sounds reading of independent_roots is already claim-independent-roots-is-bookkeeping-not-a-categorical-bar in full; the graduated historical system (half-proof, full-proof, the criminal-case exception) is already claim-roman-canon-law-rated-one-witness-equal-to-a-private-document and claim-two-half-proofs-combined-to-full-proof-except-in-grave-criminal-cases. All three are linked from the new note instead of restated. The capture's "Further leads" (the root_shared combination-rule question, Bloemberg's anti-monocausal argument, the suspicious-perfection cross-check) are curiosities, not load-bearing for anything I kept, so I left them in the capture rather than opening new 50-questions/ entries — the pile is a promise, and none of these earn one yet.

Questions. None newly routed. The one flag my new note's weakest leg still rests on — the Catholic-Encyclopedia-only "clearer than the noonday sun" criminal-case exception — already has an open question, question-verify-half-proof-combination-rule-criminal-exception-primary, from 2026-07-11. I linked it rather than duplicate it, and didn't rule on it: the capture didn't move that verification any closer, so it stays open exactly as it was.

Entities. The capture's candidate list was mostly right but had one real miss: it claimed Adhémar Esmein had "no existing entity page found." He's had one, entity-adhemar-esmein, since 2026-08-07 — the batch worker that drafted this capture apparently didn't check 40-entities/ before asserting the gap. Logged as a [defect] in 00-meta/seek-flags.md; no page action needed for Esmein, since nothing here teaches his page something it doesn't already say. I did append one dated line each to entity-constantine-i and entity-azo — both pages describe figures the "flat two-witness bar" reading leans on, and this capture's correction (the historical system was graduated, not flat) is new information about how their doctrines actually sit relative to each other, worth recording on both hubs rather than leaving it findable only through the new note. entity-r-g-bloemberg already has a page and needed nothing added.

What felt off. Mostly sound work, but two things worth flagging. First, the entity-page miss above — a small thing on its own, but it's exactly the shape of error that compounds if a batch worker is routinely not checking 40-entities/ before proposing new hubs; I didn't have the tooling here to check other recent captures for the same pattern, so I'm naming it rather than chasing it. Second, the capture itself is honest about its own limits in a way I appreciated — it explicitly says claims 2 and 3 are "already established" / "already vault-sourced" rather than papering over the overlap, which is most of why winnowing them out was easy. No tier inflation and no duplicated research beyond what the capture flagged on itself.

Skipped per headless-mode instructions. No pause for sign-off (none possible, unattended run); no git commands (the auto-commit agent handles that within 15 minutes); this journal entry stands in for the "report to Cali" step.

Second promotion: the Corbeel & De Haro corroboration-check capture

Promoted 10-inbox/raw/2026-10-22-corbeel-de-haro-2026-arxiv261028185-has-hit-the.md — a direct follow-up on question-corroborate-corbeel-de-haro-2026-arxiv-2610-28185-single-source-cap, which I opened yesterday when Corbeel & De Haro (2026, arXiv:2610.28185) hit the vault's single-source concentration cap at exactly three notes. The capture asked two things: has anyone independent engaged with the review yet, and does the parked title-slogan finding earn the fourth slot. Running headless again, so I winnowed on my own judgment.

What I kept. One genuinely new claim-note:

claim-timpson-2004-quantum-information-slogan-is-methodological-not-ontological — Timpson's own 2004 thesis, read directly, stating the ontological-vs-methodological distinction Corbeel & De Haro lean on when disclaiming their title slogan. This doesn't rest on the capped preprint at all — Timpson predates it by 22 years and is a solo-authored primary in his own right — so the cap never touches it. It's also the cleanest piece of philosophy in the whole capture: the same sentence ("it's about information") can be a toolkit-claim or a reality-claim, and nothing in the sentence says which. Worth a hub, too: entity-christopher-timpson, his first page in the vault, promoted on separate standing (a solo thesis doing real analytic work across two existing clusters) rather than mere co-authorship.

What I folded into existing notes instead of duplicating. Two of the capture's four "claims" were independent Tier-1 primaries directly confirming mechanisms the vault already had sourced only to Corbeel & De Haro's own restatement — a real strengthening, not a new atomic claim. I added each as a second independent root rather than write thin "X confirms Y" files: claim-corbeel-de-haro-2026-hayden-preskill-black-hole-as-information-mirror now carries Hayden & Preskill's own 2007 wording for "information mirror" and "delicious irony," and claim-corbeel-de-haro-2026-black-hole-central-dogma-is-unproven-modelling-assumption now carries Almheiri et al.'s own 2020 wording for the central dogma's statement, its "unproven assumption" status, and — new fact, not just new sourcing — Hawking's own named objection to it, which I also logged onto entity-stephen-hawking.

What I didn't promote, and why. The fourth "claim" — that no independent source has engaged with the review and the cap stays shut — isn't a vault claim about the world, it's the direct answer to the question that sent me looking. I appended it to that question's progress log and left it open: the cap is still undischarged (Hayden & Preskill and Almheiri et al. are sources the review summarizes, not third parties assessing it — corroborating a review's homework is not the same as corroborating the review), so the held title-slogan claim stays parked, exactly where it was. I also declined hub pages for Hayden, Preskill, and the five-author Almheiri group — co-authorship of one shared paper each, no independent standing shown this session, which the 2026-10-01 ruling says stays a mention. Same for 't Hooft: named as the oldest origin point but never read at primary here, so UNSURE, no page.

Questions. None newly opened — the one question this capture exists to answer already existed, and I ruled on it as a partial-progress update rather than closing it, since the actual thing it's waiting on (independent engagement with the preprint) still hasn't happened and may not for a while.

Entities. One new hub (Timpson), two updates to existing pages. I appended a dated line to entity-sebastian-de-haro noting the still-undischarged cap and a third De Haro-cluster paper (De Haro, Cinti & Corbeel, arXiv:2608.26282) that cites the review but shares two authors with it — not independent, but worth knowing the cluster has a third node. And the Hawking line above.

What felt off. Nothing about the sourcing — every quote in this capture was a direct primary read, Tier 1 straight down, which is about as clean as this vault's inputs get. What's worth naming instead is a near-miss the capture caught in its own commentary and I want on record in my own words too: it would have been very easy to read "an independent primary confirms the mechanism the review restates" as "the cap is discharged" — they sound adjacent, and I had to stop and separate them twice while writing the central-dogma update. The cap was never asking whether the content holds up; it's asking whether anyone besides De Haro and Corbeel has looked at this specific paper and said so. Strengthening the floor under three seedling notes is real work and I'm glad I did it, but it answers a different question than the one the question file is actually asking, and I don't want a future session skimming the "independent_roots: 2" bump and mistaking it for closure. Also logged in 00-meta/seek-flags.md: a [defect] note that philsci-archive.pitt.edu 403'd both fetch routes (one data point, not yet a pattern), and an [entity] note that three person-hubs already use "central dogma (black hole physics)" as a connects_to string with no concept hub behind it yet — not mine to build this session since the capture never flagged the term itself, but it's recurring enough that someone should.

Skipped per headless-mode instructions. No pause for sign-off; no git commands; this journal entry is the report.

Third promotion: does an independent source corroborate RIKER?

Promoted 10-inbox/raw/2026-10-22-does-an-independent-source-corroborate-rikers-roig-2026.md — a direct follow-up on question-corroborate-riker-2026-hallucination-rate-range, the single-source-cap corroboration question I opened on RIKER (Roig, 2026) yesterday. Running headless, so I winnowed on my own judgment.

What I kept. Three notes, down from the capture's four "Claim:" sections — I merged two into one per the atomicity ruling:

claim-riker-2026-no-independent-source-same-author-admits-no-replication — merges the capture's first two claims. Both candidate "corroborating" documents (a 172-billion-token same-author/same-company follow-up, and a Signal65 report run "as part of an ongoing collaboration" with RIKER's own creator) fail the independence test on the same grounds, and the follow-up's own text states directly that RIKER "has not been independently replicated with alternative frameworks." One finding, two complementary evidentiary routes — the author's own admission and an independence-test analysis of both candidates — not two claims that could be true or false independently of each other.
claim-riker-2026-self-replication-contradicts-original-fabrication-gap — the actually interesting result. The same-author follow-up, at far larger scale, does not reproduce the original paper's 40.9-point GLM-4.5-vs-Qwen3-Next-80B fabrication gap — it reports both models in the same low-fabrication tier (roughly 5-6 points apart, not 41). Not independent disconfirmation, since it's the same author, but a self-correction big enough that the original figure shouldn't circulate without it attached.
claim-riker-2026-7-9-to-96-3-range-is-128k-not-32k-table — a citation-accuracy correction. The "7.9% to 96.3%" sentence everyone (including the vault's own existing claim-note and the open question that sent me here) attributes to RIKER's 32K-context results actually traces to its 128K table. I fixed this in the existing note directly rather than leaving the wrong number live in the note it belongs to — see its new Correction history block.

Retrieve-before-write. No true collisions — the capture's claims are genuinely new findings about the RIKER lineage's corroboration status, not restatements of anything already in 30-notes/. I linked the parallel-shape notes (claim-bit2watt-no-independent-corroboration-found-as-of-2026-09-10, claim-fic-preprint-no-independent-corroboration-as-of-2026-10-20) as the same absence-finding pattern rather than treating them as overlapping content.

What I left behind. The capture's "Further leads" — Signal65's unpublished PINNACLE project, its incompletely-extracted 91-model table, a full PICARD citation-chain audit, and an unchecked Semantic Scholar citation count — are all genuine next steps but none load-bearing for anything I kept, so none became a new 50-questions/ entry. Logged as not_promoted on the capture with reasoning for each.

Questions. None newly opened. I ruled on the existing one instead: marked question-corroborate-riker-2026-hallucination-rate-range answered — the search is genuinely thorough and conclusive (no independent source exists, the closest thing to independent evidence contradicts the magnitude, and the question's own premise needed the 128K/32K correction), not a partial progress update this time.

Entities. JV Roig gets his first hub (entity-jv-roig) — sole author across the entire RIKER/PICARD lineage several claim-notes now rest on, with real standing beyond co-authorship. PICARD itself gets a watching stub (entity-picard-evaluation-framework) with first_seen stamped now rather than lost — it's RIKER's own named methodological ancestor but hasn't recurred independently yet, so a stub, not a hub. Kamiwaza AI and Signal65 stay mentions: both are single-cluster so far, and the org-promotion bar is deliberately stricter than the person bar. Mitch Lewis stays a mention too, as the capture itself already judged.

What felt off, flagged to Cali. This capture pushed arXiv:2601.08847 to the single-source concentration cap — three claim-notes now rest on it directly, with no independent corroboration in sight. That's not a sourcing problem in any one note (every quote here is a direct Tier-1 read with a sha receipt), but it's exactly the shape the cap exists to catch: a contrarian, unrefereed preprint now has a three-note subtree, and the only thing resembling a check on it is the same author's own later paper, which agrees on the qualitative finding and disagrees by 34 points on the quantitative one. I logged this as a [watch] line in 00-meta/seek-flags.md so a future promotion session checks the ceiling before adding a fourth note from this source. No tier inflation otherwise, and the capture was honest about its own limits — it flagged the 128K/32K mixup on itself rather than letting a future audit find it.

Skipped per headless-mode instructions. No pause for sign-off; no git commands; this journal entry is the report.

Fourth promotion: the Anderson/Marr-extension <-> Oaksford-Chater-yardstick-swap dup-risk log

Promoted 10-inbox/raw/2026-10-22-dup-risk-resource-rational-analysis-anderson-marr-oaksford-chater-bridge.md — a hop-batch bee's own dup-risk investigation, not a fresh research capture. It set out to check whether an unlinked cosine-0.88 pair, claim-resource-rational-analysis-extends-andersons-rational-analysis (RRA extends Anderson toward Marr's algorithmic level) and claim-resource-rational-analysis-rationality-revisited-swaps-normative-yardstick-same-move-as-oaksford-chater (RRA's §5.4 performs the Oaksford & Chater yardstick-swap), is a real bridge or an embedding-space mirage. Running headless, so I winnowed on my own judgment rather than pausing.

What I kept. Nothing new in 30-notes/. The bee re-read Lieder & Griffiths 2020 directly via extract_pdf and found a genuine, previously-unquoted §3 sentence — "Rational process models can be used to connect the computational level of analysis to the algorithmic level of analysis," paired with the bounded-optimality clause right after it — that is the paper's own stated mechanism linking the two notes' claims. That's a real finding. But the bee's own vault_novelty check on it came back max_cosine 0.841 (duplicate-risk) against note B itself, and I agree with that verdict on inspection: the conceptual territory — resource-rational analysis as a textually-bridged extension of Anderson that also performs the Oaksford & Chater swap — is already fully mapped by moc-the-yardstick-was-wrong-not-the-reasoner plus observation-oaksford-chater-uncertainty-redefinition-resource-rational-pairing-is-genealogical-bridge and claim-lieder-griffiths-2020-list-probability-theory-among-three-classical-norms-resource-rationality-supersedes, both of which already carry adjacent normative-standard quotes from the same paper's §1/§7. A new claim-note on one more sentence from a paper the vault has already quoted five times in this exact cluster would duplicate a family, not extend it. I'm overriding nothing here — the capture's own disposition already reached this conclusion; my job was to check it, not rubber-stamp it, and it holds.

What I didn't do, and why it's not mine to do here. The capture's actual ask — add a cross-link between the two seed notes (and ideally the list-probability-theory note, which sits between them citing the same paper and links to neither) carrying the §3 sentence as the connecting quote — is a real, small, well-sourced gap. I didn't make that edit. The bee that found it was sandboxed out of 30-notes/; I am not, but promote's job this session is writing new atomic notes, not amending three already-live ones, and nothing in my instructions extends the entity-page additive-update license to claim-notes. Editing live claim-notes without that kind of explicit license felt like exactly the sort of overreach "winnow conservatively" is warning against, even though the edit itself would be harmless. Routed it instead as an [entity] line in 00-meta/seek-flags.md — hub/MOC-shaped cross-linking work, which is the Warden's lane by precedent (the yardstick MOC itself exists because the Warden discharged a similar flag).

Retrieve-before-write / collisions. This was the whole point of the capture, so: checked, and the semantic-index hint's twelve related notes were all already-known members of this cluster — no new collision found, no new note needed to resolve one.

Questions. None opened, none closed. I checked 50-questions/ for anything on this exact pairing; the two closest matches (question-resource-rational-analysis-sharpest-werbos-anderson-bridge, question-verify-rational-analysis-anderson-1990-marr-computational-level) are both already _answered/ and about a different leg (Werbos, not this bridge), so there was nothing live to rule on.

What felt off. Nothing about sourcing — the bee's re-read was Tier 1, direct, sha-matched to five other notes' existing fetches of the same PDF, about as clean as this cluster gets. What's genuinely interesting is the shape of the capture itself: it's a bee correctly declining to write a note, and the only thing actually left over for promotion to do is a graph edit outside promotion's own remit. That's not a flaw in the capture, but it does mean this promotion's "output" is one declined claim and one routed flag — thinner than usual, honestly reported rather than padded.

Skipped per headless-mode instructions. No pause for sign-off; no git commands; this journal entry is the report.

Fifth promotion: conservation in every quantum measurement outcome, the Okon/Sudarsky/Wiedemann critique

Promoted 10-inbox/raw/2026-10-22-outside-arxiv261029804-conservation-in-every-quantum-measurement-outcome.md — an [outside] surface capture, chosen deliberately from outside the vault's existing network, on arXiv:2610.29804 (Okon, Sudarsky & Wiedemann, submitted the day before). This is genuinely new territory for the vault: a critique of a 2025 proposal (Collins & Popescu, Quantum 9, 1815) that a physical "preparer" system restores conservation laws to individual quantum measurement outcomes, not just statistical averages. Running headless, so I winnowed on my own judgment.

What I kept. Three claim-notes out of the capture's five "Claim:" sections — not four, because of a constraint I almost missed:

claim-okon-sudarsky-wiedemann-reference-frame-regress-undermines-preparer-requirement — the paper's reference-frame regress argument: identifying a conserved quantity's reference frame with a physical preparer generates an infinite regress, which (run through a classical analogy) would make Newtonian mechanics unable to describe any finite collection of particles.
claim-okon-sudarsky-wiedemann-preparer-never-definite-only-statistical-conservation-restored — the paper's second independent ground: even granting the mechanism, the preparer never holds a definite value of the conserved quantity before or after measurement, so only statistical conservation is shown, not the individual-outcome conservation the proposal claims.
claim-okon-sudarsky-wiedemann-noncommuting-conserved-quantities-break-compensation-mechanism — a worked two-spin-1/2 example showing the mechanism that compensates for one conserved quantity (Sz) does not simultaneously compensate for a noncommuting one (Sx) in the same state.

What I didn't keep, and why. The capture's own umbrella claim — "the paper argues the proposal is inadequately justified and does not establish what it claims" — isn't atomic beyond the two specific grounds above; I folded it into both notes as framing rather than writing a fourth thing that just restates their conjunction. More consequentially: a fourth genuinely distinct, well-sourced finding (the preparation Hamiltonian's interaction term breaks additive decomposability specifically for energy conservation, where angular momentum doesn't have this problem) is fully promotable on its own merits — but sources.md's single-source concentration cap stops at three claim-notes per unrefereed preprint, and this arXiv paper is one day old with zero chance of independent corroboration yet. I nearly wrote all four before re-reading the cap language closely enough to notice it bites per-source, not per-capture. Held it in a corroboration question instead of silently promoting past the ceiling.

Questions. One new one: question-corroborate-okon-sudarsky-wiedemann-2026-arxiv-2610-29804-single-source-cap, following the house pattern for cap questions. It carries the held energy-conservation finding as a parked fourth slot, and a second, genuinely different doubt I want on record separately: all three kept claims describe arguments made against a characterization of Collins & Popescu's proposal, and that proposal itself was only confirmed to exist via WebSearch this session, never directly fetched or quoted. If Okon et al.'s restatement of the mechanism they're attacking is uncharitable anywhere, the critique could be sound against a strawman. I didn't open a second question for this — it's the same underlying gap (nobody's independently checked this exchange yet) wearing two faces, so one question carries both rather than two thin ones.

Entities. Five new hub pages, all first appearances for the vault: entity-sandu-popescu (the actual target of the critique, with separate standing across the 2025 Quantum paper and two 2021/2023 PNAS papers with Aharonov and Rohrlich — not just one co-authorship), entity-elias-okon and entity-daniel-sudarsky (a genuine two-person research line, not a single shared paper — they have an earlier 2020 paper together with Tim Maudlin on conservation laws and semiclassical gravity), entity-yakir-aharonov (eminent on his own terms — Aharonov-Bohm effect, weak measurement — and an ancestry figure behind the reference-frame literature the critiqued proposal leans on), and a concept hub, entity-reference-frame-quantum-information, since the reference-frame-as-preparer move is load-bearing to the kept regress claim and has its own real literature (Aharonov-Susskind 1967, Aharonov-Kaufherr 1984, Bartlett-Rudolph-Spekkens 2007). I deliberately did not hub Daniel Collins or M. Wiedemann — both are co-authors of exactly one paper each in this capture with no separate standing shown this session, which the 2026-10-01 ruling says stays a mention (named on the relevant hubs instead). Also declined a hub for "Quantum (journal)" as a venue-level candidate not yet load-bearing.

What felt off. Nothing about the sourcing itself — every quote is a direct extract_pdf fetch of the primary with a sha receipt, Tier 1 straight down, and the capture is honest that the critiqued paper's own text hasn't been read yet (flagged on itself as [unverified-quote — needs direct read] rather than papering over it). What's worth naming is the cap near-miss above: this is exactly the kind of brand-new, structurally rich preprint that invites writing every distinct argument as its own note, and the cap exists precisely because that instinct is how a single contested source quietly grows a whole subtree before anyone's checked whether the critique itself holds up. Logged a [watch] line in 00-meta/seek-flags.md so a future session checks the ceiling — and the characterization doubt — before this branch grows a fourth note.

Sixth promotion: does Table 1 or Table 3 actually bridge Liang's 2009 dating and the 2008 book?

Promoted 10-inbox/raw/2026-10-22-what-genuinely-connects-liang-zhongtang-dates-the-one.md — a bridge-seed task testing the cosine-0.88 unlinked pairing of claim-liang-2009-dates-one-child-policy-origin-to-june-1978-three-day-meeting against claim-greenhalgh-2008-book-table-1-names-liang-zhongtang-as-key-actor. Running headless, so I winnowed on my own judgment.

What I kept. Two notes, down from the capture's three "Claim:" sections:

claim-greenhalgh-2008-book-table-3-independently-dates-june-1978-policy-rule — the actual bridge. Table 1 (a bare name-to-era roster entry) isn't what connects the two seed notes; Table 3, one page over in the same chapter and read by no prior session, independently gives June 1978 / October 1978 for the same policy rule Liang's essay dates the same way from an entirely separate document. This closes a real gap: the vault's only prior independent corroboration of June 1978 ran through Greenhalgh's 2003 article alone.
claim-greenhalgh-2008-book-repeats-slogan-not-origin-framing-for-june-1978 — the honest second half. Having found the date match, Greenhalgh's own prose still calls June 1978 a "slogan," not Liang's "instigator" — the existing dating fork (claim-decade-vs-seven-years-fork-depends-on-policy-origin-anchor) recurs in a second of her documents rather than closing.

What I folded rather than promoted. The capture's second "Claim" — Table 1 independently placing Chen Muhua in the same era-row as Liang — isn't atomic enough for its own note: it's context for the Table 3 finding, not a fact that could be true or false on its own against anything the vault doesn't already hold (that Chen Muhua headed the Leading Group from June 1978 has been on file since 2026-08-09). I folded it into the Table 3 note as a supporting paragraph and logged it on entity-chen-muhua's Updates instead of writing a third thin note. Recorded as not_promoted on the capture with that reasoning.

Retrieve-before-write / collisions. None. Checked the semantic-index hint's twelve related notes plus the entity pages directly — the Table 3 material is genuinely unread territory; no existing note covers it.

Questions. None newly opened. Both new notes repeat an English rendering of Liang's 始作俑者 sentence already carrying [unverified-translation] on claim-liang-2009-dates-one-child-policy-origin-to-june-1978-three-day-meeting; I linked both to the existing question-verify-liang-zhongtang-2009-essay-translation rather than opening a new one — same doubt, no new translation surface. The capture's own unreconciled detail (Table 3's second codification date, "January 1979, BP directors," against Document No. 69's 26 October 1978) isn't load-bearing for either kept claim, so it stays a loose end in the capture rather than a question-pile entry.

Entities. Three pages updated, one considered and declined. Appended dated lines to entity-chen-muhua (the Table 1/Table 3 placement), entity-susan-greenhalgh (a second independent instance of her own text dating June 1978, the fork recurring in a second document), and entity-liang-zhongtang (a second Greenhalgh document landing on his date without citing him). Considered and declined an append to entity-song-jian for his own Table 1 "Missile Ministry" entry — true but adds nothing his page doesn't already document about this dispute; a reasoned skip, named here rather than left silent.

What felt off. Sourcing was clean — both PDFs re-fetched this session, sha256 unchanged from every prior citation, no addressed-to-AI language on either document, same as the capture's own safety-flags section reports. The one thing worth naming: this is the cluster's fourth or fifth "shared landlord, not shared argument" verdict in two months (the observation note on this exact shape is already on file, observation-liang-2009-and-greenhalgh-2003-are-symmetric-shared-landlord), and I notice I'm starting to expect that shape before I've read the primaries each time. It held again here, correctly, but it's worth someone checking in a future session that the pattern-match isn't substituting for the actual read — this time it wasn't; I re-fetched and quoted both documents rather than assuming the shape.

Skipped per headless-mode instructions. No pause for sign-off; no git commands; this journal entry is the report.

Seventh promotion: Hal Puthoff, in his own words

Promoted 10-inbox/raw/2026-10-22-who-is-hal-puthoff-in-his-own-words.md — a direct follow-up on claim-hal-puthoff-bridges-1972-stargate-and-2010-aatip-antigravity-research, which established that the same physicist founded 1972 Stargate and authored a 2010 DIA antigravity report but never explained why the same person recurred across 38 years. This capture went looking for Puthoff's own account of the why, in his own writing and his own spoken words, not the DIA reports themselves. Running headless, so I winnowed on my own judgment.

What I kept. Three claim-notes, one per distinct source and distinct explanatory thread — each could be true or false independently of the other two, so none folds into another under the "atomic is not many" ruling:

claim-puthoff-1972-cia-recruitment-rested-on-pre-existing-intelligence-pedigree — his 1996 primary account attributes the CIA's 1972 approach to his pre-existing Naval Intelligence/NSA pedigree, not to anything about the psi results themselves.
claim-puthoff-2026-account-names-bigelow-aerospace-as-the-aatip-recruitment-channel — his own 2026 podcast account names chief science advisor to Robert Bigelow, not any lingering Stargate-era tie, as the actual door DIA walked through in 2010.
claim-puthoff-2010-zpe-antigravity-paper-continues-research-program-since-1989 — his 2010 DIA-era paper's own footnotes show the theoretical apparatus behind it was a personal research program running back to a 1989 Phys. Rev. A paper, not work generated to order for the assignment.

Together they complicate the neater story I'd have liked — "Puthoff is the Pentagon's standing point of contact for fringe physics" — with something more interesting: two unconnected recruitment channels, 38 years apart, and one constant underneath both, a self-built specialist habit of showing up where mainstream physics runs out of explanations.

Retrieve-before-write. Checked the semantic-index hint's twelve related notes first. No collisions — the existing bridge note states the 38-year fact without explaining it; these three notes are genuinely the explanation, not a restatement.

What I left behind. Recorded on the capture's not_promoted: list with reasoning each: the JBIS-2002-vs-2010-arXiv-submission date ambiguity (folded into claim 3 as a non-load-bearing caveat — the claim holds either way, and a document's own publication metadata isn't a 30-notes claim on its own); Jay Stratton's AATIP-co-founder/UAP-Task-Force role (real candidate, too thin here, flagged below); Robert Forward's USAF precedent (folded into claim 3, but earned him his own hub on separate standing); Dan Farah and Steven Bartlett (interview roles only, already judged thin by the capture itself); and the Brian Dunning/SPR-Encyclopedia leads the capture explicitly declined to use.

Questions. None newly opened. No [unverified-*] flags in this capture, and the one open unresolved point (the JBIS dating) isn't load-bearing for anything kept, so per the intake discipline it doesn't earn a question-pile entry. Checked 50-questions/ for anything already tracking Puthoff/AATIP/ Stargate/Bigelow — nothing exists, so no closure to rule on either.

Entities. Two new hubs: entity-robert-bigelow (the actual, named 2010 recruitment channel — real standing beyond a mention) and entity-robert-forward (institutional-precedent physicist Puthoff's own 2010 paper cites directly, separate standing via his own USAF study and Casimir-effect work). Three existing hubs updated additively rather than left silent: entity-hal-puthoff (all three new findings), entity-andrei-sakharov (a new, unrelated-to-Nobel fact — he's also the foundational physicist behind Puthoff's gravity-as-vacuum-fluctuation framework), and entity-ingo-swann (Puthoff's own account names Swann's PK effects as the proximate trigger for the CIA's approach, while attributing the CIA's choice of Puthoff specifically to his intelligence pedigree, not to Swann). Declined hubs for Dan Farah and Steven Bartlett (thin, interview roles only). Jay Stratton logged as an [entity] line in 00-meta/seek-flags.md rather than promoted or silently dropped — real candidate, not enough here to clear the bar yet.

What felt off. Sourcing was clean and I want to say so plainly: three Tier-1 primaries (Puthoff's own 1996 essay, a third-party transcript of his own 2026 recorded words, his own 2010 arXiv paper), each with a direct-fetch sha receipt, no aggregator or paraphrase layer anywhere load-bearing. The one thing I changed from how the capture itself quoted the paper: its captured source_quote_5 joined two sentences across an ellipsis, which the 2026-10-01 quote-shape ruling now disallows for a freshly-written note — I used only the one complete sentence that actually grounds the claim rather than carry the ellipsis forward into a new note. Nothing else needed correcting; the capture was honest about what it didn't chase (Jay Stratton, the companion arXiv paper, the JBIS dating) rather than papering over the gaps.

· raw markdown