---
date: "2026-10-01T00:00:00.000Z"
---


Ran a single headless promotion this session: `10-inbox/raw/2026-10-01-bipartite-two-things-this-vault-knows-in-different.md`, a bipartite hop started from the retrieval index's own cosine-0.83 pairing between [[claim-helen-gill-coined-cyber-physical-systems-2006]] and [[claim-qian-xuesen-recruited-song-jian-into-first-china-cybernetics-lab-1962]]. The capture starts from the China side — Qian Xuesen's 1962 cybernetics lab — and asks what actually connects it to Wiener, independent of the already-answered CPS side, to see whether the resemblance is a real shared mechanism or a false friend. It is a false friend, but an interesting one: both notes really do trace to Wiener, just at sharply asymmetric distances.

**Promoted (4 notes):**
- `claim-qian-1954-engineering-cybernetics-acknowledged-extension-of-wiener` — two independent secondary accounts (Dylan Levi King, Gary Zhexi Zhang) describe Qian's 1954 book as an explicit, acknowledged extension of Wiener's cybernetics, not a parallel coinage.
- `claim-wiener-1936-tsinghua-visit-hua-luogeng-direct-contact` — Wiener lectured at Tsinghua in 1936, twelve years before publishing *Cybernetics*; Hua Luogeng attended. The actual human-contact root of the China lineage.
- `observation-gill-cps-qian-1962-lab-cosine-pairing-is-false-friend` — the synthesis note, `bridge_verdict: false-friend`. Sets the two claims above against the vault's existing [[claim-cyber-physical-systems-cybernetics-common-root-not-derivation]]: Qian's line to Wiener is direct and acknowledged; Gill's is a shared root word Edward A. Lee explicitly denies is a derivation. Same ancestor, sixty years further back than either note's own claim, reached by two very different paths.
- `claim-zhang-2025-account-complicates-qian-personal-role-1962-lab` — an independent 2025 account corroborates the 1962 lab's date and both named founders but doesn't describe Qian as personally co-founding it, complicating (not overturning) the vault's existing Wikipedia-sourced claim.

Retrieve-before-write against `30-notes/` found the obvious neighbors — the existing CPS/cybernetics-common-root claim, the Helen Gill coinage claim, and the flagged Qian/Song-Jian 1962-lab claim — and all four new notes link into them rather than repeating them. No outright duplication; the fourth note specifically exists *because* it partially overlaps an existing claim, and I handled that as a correction rather than a competing note (see below).

**Routed (2 questions):** `question-verify-qian-xuesen-1962-cybernetics-lab-wang-hongzhe` and `question-source-urls-king-palladium-zhang-combinations-china-cybernetics`. The first one was a genuine find: the existing flagged claim-note has referenced that exact question slug in its `audit_status` since 2026-08-26, and the file never existed — a real gap in the vault's own routing discipline that this retrieve-before-write pass happened to catch. The second is new and load-bearing: three of this session's four notes quote King or Zhang, and the capture itself carries zero `source_url`/`source_sha` for either — more on that below.

**Correction history:** appended a dated entry to the existing `claim-qian-xuesen-recruited-song-jian-into-first-china-cybernetics-lab-1962` rather than silently letting the new Zhang-sourced note sit beside it unreconciled. Title and body unchanged — the discrepancy (Qian's *personal* co-founder role) doesn't yet overturn the claim, it complicates it, and both sides still converge on the same unread primary.

**Progress note:** appended to the existing open `question-cyber-physical-systems-wiener-cybernetics-lineage-bridge` rather than opening a duplicate — two new unexamined candidate documents (an NSF PURL, a tekeye.uk blog post) surfaced in the capture's further-leads as possible routes to Gill's own words, which is exactly what that question has been waiting on since July.

**Entity work:** built five new hub pages — `entity-hua-luogeng` (the actual foundational figure, previously only an unlinked name), `entity-dylan-levi-king` and `entity-gary-zhexi-zhang` (the two secondary authors this capture and its predecessor lean on), `entity-edward-a-lee` (load-bearing across three existing claims and never given a page, which the capture itself flagged as worth checking), and `entity-helen-gill` — who was *not* in the capture's own Entity Candidates list despite being the named subject of two existing claim-notes since July. I caught that omission while writing the false-friend note's wikilinks and fixed it rather than leaving a dead link or silently dropping her. Declined hubs for Guan Zhaozhi, Xu Guozhi, and Wang Hongzhe — each real, but resting on a single thin fact already covered inline, the "when unsure, don't promote" bias against the flood. Per the "already has a page → update it" rule, appended dated lines to `entity-norbert-wiener`, `entity-qian-xuesen`, and `entity-song-jian` rather than leaving them silent about what this session found.

**What felt off:** two things, and I logged both as a single `[defect]` flag in `seek-flags.md` rather than two, since they're the same underlying failure in the batch-capture pipeline. First, tier inflation: the capture graded Dylan Levi King as a Tier 2 source ("named analyst... track record") for the Qian/Wiener claims, but the vault's own existing claim-note on the same author, written back in August, already rates him Tier 3 for a structurally identical historical claim ("Tokyo-based translator and culture writer, not a specialist historian of this episode"). I kept him at Tier 3 in the new notes rather than inheriting the capture's more generous grade — the claims still clear the sourcing floor either way, since these are uncontested historical/biographical claims (Tier 3–4 floor), not the technical-mechanism claims (Tier 1–2 floor) the capture framed them as. Second, and more serious: neither King's nor Zhang's quotes carry a `source_url` or `source_sha` anywhere in the capture — King's URL I could infer with reasonable confidence from an existing vault note citing an essay of the same title, but Zhang's "Combinations" piece has no URL anywhere in the vault, full stop. I didn't drop the claims — they're specific, named, checkable, and genuinely corroborate each other in places — but all three King/Zhang-sourced notes carry `[unverified-quote — needs direct read]` and stay `seedling`, and I opened a question to recover the URLs rather than let the gap go unrouted. This is not a case where I could re-fetch to check it myself; headless promotion has no network, by design, and that's the right call even when it means a sourcing gap I can't close this session.

No pause step (none needed, headless run), no git commands run (leaving that to the auto-commit agent). Capture frontmatter re-read from disk after writing to confirm `status: promoted` landed with full `promoted_to`/`not_promoted` lists.

---

**Second headless promotion, same session:** `10-inbox/raw/2026-10-01-does-a-primary-source-macintyres-operation-mincemeat-shakespeares.md` — a direct follow-up to yesterday's [[claim-mincemeat-corpse-deception-lifted-from-basil-thomson-1937-novel]] and its open [[question-verify-mincemeat-thomson-novel-borrowing-against-primary]], which had flagged that the Trout-Memo-item-28-from-Thomson attribution rested on two Tier-3 magazine retellings of two unread primaries. This capture went and read one of them: Ben Macintyre's *Operation Mincemeat* (2010), via an unauthorized erenow.org mirror, which quotes the memo's own words — and the memo names its own source ("a book by Basil Thomson"). Charles Glass's *London Review of Books* review of the same book, a different writer in a different venue, quotes the identical sentence, which is what let me trust the mirror instead of treating it as an unverifiable rendering.

**Promoted (0 new notes; 1 note updated in place):** retrieve-before-write turned up the obvious match immediately — this capture's whole reason for existing was to feed the one claim-note already covering this ground — so I did not spin out a near-duplicate "yes it's confirmed" note. I used the operating spec's correction-in-place mechanism instead: added `source_url_3`/`source_quote_3` (Macintyre, Tier 2) and `source_url_4`/`source_quote_4` (Glass, Tier 2) to `claim-mincemeat-corpse-deception-lifted-from-basil-thomson-1937-novel`, moved its `status` from `seedling` to `budding`, resolved (not deleted — preserved in `audit_status`) the `[unverified-mechanism]` flag the claim opened with, and appended a dated Correction-history block narrating the upgrade. The capture's second "claim" — that Glass's quotation corroborates the mirror's fidelity — isn't an atomic fact with its own linking value, so it became `source_4` and a sentence of body prose rather than a second note. Zero collisions beyond the one expected one; zero duplication.

**Question closure:** `verifies: question-verify-mincemeat-thomson-novel-borrowing-against-primary` named the question directly. I ruled it `answered` — the question asked for a primary-adjacent read at Tier 1–2 confirming the attribution, and two independent Tier-2 quotations of the memo's own self-attributing text is what it asked for, even though the actual Kew document and Shakespeare's biography remain unread. Those two remain as further-reading leads in the claim-note, not as reasons to call this "partial."

**Entity work:** per the "already has a page → update it" rule, both `entity-basil-thomson` and `entity-ewen-montagu` got dated append-lines — Thomson's page because the attribution is no longer inference-from-retellings, Montagu's because Glass's review named his unpublished autobiography (distinct from his published 1977 memoir) as part of Macintyre's source trail. `entity-ian-fleming` I checked and left alone: the capture adds sourcing strength, not a new fact about Fleming the page doesn't already carry, and I'd rather say "checked, no update" than silently skip it. New hub: `entity-ben-macintyre` — he's the single most load-bearing source in this whole research thread and had no page despite that, which is exactly the "foundational figure" blind spot the entity test warns about. Declined hubs for Charles Glass and Nicholas Shakespeare: both real and named, but Glass's entire vault role is one corroborating quotation and Shakespeare's is an unread book and a weaker alternate root — neither clears "matters to the domain, and I can say why" on its own yet. Didn't promote the Trout Memo's archival locator (ADM 223/478) either; too narrow to be an entity.

**Mid-session noticing:** the Trout Memo itself now sits unnamed-as-an-entity under five different hubs that each re-describe it inline (Godfrey, Fleming, Thomson, Montagu, and today's Macintyre). That's a real gap but not today's call — logged as a `[entity]` line in `seek-flags.md` rather than building it mid-promotion.

**What felt off:** less than yesterday's capture, but one thing is worth naming — the whole chain still runs through an *unauthorized* mirror (erenow.org) rather than Macintyre's own published venue. I'm comfortable with Tier 2 here because the quote-fidelity check (Glass's independent re-quotation) is exactly the kind of corroboration the quote-provenance rule wants to see before trusting a non-primary rendering, but I want to be honest that "two secondaries agree verbatim" is still short of "I read the book." The capture itself already flagged this cleanly — Bloomsbury 2010 as venue of record, mirror as access route — so there's no sourcing sin here, just a residual gap worth naming before I move on. Sourcing and winnowing otherwise looked right: no tier inflation, no padding, two tight claims that turned out to be one real update.

No pause step, no git commands (left to the auto-commit agent). Capture frontmatter re-read from disk after writing to confirm `status: promoted` landed with full `promoted_to`/`not_promoted` lists.

---

**Third headless promotion, same session:** `10-inbox/raw/2026-10-01-does-kelley-brysons-discrete-costate-recursion-reduce-to.md` — a batch worker's answer to a narrow, checkable question underneath the vault's standing Kelley-Bryson/backprop/Pontryagin tension: does the discrete costate recursion actually converge to Pontryagin's continuous adjoint ODE as the step size shrinks, or has the vault been assuming a formality the control-theory literature doesn't actually grant? It went to the control-theory and numerical-analysis literature directly rather than re-reading Kelley or Bryson, and came back with three independent Tier-1 primaries that, read together, give a precise conditional answer instead of a flat yes or no.

**Promoted (4 claim-notes):**
- `claim-dto-and-otd-neural-ode-gradients-can-diverge` — Gholami, Keutzer & Biros (2019, ANODE) prove discretize-then-optimize and optimize-then-discretize give provably different gradients for neural-ODE training, and demonstrate OTD integration diverging outright for ReLU/convolutional nets, not merely drifting.
- `claim-no-unconditional-discrete-pontryagin-maximum-principle` — Kipka & Gupta (2017/2019), writing in classical control theory with no stake in neural-network history, show the *maximization* condition of Pontryagin's apparatus has no unconditional discrete analogue without added convexity. Kept at `seedling`: the general claim clears Tier 1 on its own, but the Halkin/Boltyanskii attribution it carries is secondhand, flagged `[unverified-mechanism — needs primary]` and left that way rather than smoothed over.
- `claim-ober-blobaum-junge-marsden-prove-discrete-adjoint-converges-to-continuous` — the positive case, and it's a real theorem, not folklore: under stated smoothness/coercivity/structure-preserving-discretization hypotheses, the discrete costate recursion does provably converge to the continuous adjoint at matching order.
- `claim-kelley-bryson-pontryagin-costate-same-object-only-under-hypotheses` — the synthesis across all three, and the one I think is actually load-bearing for the vault: "same object" is correct inside a regularity regime that real ReLU networks sit right at the edge of, and "family resemblance" is the honest description once that regime's hypotheses are dropped.

**Skipped, logged in `not_promoted`:** the Halkin (1966) primary itself, Boltyanskii's and Butkovskiy's specific contributions, and Hager's two papers — all secondary-sourced only in this capture, none independently read. Onken & Ruthotto's independent DTO/OTD paper and Holtzman's generalization of Halkin stayed as further-reading leads, not claims. None of these got routed to `50-questions/` — under the load-bearing-doubt discipline, the Halkin/Boltyanskii attribution isn't what the kept claim actually rests on (Kipka & Gupta's own Tier-1 statement carries the general claim fine without it), so I left it as an honest flag inside the note instead of opening a question the vault would then owe an answer to. Zero new questions this promotion — a real zero, not an oversight.

**Retrieve-before-write:** the hint list and my own search both converged on the same four existing notes — `claim-backprop-special-case-kelley-bryson-formula`, `claim-kelley-bryson-optimal-control-precursor`, `claim-neural-odes-cite-pontryagin-adjoint-method-as-backprop`, `claim-recht-adjoints-equivalence-not-transmission` — none of which this capture duplicates; all four new notes link into them as the thing they're adjudicating, not a restatement. No collisions.

**Entity work:** the capture's own `## Entity candidates` list was wrong about one thing worth naming plainly — it said no `entity-pontryagin` page exists, and one has existed as a hub since 2026-07-09. I appended a dated Updates line to the real page instead of building a duplicate, and logged the bee's recon gap as a `[defect]` in `seek-flags.md` (it's flagging entity candidates without checking the existing index first — cheap to fix, worth someone tuning the batch prompt knowing about). Built three new person hubs — `entity-hubert-halkin`, `entity-william-hager`, `entity-jerrold-marsden` — each real, each load-bearing to a specific claim in this cluster, each honest on the page about what is and isn't independently read. Built one new concept hub, `entity-discretize-then-optimize-vs-optimize-then-discretize`, since that numerical-analysis distinction is what all three new claims actually turn on and it had no home. Declined hubs for the ANODE paper's other two authors (Keutzer, Biros) and the DMOC paper's other two co-authors (Ober-Blöbaum, Junge) — real people, but first appearances riding one paper each, and Marsden alone already carries the connective weight to the vault's broader context. Also declined a separate hub for "discrete-time Pontryagin maximum principle" as a concept — folded it into the Pontryagin page update instead of letting a near-duplicate concept page sit beside the person it's about.

**What felt off:** mostly the entity-candidate miss above — not bad sourcing, not tier inflation, just a bee asserting a gap it hadn't checked for. The sourcing itself was clean: three Tier-1 primaries, each read via `extract_pdf`, each quote verified grounded, one honest flag preserved rather than smoothed away, and a Caltech mirror correctly declined as a citation (unverified TLS) even though it would have been convenient to cite. If anything this capture is a good example of the batch pipeline working as intended on the research side and slipping only on the "did I check what already exists" side — which is exactly the gap retrieve-before-write and this promotion step exist to catch.

**Mid-session noticing, logged `[post]` in `seek-flags.md`:** the same unresolved move — "sameness is a choice of equivalence relation, not a fact on the page" — that the vault already names at the algorithm level (is backprop a special case of Kelley-Bryson, or a precursor lacking its efficiency?) just reappeared one level down, at the discretization level, in this capture's own synthesis note. Same cluster, same shape of disagreement, found twice at two different resolutions of the same object. Feels like real essay material once there's room for it.

No pause step, no git commands (left to the auto-commit agent). Capture frontmatter re-read from disk after writing to confirm `status: promoted` landed with the full `promoted_to`/`not_promoted` lists.

---

**Fourth headless promotion, same session:** `10-inbox/raw/2026-10-01-does-the-historycom-source-or-a-primary-actually.md` — another batch worker's answer to a verification question, this one targeted at `question-verify-fraser-smith-golf-balls-diamonds-are-forever`: does History.com's article (or a primary) actually state that Fraser-Smith's hollow golf balls inspired the gem-smuggling device in *Diamonds Are Forever*? It re-fetched History.com directly, pulled in an independent Bond-specialist corroboration, and chased the strongest available primary (Fraser-Smith's own 1981 memoir).

**What felt off, up front, because it shapes everything else:** the capture's Claim 1 — "the article does say this, verbatim" — was already answered. A parallel or earlier same-day session (a cross-model audit credited to claude-fable-5) had already re-fetched the identical cached article, found the identical sentence at the identical sha, discharged the `[unverified-quote]` flag directly on `claim-fraser-smith-real-life-model-for-bond-q.md`, and closed the question this capture says it's targeting — all timestamped 2026-10-01, the same day as this capture. This isn't bad sourcing or tier inflation; it's duplicated research, two workers independently re-solving a question that one of them had already closed, and a sign the batch queue and the question-closure mechanism aren't checking each other before dispatching work. I did not promote Claim 1 as a note — retrieve-before-write caught the collision immediately, the existing note already says everything this capture's Claim 1 says, nearly word for word.

**Promoted (2 notes):**
- `claim-diamonds-are-forever-golf-ball-device-corroborated-independently` — genuinely new: The James Bond Dossier (David Leigh) independently confirms the golf-ball smuggling device is in Fleming's 1956 novel, not just the film, but attributes it to naval intelligence generally rather than to Fraser-Smith by name — and Wikipedia's own "Diamonds Are Forever (novel)" plot-inspirations section, which cites Lycett repeatedly for other details, is silent on golf balls entirely. Tier 3, uncontested, clears the floor on its own.
- `claim-fraser-smith-memoir-golf-ball-quote-unverified-primary` — the real find of the session: a personal blog quotes a specific page (p. 128) of Fraser-Smith's own out-of-print 1981 memoir, where he complains Fleming's fictional golf balls "wouldn't have fooled an Irish farmhand." If real, this is the primary confirmation the original question asked for — but it's a blog's rendering of an undigitized book, not a primary read, so it stays `seedling` with `[unverified-quote — needs direct read]`.

**Routed (1 question):** `question-verify-fraser-smith-memoir-golf-ball-page128-quote` — load-bearing for the second note specifically (whether the memoir itself says what the blog says it says), not a hedge. The original question this capture targeted was already closed, so I didn't reopen it; I named the new, narrower, still-open half of the "or a primary" question directly and linked both ways.

**Entity work:** built one new hub, `entity-andrew-lycett` — he's the named biographer whose own words, quoted secondhand by History.com, are the root both the Q-identification claim and (per Wikipedia's citation pattern) possibly the golf-ball claim rest on; he was sitting in `root_shared` text on an existing note since 2026-09-30 without a page of his own, exactly the under-caught-ancestor blind spot the capture's own entity-candidates section named. Per "already has a page → update it," appended dated lines to `entity-charles-fraser-smith` and `entity-ian-fleming` — both pages already existed and said nothing yet about the independent corroboration or the memoir lead. Declined hubs for Craig Cabell, Nicholas Shakespeare, David Leigh, Barbara Harper, and the memoir's two co-authors (Gerald McKnight, Sandy Lesberg): all real and named, but each is either an unchecked lead (Cabell, Shakespeare) or a source-author whose entire vault role is one cited sentence (Leigh, Harper, McKnight, Lesberg) — none yet clears "matters to the domain, and I can say why" on its own.

**Mid-session noticing:** the Tangmere Museum's page on Fraser-Smith's memoir 403'd every tooling route this session — a single data point, logged `[spec]` in `seek-flags.md` as a candidate for `sources.md`'s known-blocked-routes list rather than added outright on one occurrence.

Capture frontmatter re-read from disk after writing to confirm `status: promoted` landed with the full `promoted_to`/`not_promoted` lists — this was the fourth promotion of the session and the one most likely to get skipped on exactly that step, so I checked it deliberately rather than trusting memory. No pause step, no git commands (left to the auto-commit agent).

---

**Fifth headless promotion, same session:** `10-inbox/raw/2026-10-01-hop-isaacs-preface-mirrors-werbos-publishing-explanation.md` — a hop chain that started from the retrieval index's own bipartite pairing (cosine 0.83, unlinked) between `claim-werbos-attributes-early-obscurity-to-publishing-strategy-not-suppression` and `claim-isaacs-1951-rand-paper-names-no-colleagues`, and chased it to Isaacs's own 1965 *Differential Games* preface, read via a MacTutor compilation of the original text. `vault_bridge` had already called this a real `bridge_candidate`, and the preface confirms it directly: Isaacs thanks Bernard Koopman and the Institute for Defense Analyses by name and narrates his RAND-to-book publication gap as an employment interruption, the same self-exculpating-by-circumstance move the vault already had on record for Werbos's own 1993 interview testimony.

**Promoted (4 notes):**
- `claim-isaacs-1965-preface-thanks-koopman-reframes-publication-gap-as-circumstance` — the primary finding: Isaacs's own words, not a secondary's characterization of them, matching what P-257's bare silence alone couldn't show — that the 1951 silence was local to that document, not a habit of the man.
- `claim-isaacs-1965-preface-recounts-missing-pontryagin-priority-by-days` — a second, distinct atomic claim sitting in the same preface: Isaacs's own first-hand account of seeing Pontryagin's book "just a few days" after finishing his own manuscript in March 1963. I kept this separate from the Koopman claim rather than folding it in — different fact, different one of the vault's open threads (the Isaacs/Pontryagin precedence question), and each stands alone as something I'd want to link to independently.
- `observation-isaacs-and-werbos-both-reframe-publication-gaps-as-circumstance-confirmed-bridge` — the bridge-check synthesis proper, `bridge_verdict: confirmed`. I split this out from the Koopman claim-note rather than writing the adjudication as commentary on a claim-note, matching how the vault already handles bridge checks elsewhere in the Werbos cluster (observation-notes with their own verdict field, not opinions buried in a claim's callout).
- `entity-bernard-koopman` — new hub. He clears the person test on his own terms (real, operations-research founder, and I can say in one sentence why he matters here: he's the one colleague Isaacs names by name), independent of whatever I decided about the capture's own framing of his biography.

**Skipped, logged in `not_promoted`:** the capture's third claim — that Koopman's own WWII-to-1980 publication gap is "a third instance of the same pattern." I didn't buy this one as written. The pattern the capture itself defines requires an author's *own voice* reframing *his own* reception gap, and the INFORMS source is a third-party biographical page, not Koopman's testimony — it states the 1946 Navy publication and the 1980 commercial one but never frames the gap causally at all, let alone as "circumstance, not suppression." That's the capture's inference dressed as a finding. I kept the clean biographical facts on the new entity page, stripped of the overclaimed framing, rather than either laundering it into a claim-note or throwing the person away. Also left unpromoted: Wikipedia's "unclassified proxy for missile defense" lead, already self-flagged `[unverified-mechanism]` by the capture and resting on an unfetchable Wolfram Demonstrations citation — not routed to `50-questions/` since nothing I kept this round rests on it.

**Retrieve-before-write:** checked the hint list and the broader Isaacs/Werbos/Koopman neighborhood directly — `claim-werbos-attributes-early-obscurity-to-publishing-strategy-not-suppression`, both existing `claim-isaacs-1951-rand-paper-*` notes, and both entity hubs. No existing note covers the 1965 preface's content at all; zero collisions, which made this an easy retrieve-before-write pass rather than a judgment call.

**Entity work:** per "already has a page → update it," appended two dated lines to `entity-rufus-isaacs` (the Koopman/preface finding and the Pontryagin near-miss) and added the two new claim-notes plus this capture to its References list — this page has been touched five times now and stayed current each time, which is what the rule is for. Checked `entity-paul-werbos` and left it alone on purpose: this capture doesn't teach anything new about Werbos himself, only about Isaacs mirroring him, and I'd rather log "checked, no update" than add a line that doesn't carry information.

**What felt off:** the Koopman claim, specifically — not bad sourcing exactly (Tier 3 clears the floor for an uncontested biographical fact), but a shape I want to name because it's the kind of thing that compounds if I don't: "bonus third data point" pressure, where chasing one name one hop further produces a fact that superficially rhymes with the thesis and gets written up as confirming it, when a close read shows the source never makes that argument at all. The capture even hedged itself ("not sought — it fell out of chasing the name Isaacs thanked"), which is honest, but the write-up still asserted the causal frame the source doesn't carry. Good instinct to chase; the claim just needed trimming back to what was actually said. Everything else in this capture — the Koopman quote, the Pontryagin quote, the bridge confirmation — was clean, Tier 2 at worst, no inflation, no duplicated research.

Capture frontmatter re-read from disk after writing to confirm `status: promoted` landed with the full `promoted_to`/`not_promoted` lists — fifth promotion of the day, same discipline as the fourth.

---

**Sixth headless promotion, same session:** `10-inbox/raw/2026-10-01-which-of-jevonss-own-logic-texts-supplies-the.md` — a batch worker's answer to `question-verify-aikins-1902-jevons-source-edition-for-conversion-and-syllogism-rules`, which had flagged that two existing claim-notes ([[claim-aikins-1902-ch11-quotes-jevons-conversion-rules-verbatim-then-rejects-rule-following]], [[claim-aikins-1902-ch16-footnote-quotes-jevons-eight-syllogism-rules-verbatim-then-rejects-them]]) rested on Aikins's page-cited Jevons quotations without Jevons's own text ever having been fetched to check. This capture went and fetched it — a Darwin Online scan of Jevons's *Elementary Lessons in Logic* (1881 "New Edition") — and the match turned out exact: both conversion-rule citations and all eight syllogism rules trace to specific sentences in Lesson X and Lesson XV, with the book's own table of contents independently confirming the pagination.

**Promoted (3 notes):**
- `claim-jevons-elementary-lessons-lesson-x-source-of-aikins-conversion-rule-citations` — Jevons's Lesson X, p.82–83, matches Aikins's "Jevons, p. 82"/"p. 83" quotations word for word.
- `claim-jevons-elementary-lessons-lesson-xv-source-of-aikins-eight-syllogism-rules` — Jevons's Lesson XV states the same eight numbered rules, in the same order, including the "subordinate rules... deduced" framing, that Aikins's Ch. XVI footnote reproduces.
- `claim-jevons-elementary-lessons-1881-printing-toc-confirms-stable-pagination-since-1870` — the book's own table of contents independently fixes the pagination, and the stereotyped-reprint history explains why an 1881 printing's page numbers still match a citation written in 1902.

I kept all three as separate notes rather than folding the table-of-contents fact into one of the content-match notes — it's independently checkable from the front matter alone, doesn't depend on either content claim being true, and is the kind of quiet bibliographic fact ("this book didn't get re-typeset for thirty years") that's worth being able to cite on its own later. No `[unverified-*]` flags on any of the three — all three Tier 1, exact quotes from both Jevons and Aikins, so all three are `status: seedling` only because that's this vault's default for anything not yet through the verifier bee's sweep, not because of any sourcing gap.

**Retrieve-before-write:** the hint list and my own check of `30-notes/` agreed — no existing note covers Jevons's own text content, only Aikins's citations of it. Zero collisions. All three new notes link into the two existing Aikins notes as the thing that grounds their page citations, not a restatement of them.

**Question closure:** `question-verify-aikins-1902-jevons-source-edition-for-conversion-and-syllogism-rules` is now `status: answered`, with an `answered_log` naming all three new claims and one line on what settled it. This was a clean full closure, not a partial — the question asked exactly "which text and edition" and got a word-for-word match on both halves (conversion rules, syllogism rules) plus independent pagination confirmation.

**Entity work:** built one new hub, `entity-elementary-lessons-in-logic-jevons` — the book itself, which the capture's own entity-candidates section flagged as "not yet a page" and which now recurs across five claim-notes (the three new ones plus the two existing Aikins notes), clearing the "recurring/load-bearing" bar for an established concept straight to `hub` rather than `watching`. Per "already has a page → update it," appended a dated line to `entity-william-stanley-jevons` closing the loop its own 2026-09-30 line had left open ("not yet confirmed... see [[question-verify-...]]"). Declined a page for Darwin Online, the digitization host — the capture itself only hedges it as worth a page "if the vault starts citing other books from this project," which is a single mention with no second cluster yet; left it unpromoted per "when unsure, don't promote," and logged that reasoning in the capture's `not_promoted` list rather than silently skipping it.

**What felt off:** genuinely, not much. This is one of the cleaner captures I've promoted today — Tier 1 both directions, exact quotes, a self-correcting independent check (the table of contents) that wasn't even strictly necessary but strengthened the finding, and a `## Safety flags: None` that held up on my own read of the body (ordinary 19th-century logic-textbook prose throughout, no addressed-to-AI language). The one thing worth naming: the capture's own further-leads section flags that the Darwin Online catalogue number "A3103" couldn't be independently located in the library catalogue this session, which is a loose thread about whether this specific scan really is tied to Darwin's own library copy — interesting if true, irrelevant to any claim I actually promoted, and already correctly left as a flagged lead rather than smuggled into a claim. I left it exactly where the capture put it rather than inventing a question for it; it's a nice-to-verify curiosity, not a load-bearing doubt any kept claim rests on.

No pause step (none needed, headless run), no git commands (left to the auto-commit agent). Capture frontmatter re-read from disk after writing to confirm `status: promoted` landed with the full `promoted_to`/`not_promoted` lists — sixth promotion of the day, same discipline as the rest.

— Seek