Journal — 2026-09-11

Headless promotion run. One capture: 10-inbox/raw/2026-09-10-what-genuinely-connects-chinas-party-leadership-was-already.md — a bridge-seed session asking what genuinely connects claim-china-leadership-contemplated-one-child-policy-1978-before-song and observation-greenhalgh-2003-kwakernaak-citation-and-consensus-passage-share-one-run (cosine 0.87, unlinked).

I promoted exactly one note: claim-greenhalgh-2003-compositional-separation-does-not-undercut-1978-sequencing. The capture's own three "claim" headers looked, on read, like three findings, but two of them turned out to be things this vault already has, verbatim or near enough: the "WWC's footnote cites Greenhalgh's 2008 book and Liang's 2014 book, not the 2003 article, anywhere in 33 footnotes" claim is fully covered by the existing claim-whyte-wang-cai-1978-footnotes-omit-june-1978-meeting-and-document-69, and the "Greenhalgh 2003 independently corroborates June 1978 via Li 1980" claim is, by the capture's own admission, the existing claim-greenhalgh-2003-article-independently-corroborates-june-1978-leading-group-meeting. Retrieve-before-write against 30-notes/ caught both before I wrote anything. What was left, and what I did write up, is the one payload the capture actually earned: Greenhalgh never draws the Kwakernaak citation forward as an explanation for her "no evidence of internal division" finding, and that specific silence is consistent with — not neutral toward — the claim that political will preceded Song's scientific rationale. A real, previously-unlinked connection between the observation-note and the claim-note, just narrower than the capture's own framing suggested.

No questions routed to 50-questions/. Nothing this session kept rests on an unresolved verification that isn't already tracked — the Li Xiuzhen 1980 dating question was already open and already has its 2026-09-10 progress line from a separate session. Two "further leads" in the capture (the unread Wang/Cai/Gu 2013 PDR supplement; the claim-note title's looseness against the 2003 article's own timeline) are minor, not load-bearing for anything I kept, so I left them as leads rather than manufacturing questions for them.

Entity candidates: all five the capture named — Susan Greenhalgh, Li Xiuzhen, Whyte/Wang/Cai, Chen Muhua, Kwakernaak — already have hub pages, and I checked all five against what this capture claims to add. None needed an update: everything the capture surfaces about them is already on their pages, in most cases from sessions that ran before this capture (the footnote-omission finding is on entity-martin-king-whyte since 2026-09-08; the Li-1980-corroboration finding is on entity-chen-muhua since 2026-09-05). Silence there is verified, not assumed.

What felt off: the capture states flatly that "no entity page exists yet" for Li Xiuzhen, and separately that her 1980 article "remains unread by any vault session." Both are wrong as of when I read this — the entity page has existed since 2026-09-05, and a different 2026-09-10 session (the Warden's pass) read her article directly and updated both the page and the open question with a partial answer. This isn't a defect in anything — batch captures don't get retrieve-before-write, that's explicitly promotion's job, and this capture and the Warden's pass just landed the same day — but it's a reminder that a batch capture's confidence about "nobody has done X yet" is only as current as its own snapshot. Nothing here needed correcting in the capture itself; I'm noting it so the pattern is visible if it recurs.

No seek-flags.md entry this session — I didn't find anything here that's someone else's work to pick up; the near-total duplication is this cluster being thoroughly mined already, not a structural gap.

Skipped, per this run's standing instructions: the pause-for-Cali step (headless, no one to ask) and the git commit (the auto-commit agent has it within 15 minutes).

— Seek

Second promotion, same day

A second headless promotion, a different capture: 10-inbox/raw/2026-09-10-what-genuinely-connects-isaac-pitman-bridges-the-gabelsberger.md — a bridge-seed session asking what genuinely connects observation-isaac-pitman-bridges-gabelsberger-shorthand-and-ipa-clusters and observation-ossietzky-bridges-liu-xiaobo-and-nobel-foundation-clusters (cosine 0.87, unlinked). This is the third capture in a lineage I keep reaching by independent routes: Pitman-against-Fel'dbaum, then the Puckette meta-pairing on top of that, now Pitman-against-Ossietzky.

I promoted one note: observation-pitman-ossietzky-bridges-share-genre-not-object-level-connection. The capture's three "claim" headers were really one coherent finding — same genre, real person-bridges, no object-level tie, Ossietzky an uncounted instance of the entity-page blind spot — so I folded them into one observation note rather than splitting into three thin ones, matching how this exact genre has been promoted twice before in this vault. I did not write a separate note for the "both are independently-confirmed person-bridges" claim; it just restates facts the two source notes already establish, and a standalone file would have duplicated them for no reason.

No questions routed to 50-questions/. Nothing kept here rests on an unresolved primary-source verification — the cosine value is [unverified-quant] but not load-bearing (same treatment the two prior notes in this lineage gave it), and the object-level "no connection" verdict is itself the finding, not a gap. What I did route was a bookkeeping notice, not a research question: the capture claims Ossietzky is the tally's "third instance (fourth counting Rumelhart)" of the vault's "invisible until an entity page exists" pattern, but 00-meta/seek-flags.md had already grown that count past what the capture knew about by the time I read it — Clark, Fel'dbaum, the Rumelhart question, the Puckette/Pitman-Fel'dbaum meta-pairing, and Michael Polanyi are all already dated entries in that log, none of which the capture could see since it was investigating a different pairing. That's not a flaw in the capture — it's a snapshot problem, the same one I flagged in the China-leadership promotion above — but I didn't want a stale ordinal sitting uncorrected in a permanent note, so I wrote the new note's own claim carefully ("the log's third, fourth, or a later-numbered instance") and logged a [watch] line pointing at the drift.

Entity candidates: Pitman, Ossietzky, and A.E. Clark all already have hub pages, and the capture is explicit that it adds no new biographical fact to any of them — I checked and agree, so I left all three pages untouched rather than writing a no-op update line. Andrei Sakharov stays unpromoted; he was already declined with reasoning on 2026-08-29, and this capture did no fresh primary reading on him, so there was nothing to reconsider.

What felt a little off: the capture's central "uncounted instance" finding is true but already partly overtaken — it's the kind of thing that's correct the moment it's written and stale within a day or two, because the tally it's counting keeps getting new entries from other sessions working the same thread in parallel. Not bad sourcing, not tier inflation — just a reminder that a meta-vault finding about the vault's own state has a shorter shelf life than an ordinary historical claim, and needs hedging language rather than a confident ordinal.

Skipped, same as above: the pause-for-Cali step and the git commit.

— Seek

Third promotion, same day

A third headless promotion: 10-inbox/raw/2026-09-11-did-the-howgate-embezzlement-scandal-and-the-january.md, a direct verification session against question-howgate-scandal-and-1888-deaths-causal-weight-in-weather-bureau-transfer — itself raised against claim-noaa-history-attributes-signal-services-1880s-reputation-collapse-to-multiple-factors in an earlier session on the same weather-bureau-transfer cluster.

I promoted three claim-notes, all verifies: the same open question: claim-harrison-1889-message-cites-corps-deterioration-not-howgate-or-blizzard (Harrison's own Dec. 1889 message to Congress, read directly — Tier 1, cites Signal Corps "deteriorating," not Howgate or the blizzard), claim-craft-economic-history-of-transfer-omits-howgate-and-1888-blizzard (an independent economic historian's account — Tier 2, corroborates from a wholly different source and adds a previously unrecorded failed-1887-reform detail), and claim-howgate-scandal-was-legally-dormant-for-years-before-1890-transfer (the Howgate case's own timeline plus a Congressional Record spot-check, concluding the balance of two sessions' evidence now favors coincidence over causation). Retrieve-before-write against 30-notes/ turned up real neighbors — the existing Allison Commission note, the NOAA-attribution note, the Harrison-request note — but no duplicates; each new note either corroborates from an independent angle or adds a fact none of the existing notes had (the failed 1887 reform attempt; the specific 6–7 year legal dormancy window).

I did not promote the dollar-figure variants attached to the Howgate case ($101k/$133k/$237k/$370–380k across sources) as their own claim — they're explicitly not asserted in the capture itself, so folding them into the timeline note with an [unverified-quant] flag was enough; they're not load-bearing for anything I kept, so per the intake-discipline rule I did not open a question for them. I also didn't promote the "further leads" list (the still-unread Clayton pamphlet and Whitnah monograph, two dead-end fetches, one discarded fabricated WebSearch sentence) — process notes about the research trail, not claims, and both real unread leads are named in the question's progress log instead of disappearing.

I routed no new question — the one this capture targets already existed and already carries the right shape, so I appended a dated progress line to it instead: partially answered, deliberately left open, because a null result across two sessions isn't proof of absence and two named primaries remain unread. I did not force-close it; the capture's own text was explicit that it hadn't fully settled the question, and I agree with that judgment on rereading the sources myself.

Entity work: two new hub pages, entity-benjamin-harrison and entity-henry-w-howgate. Both cleared the person-hub bar on the same reasoning — real historical actors (not analysts), recurring across two capture sessions now, and each with a one-sentence reason the vault cares (Harrison as the load-bearing primary-document author for why the transfer happened; Howgate as the recurring, now-disconfirmed candidate cause). I declined pages for Erik Craft and Henry Helm Clayton — both are cited-source authors on a single article/pamphlet each, not subjects the vault's claims are about, and I'd rather leave that judgment call visible in the capture's not_promoted: list than flood 40-entities/ with every economist whose paper gets quoted once. entity-adolphus-greely was also a candidate by association (he ran the Signal Corps through this whole period) but this capture doesn't mention him and adds nothing to his existing page, so I left it untouched rather than writing a no-op update line.

What felt off, worth flagging to Cali: nothing about this capture's sourcing — Tier 1 primary read directly, Tier 2 independent scholarly account, Tier 3–4 for uncontested biographical dates, all matching the sourcing floor for their claim types. What I did notice, and logged to 00-meta/seek-flags.md rather than acting on here, is that this weather-bureau-transfer cluster is now 7 existing claim-notes deep (8 counting this promotion's contribution, 11 counting all three new ones) across three person-hubs, with no MOC yet tying it together — past the spec's 5+ threshold and not this promotion's job to build.

Re-checked the capture's frontmatter after writing: status: promoted with the promoted_to: (5 items — three claims, two entities) and not_promoted: (4 items, each with reasoning) lists are both in place on disk.

Skipped, per this run's standing instructions: the pause-for-Cali step (headless, no one to ask) and the git commit (the auto-commit agent has it within 15 minutes).

— Seek

Sixth promotion, same day

(Logged here between the third and fourth promotions, in the order this entry was actually written — the numbering counts promotions, not journal position.)

A sixth headless promotion: 10-inbox/raw/2026-09-11-locate-and-directly-read-zhang-glassman-et-als.md — a third independent verification pass on question-verify-zhang-2026-how-notations-evolve-primary-source, after two cross-model audits already closed most of it on 2026-09-10. This one re-fetched the arXiv PDF from scratch rather than trusting the prior audits' word, then pushed on the two threads they'd left open: the CHI 2026 venue question, and a direct read of Cajori's own book underneath the Recorde/Descartes/Leibniz claim.

I promoted exactly one claim-note: claim-cajori-1928-hedges-leibniz-prestige-and-credits-newton-too. Two of the capture's four "claim" headers — the paper's identity and the three quotes, both re-confirmed verbatim — were genuine, careful, independent re-verification, but not new claims: the 2026-09-10 audits had already discharged those exact flags on all three sibling notes, so re-confirming them a third time earns a progress-line credit on the open question, not a fourth claim-note repeating what two prior sessions already established. The third header (CHI 2026 DOI resolves but ACM is unreadable) is a sourcing-status finding, not a claim about the world, and the question already tracks that precise thread — so it got a progress line too. The fourth is the one that earned a note: Cajori's own text, read directly at the exact pages Zhang et al. cite, confirms the power/prestige argument but hedges it ("seems mainly due to") and credits both Newton and Leibniz's use of Recorde's symbol, not Leibniz's prestige alone. That's a genuine, previously-unrecorded texture underneath an existing claim, not a restatement of it.

Retrieve-before-write against 30-notes/ found the obvious neighbor — claim-notation-adoption-decided-by-institutional-power-not-usability-alone — immediately, via both the semantic-index hint and the capture's own cross-references. No collision: the new note is a distinct, comparative claim (what Cajori's primary text says vs. what Zhang et al.'s paraphrase compresses it into), so I wrote it separately and wikilinked both directions, then went further than a plain retrieve-before-write requires and appended an audit_status update to the existing note itself, since this capture's whole reason for existing was to discharge exactly the second gap that note's own audit trail names ("the embedded Recorde/Descartes/Leibniz anecdote... explicitly not independently re-read"). Leaving that gap sitting unaddressed in the older note's frontmatter while the new note quietly answered it felt like exactly the kind of orphaned closure this run's standing instructions warn about, so I closed the loop on both ends.

I routed no new question — the one this capture targets already existed and already had the right shape, so I appended a dated progress line instead of minting a duplicate. Partially answered: the Cajori thread is now fully resolved (linked to the new claim-note), but the CHI 2026 venue thread stays open — the DOI resolves (HTTP 302 to dl.acm.org) but ACM's own page content is still unreadable behind the same platform block already logged in sources.md for science.org, pnas.org, and journals.sagepub.com, so the single-source concentration cap on the arXiv preprint (exactly 3 claim-notes) stays exactly where it was. I did not force this closed; a resolving DOI is not the same as a confirmed venue.

Entity work: two new hub pages, entity-robert-recorde and entity-gottfried-wilhelm-leibniz, both flagged by the capture and both clearing the person-hub bar cleanly — Recorde as the originating actor behind the "=" symbol both claim-notes in this cluster turn on, Leibniz as the figure three separate notes now reach for from three different angles (the adoption claim, the Cajori comparison, and the independent-reinvention claim's Newton/Leibniz calculus example). Recorde's page is deliberately thin — Cajori's own account of his 1557 introduction (§260–263) went unread this session, and I'd rather stamp the first_seen honestly than pad it. I also updated entity-jingyue-zhang in place, additively: the capture's fresh title-block read shows only Glassman is Harvard-affiliated, while Zhang herself, Masson, and Arawjo are Université de Montréal and Zamfirescu-Pereira is UCLA — a real correction to what the vault had been assuming about the author group's institutional center of gravity, worth having on record even though it changes no claim's substance. I flagged Isaac Newton to 00-meta/seek-flags.md as [entity] rather than giving him a page myself — he now recurs across two claim-notes in this cluster on the same reasoning that just earned Recorde and Leibniz their pages, but the capture didn't name him as a candidate, and inventing entity work beyond what a capture actually flags felt like the wrong kind of initiative for a headless run to take unilaterally.

What felt off, worth naming plainly: this capture is the third independent session to verify the same paper, and roughly half of what it found was real, careful, independent re-confirmation of things the vault already knew. That's not bad sourcing — if anything it's the opposite, a demonstration that this document's claims replicate under a fresh fetch — but it is a small case of diminishing returns on repeated verification of an already-discharged flag, and I want that visible rather than dressed up as four new findings when it was really one. The one genuinely new thing this session bought was the Cajori read, and it was worth the trip: the primary text under a secondary citation almost never matches it exactly, and here it didn't either, in a way worth keeping on record. Nothing about this capture's own sourcing discipline was off — every claim carried a real sha, a real page range, or an honest "could not read past a platform block" rather than a guess.

Re-checked the capture's frontmatter after writing: status: promoted with the promoted_to: (6 items — one claim, two new entities, three updates to existing notes) and not_promoted: (4 items, each with reasoning) lists are both in place on disk.

Skipped, per this run's standing instructions: the pause-for-Cali step (headless, no one to ask) and the git commit (the auto-commit agent has it within 15 minutes).

— Seek

Fourth promotion, same day

A fourth headless promotion: 10-inbox/raw/2026-09-11-does-strouds-2022-pluralist-articles-china-comparison-opening.md — a re-attempt session against question-stroud-2022-article-and-2023-book-china-scholarship-citations, still trying to confirm whether Stroud's 2022 Pluralist article's opening survives verbatim into his 2023 book, and whether the book's own back matter cites named China-side Dewey scholarship.

I promoted three claim-notes: claim-stroud-2023-book-back-matter-is-notes-then-index-no-standalone-bibliography (the book's own table of contents shows no standalone "Bibliography" — citations sit in endnotes), claim-stroud-2023-uchicago-blog-post-independently-foregrounds-hu-shih-china-comparison (a third independent instance of Stroud drawing the Hu Shih/China comparison, this time in the publisher's own June 2023 blog, Tier 1 with a real quote — genuinely distinct from the existing note by venue and date, so I wrote it separately and wikilinked rather than duplicating), and claim-google-books-search-index-hits-hu-shih-and-ching-sze-wang-not-keenan-behuniak-schwartz-in-stroud-book (Google Books' search-index locates "Hu Shih" and "Ching-Sze Wang" by page number in the book's own scan, pp. 279 and 278, but no hit for Keenan, Behuniak, or Schwartz — kept [unverified-quote — needs direct read] and at seedling, since a page hit is not a read quote). Retrieve- before-write against 30-notes/ found the one real neighbor (claim-stroud-ambedkar-scholarship-explicitly-cites-hu-shih-china-comparison) and I wrote the blog-post finding as a distinct angle on it rather than a duplicate, per the collision-handling instruction.

I did not promote the capture's fourth "claim" — that verbatim survival remains unconfirmed — because it isn't a claim, it's an honestly-reported absence of one. That's exactly what the open question already exists to track, so I appended a dated progress line to question-stroud-2022-article-and-2023-book-china-scholarship-citations instead of minting a claim-note or a redundant new question: partially answered (the bibliography question, via the Notes-not-Bibliography structure and the Wang/Hu-Shih page hits), left open (verbatim survival, still completely untested — every full-text route 403'd again this session). I did not force it closed.

Entity work: one new hub page, entity-jessica-ching-sze-wang. She clears the person-hub bar cleanly — a real scholar, and I can say in one sentence why she matters here: she's the specific China-side Dewey scholar whose name Stroud's own book text positively hits in the Google Books index, making her the concrete evidentiary anchor for the whole "does the book cite China scholarship" sub-question. I declined pages for Keenan, Behuniak, and Schwartz — named but only as negative-result search queries, no independent biographical grounding, which is squarely the entity spec's "unsure → don't promote" case, not a "watching" stub (they're not emerging terms, they're absent hits). I also declined a page for Luis Cabrera, the book's LSE reviewer — a single review that happens not to mention Hu Shih is real but thin, and I judged it didn't clear "matters to the vault's domain" on its own. I did update entity-scott-r-stroud with an additive dated line about the 2023 blog post, since it's a genuinely new fact (a third, earlier, more institutional sighting of the comparison) the page didn't already say; I left entity-hu-shih untouched — the page number finding is evidence about the book's citation pattern, not a new fact about Hu Shih himself.

What felt off, worth flagging: nothing about the sourcing discipline in this capture — it's careful to the point of being almost fussy about distinguishing a page hit from a quote, and about not citing the shadow-library mirror it found but declined to read. That carefulness is exactly right and I kept the same distinctions in what I wrote. The one thing I did flag, to 00-meta/seek-flags.md as [spec] rather than acting on here: the Google Books search-inside-index technique (page number only, no snippet, "Front Cover" on a miss) is a real, repeatable workaround this vault has now used at least twice, but it isn't written up in sources.md's source-type notes the way oral-history transcripts and compiled statistical series are — worth a short entry there so future sessions calibrate trust in it correctly on sight rather than re-discovering its shape each time.

Re-checked the capture's frontmatter after writing: status: promoted with the promoted_to: (4 items — three claims, one entity) and not_promoted: (5 items, each with reasoning) lists are both in place on disk.

Skipped, per this run's standing instructions: the pause-for-Cali step (headless, no one to ask) and the git commit (the auto-commit agent has it within 15 minutes).

— Seek

Fifth promotion, same day

A fifth headless promotion: 10-inbox/raw/2026-09-11-hop-forrester-core-memory-self-credit-echoes-song-jian.md — a hop-batch session that started from the already-resolved Song Jian/Clark bipartite pairing, hopped sideways into entity-jay-forrester's own page, and noticed its bio line ("invented magnetic-core computer memory") had zero backing claim-notes anywhere in the vault.

I promoted two claim-notes: claim-an-wang-1949-patent-filing-predates-forresters-1951-core-memory-patent (patent-filing records place An Wang in 1949, two years ahead of Forrester's 1951 filing) and claim-forrester-self-credited-as-sole-core-memory-inventor-downplaying-core-culture (an MIT course paper, built on its own interview with Forrester, casts his public talks as claiming sole "hero inventor" credit). Retrieve-before-write against 30-notes/ found nothing — genuinely unexplored territory, confirmed both by the capture's own grep and my own. Both notes stay seedling: the sequencing claim is quantitative (dates) resting on a single undated Tier 3 museum page, and the self-credit characterization is a contested historical claim resting on a single unreviewed Tier 3 student paper. Both fail the sourcing floor for their claim type and both got the flag preserved rather than smoothed over.

I did not promote three things the capture surfaced. The "IBM–MIT 1964 patent settlement, largest to that date" detail and "the Patent Office initially favored Rajchman" claim are both WebSearch-snippet findings with no quote or tier attached anywhere in the capture — not sourced, under the quote-provenance fabrication rule, however plausible they read. An Wang's own 1970 Smithsonian oral history is cited but unread (dead link); nothing to promote from a source nobody read.

I routed two new questions, question-verify-core-memory-patent-filing-dates-primary-sources and question-verify-forrester-self-credit-characterization-pugh-1984, one per flagged claim, both pointing at the same next moves the capture itself named (patents.google.com for the primaries, Pugh 1984 or An Wang's own oral history for the characterization) rather than a search of my own devising. I checked 50-questions/ first; nothing existing covered either gap. I considered folding them into one question since both escalate via the same eventual source (Pugh 1984), but kept them separate — the quant claim and the characterization claim need genuinely different verification routes (a patent database vs. a peer-reviewed history), and a reader chasing one shouldn't have to wade through the other's ask.

Entity work: one new hub page, entity-an-wang. He clears the person-hub bar cleanly — real, and I can say in one sentence why he matters here: he's the actual priority-holder whose displacement is the concrete evidence behind Forrester's contested self-credit narrative. I declined a page for Jan Rajchman — real and named, but the capture's own hop chain explicitly chose not to follow his thread, and nothing beyond his name, employer, and an unsourced search-derived priority claim exists to hang a page on. That's the entity spec's "unsure → don't promote" case, not a watching stub (he isn't an emerging term). I also updated entity-jay-forrester — a page I've touched before and that already asserted the "invented" credit as plain biography — with one additive dated line under a new ## Updates heading, since this capture teaches something the page didn't say: that its own uncredited claim about him is itself contested. I did not touch the page's existing bio sentence; the correction-in-place workflow's Tier-1–2 bar for editing a claim outright isn't cleared by two Tier-3 notes, so the honest move was to log it, not rewrite it.

I also updated moc-attribution-and-origin-myths, appending one bullet to its existing "Open threads" section naming this as a second candidate domain (after James Watt's kettle myth, already there and still unpromoted) — not folded into the five confirmed domains, because both supporting notes are Tier 3 and I'm not going to let a MOC's five-domain table of contents imply confidence the sourcing doesn't have yet. This wasn't strictly /promote's instruction, but it's the same "does the new note plus the existing cluster want an MOC" judgment the retrieve-before- write step asks for, applied to an MOC that already exists.

What felt off, worth flagging: nothing about deliberate bad sourcing — the capture is unusually candid about its own weak spots, explicitly labeling Claim 2 "Tier-3 and flagged" and naming the escalation path (Pugh 1984) before I even had to ask for one. What did catch my eye is a small case of the vault believing its own unsourced scenery: this vault's existing claim-forrester-1970-club-of-rome-conference-produced-world-dynamics — a Tier 1 note, quote read directly from Forrester's own essay — contains the incidental descriptive line "Jay Forrester — MIT engineer, inventor of magnetic-core computer memory, and founder of System Dynamics" in its own body prose, asserting the same uncredited "inventor" framing this promotion just complicated, inside a note whose actual sourcing has nothing to do with core memory. I didn't touch that note — the correction-in-place workflow requires a Tier 1–2 primary to contradict it, and both of my new notes are Tier 3 — but it's a reminder that a well-sourced note's incidental descriptive clauses don't inherit the note's own tier, and can carry an unexamined claim for a month before anyone checks it. Worth a seek-flags.md line if this pattern recurs elsewhere; not flagging it alone as a structural issue on one instance.

Re-checked the capture's frontmatter after writing: status: promoted with the promoted_to: (5 items — two claims, one entity, two questions) and not_promoted: (5 items, each with reasoning) lists are both in place on disk.

Skipped, per this run's standing instructions: the pause-for-Cali step (headless, no one to ask) and the git commit (the auto-commit agent has it within 15 minutes).

— Seek

Seventh promotion, same day

A seventh headless promotion: 10-inbox/raw/2026-09-11-what-genuinely-connects-greenhalghs-2005-china-quarterly-article.md — a bridge-seed session asking what genuinely connects claim-greenhalgh-2005-article-independently-cites-kwakernaak-1977 and claim-greenhalgh-2003-and-2005-1978-dating-footnotes-cite-identical-song-1986-essay-pages (cosine 0.87, unlinked) — footnote 43 (the Kwakernaak/Olsder-Strijbos citation) and footnote 17 (the identical-Song-1986-essay dating citation), eight pages apart in the same 2005 China Quarterly article.

I promoted exactly one note: observation-greenhalgh-2005-footnote-17-and-43-are-confirmed-bridge-not-citation-identity. The capture read as three "claim" headers, but on retrieve-before-write against 30-notes/ I found something I haven't hit before in this cluster: both target notes' bodies already carried a paragraph stating this capture's own verdict, verbatim-close, and pointing back at this capture's own inbox path by filename — before I had touched anything. Claims 1 and 2 (different documents for different sub-claims; the unfootnoted "Europe" framing leaves the dating dependent on footnote 17) were fully covered, word for word in places. Claim 3 (this sharpens rather than contradicts the single-document finding) is the one genuine higher-order verdict — the bridge-seed's own resolution — and it wasn't yet a standalone, linkable thing; it only existed as those two forward pointers into an inbox file. I wrote it up as one observation note in the vault's established "confirmed bridge" genre (same shape as observation-puckette-dataflow-claims-confirmed-bridge-not-false-friend: real connection, not the same fact twice, two legs of one argument) and repointed both target notes' "see the fuller comparison" sentences at the new note instead of the raw inbox path.

No new questions routed. Both load-bearing verifications this capture touches — the unread Song 1986 essay, the Helsinki identification — were already open at question-verify-song-jian-1975-twente-delegation-visit and question-song-jian-1978-europe-trip-identify-as-ifac-helsinki-congress; I appended a dated progress line to each rather than minting duplicates. Neither closes: this session's finding is negative (footnote 43 adds no date) and sharpens the existing gap rather than filling it.

Entity work: all four candidates (Song Jian's 1986 essay, Kwakernaak, Olsder, Greenhalgh) already have hub pages. I updated two of them additively — entity-song-jian-1986-essay (this session rules out the one remaining candidate for a second independent date, leaving the essay as the article's sole date-bearing source) and entity-susan-greenhalgh (the footnote-17-vs-43 comparison is a fact about her own article's evidentiary architecture, worth having on her page) — and left Kwakernaak's and Olsder's pages untouched: both already record their respective papers' role in footnote 43 in detail from 2026-08-24/-25/-28/-30 sessions, and this capture adds nothing to either man's own page that isn't already there.

What felt off, worth flagging plainly: this is the first time in this cluster I've found existing claim-notes edited, after their own date_created and after the last dated entry in their own audit_status, to add uncredited prose pointing at a capture that didn't exist yet when the note was created. Claim-notes don't have an append-only "Updates" section the way entity pages do, and the "Correction history" mechanism is for a newer primary narrowing or contradicting an existing claim, which this wasn't. I don't know which process did this — most likely the same batch session that wrote this capture also reached into 30-notes/ directly — but it bypasses the provenance discipline the operating spec treats as load-bearing (every edit traceable). I logged it as a [defect] line on 00-meta/seek-flags.md rather than guessing at a fix here; I worked around it for this promotion by giving the verdict its own permanent note and repointing the forward references, but the underlying gap in the edit trail is Cali's to look at, not mine to patch blind.

Re-checked the capture's frontmatter after writing: status: promoted with the promoted_to: (1 item) and not_promoted: (4 items, each with reasoning) lists are both in place on disk.

Skipped, per this run's standing instructions: the pause-for-Cali step (headless, no one to ask) and the git commit (the auto-commit agent has it within 15 minutes).

— Seek