Journal — 2026-10-12
Headless promotion session. Promoted one capture:
10-inbox/raw/2026-10-09-what-genuinely-connects-carl-von-ossietzky-person-bridges.md
— the Ossietzky/Polanyi cosine-0.88 bridge investigation.
What I wrote. Two notes, not four. The capture's first three claims (no
object-level connection between Ossietzky and Polanyi; the cosine proximity
explained by shared write-up genre; vault_bridge's own re-run anchoring on
an unrelated surname pattern) are one finding wearing three headings — a
result, its caveat, and the instrument reading that produced it — so I
folded them into a single observation note,
observation-polanyi-ossietzky-bridges-share-genre-not-object-level-connection.
It keeps the [unverified] flag on the no-connection claim and stays at
status: seedling for exactly that reason; the search was genuinely
exhausted (nomination-archive nominator lists, AAC/SPSL rescue-body
accounts) and there's no further specific primary to point a question at,
so I didn't manufacture one for 50-questions/ just to have somewhere to
put the doubt. The fourth claim — Polanyi's twelve Nobel nominations,
1959–1974, never a win, read straight off the Nobel Foundation's own
nomination archive with a verified quote and sha — is a genuinely separate,
solidly Tier-1 fact with nothing flagged, so it got its own note:
claim-michael-polanyi-twelve-nobel-nominations-1959-1974-never-won.
What I didn't promote, and why. The two further-reading leads (the
Hartl article on Polanyi and freedom of science; the Hayek/Mont-Pelerin
1974-Nobel angle) stayed leads — one needs a primary read before it's
anything, the other turned out to be already sitting on
entity-friedrich-hayek.md almost verbatim, so writing it again would have
been noise. The "the vault's own bridge-tally is scattered with no index"
observation is real but not new: I counted at least seven prior log lines
on exactly this, including an explicit 2026-10-04 warden-pass declining to
re-raise it because it's already routed to Cali as a pending proposal. Adding
an eighth echo would have been the kind of noise this discipline exists to
prevent, so I left it out and said so in the capture's not_promoted: list
instead of in the flags ledger.
Of the capture's four entity candidates: Michael Polanyi and Friedrich Hayek
already have hub pages. Hayek's needed nothing — the 1974 Nobel fact was
already there. Polanyi's got a new dated line for the nominations finding.
Hilde Walter and Hellmuth von Gerlach — the actual organizers of Ossietzky's
1935–36 nomination campaign — I declined to promote to hub pages. The
capture names them, but only as background inside the [unverified]
search-exhaustion claim, with no citation attached to their own role. That's
a real gap in the capture, not a small one: if their organizing work is
ever going to anchor anything, it needs its own primary source first, not a
hub built on an unsourced aside. The "Nobel nomination archive" entity
candidate I redirected to a sources.md addendum instead of a 40-entities
page — it's a routing fact (this one path resolves where the rest of the
domain 403s), which is exactly the shape sources.md's known-blocked
catalog already exists to hold.
One new thing worth flagging, one thing not worth re-flagging. I logged
a [watch] line in seek-flags.md for the sharpest concrete instance yet
of vault_bridge anchoring on a bare surname-token pattern — asked directly
about Ossietzky and Polanyi, it handed back Ludwig von Bertalanffy and
Leopold von Ranke, two men who share nothing with the query but "von
[German surname]," while missing Polanyi entirely. That's new evidence, not
a repeat. I did not add a ninth line about the scattered-tally problem,
for the reason above — the ledger already has the thread, and Cali already
has the decision in front of her.
What felt off about the capture. Nothing structurally — sourcing was clean, the Tier-1 nomination-archive fetch was properly sha'd and quote-checked at capture time, and the capture was honest about its own search limits rather than overclaiming a negative. If anything, the capture's strongest instinct was restraint: it explicitly declined to add a wikilink between the two bridge notes, the same restraint I kept. The one soft spot is the unsourced mention of Walter/von Gerlach, which I've left out rather than let ride on vibes.
Promoted → two notes, zero questions opened (none of the flags were
load-bearing enough to warrant one), one entity page updated, one
sources.md addendum, one seek-flags.md watch line. Capture marked
status: promoted with full promoted_to:/not_promoted: lists; not
deleted.
— Seek
Second promotion pass, same day
A second headless promotion session. Promoted
10-inbox/raw/2026-10-09-what-genuinely-connects-liang-zhongtang-proposed-a-two.md
— the bridge-seed check between
claim-liang-zhongtang-1979-two-child-proposal-and-1985-yicheng-pilot and
claim-whyte-wang-cai-and-greenhalgh-2003-converge-on-same-three-stage-timeline
(cosine 0.88, unlinked).
What I wrote. Three notes, matching the capture's own three ## Claim:
sections — each one independently true or false of the other two, so I
didn't fold them. observation-greenhalgh-2003-bridges-liang-1979-yicheng-and-three-stage-timeline-seed-notes
records the clean half: a note the vault already had on file
(claim-greenhalgh-2003-pdr-article-independently-names-liang-zhongtang)
turns out to sit directly inside the overlap of both seed notes and had
simply never been linked to either — the "proof already in the vault, just
not pointed at the question" shape. claim-whyte-wang-cai-cite-yan-wei-li-zhang-not-liang-for-yicheng-pilot
records the messier half: Whyte-Wang-Cai's essay opens its whole argument
through the Yicheng pilot but reaches it through Yan Wei & Li Zhang's 2014
re-examination, never through Liang himself — a citation-gap pattern this
cluster has now logged more than once. claim-whyte-wang-cai-and-liang-disagree-on-yicheng-pilot-start-date
keeps the one live doubt: the two sources disagree by five years (1980 vs.
1985) on when the pilot actually began, and nothing in the capture resolves
which is right, so that note stays seedling with an
[unverified-historical] flag carried forward honestly rather than
split the difference.
What I routed. One new question, question-verify-yicheng-pilot-start-date-yan-wei-li-zhang-2014 — the date conflict is load-bearing for how the vault describes the pilot going forward, and there's a single, named, specific document (Yan Wei & Li Zhang's 2014 China Journal re-examination, paywalled, not yet read by any session) that would likely settle both the date and the authorship-credit question at once. I checked the two existing Liang-translation questions first and neither covers this — this is a date-and-citation doubt, not a translation doubt — so a new question was the right call, not a third hedge stapled onto an unrelated thread.
What I skipped, and why. The capture's own top-level "the bridge is
real, through both halves" framing isn't a fourth claim — it's the verdict
the three notes above collectively constitute, and writing it separately
would have been the same finding twice. Yan Wei and Li Zhang went in as
named mentions inside the two Yicheng notes, not as entity stubs: they're
co-authors of one single, unread, paywalled paper, which is a mention under
Cali's 2026-10-01 ruling, not a hub, no matter how load-bearing that one
paper turns out to be for this cluster. Two further-reading leads from the
capture (whether the Liang 1985 book chapter is the written form of the
1979 Chengdu proposal; whether White's or Scharping's general histories
mention Yicheng or Liang) stayed leads — neither was read this session,
and a lead that wasn't followed doesn't earn a claim-note just because it's
interesting. All four left-behind items are named in the capture's
not_promoted: list, not just dropped silently.
Entity pages. Three already-existing hubs — Susan Greenhalgh, Liang
Zhongtang, Martin King Whyte — each got one dated, additive ## Updates
line and a matching ## References entry pointing at whichever of today's
three notes concerns them; none needed anything rewritten. No new entity
page: the only candidates without existing pages (Yan Wei, Li Zhang) didn't
clear the hub bar, per above.
What felt off about the capture. Less than usual, and worth saying so
plainly. Sourcing was genuinely clean — all three primaries re-fetched this
session with matching shas, the one Tier-3 source (Sixth Tone) correctly
held to Tier 3 rather than inflated, and the capture was explicit about what
it couldn't check (the paywalled Yan Wei/Li Zhang article) instead of
quietly treating WWC's secondhand citation of it as settled. If I have a
complaint, it's a small one: the capture's "Further leads" section bundles
one real unresolved thread (the 1985 book-chapter identity question) with
two that are really just "didn't check this," and I had to read the whole
capture closely to tell which of its loose ends were load-bearing enough
for a question versus just background texture for not_promoted:. Nothing
here rises to a seek-flags.md entry on its own.
Promoted → three notes, one question routed, zero collisions requiring a
skip (the one apparent overlap, with the existing Greenhalgh citation note,
turned out to be the actual finding rather than a duplicate), three entity
pages updated additively. Capture re-read from disk after marking — confirmed
status: promoted with full promoted_to:/not_promoted: lists landed
correctly; not deleted.
— Seek
Third promotion pass, same day
A third headless promotion session. Promoted
10-inbox/raw/2026-10-09-what-genuinely-connects-whyte-wang-cais-2015-china.md
— a bridge-seed check between
claim-whyte-wang-cai-cite-liang-2014-books-not-2009-essay and
claim-greenhalgh-2003-pdr-article-independently-names-liang-zhongtang
(cosine 0.88, unlinked). Near-identically titled to the capture I promoted
earlier this week (2026-08-19, a different pairing of the same
Whyte-Wang-Cai note against the Olsder dating dispute) — the capture itself
flagged this and I confirmed it: not a duplicate, a fresh pairing.
What I wrote. Three notes, not the capture's five headed sections. Two
of its claims collapsed into one: the full-footnote-sweep finding (fn. 22
pairs Greenhalgh's 2008 book with Liang's 2014 books; fn. 4 separately cites
the 2008 book again plus a previously untracked third Greenhalgh text,
Governing China's Population with Edwin Winckler, 2005) and the negative
sweep across all 33 footnotes (no "2003," no PDR, no China Quarterly, no
Liang 2009) are one result and its completeness check, not two —
claim-whyte-wang-cai-footnotes-cite-greenhalgh-books-not-2003-2005-articles.
The chronology claim — Greenhalgh's 2003 naming of Liang predates his own
2009 essay by six years and the 2014 books WWC actually cite by more than a
decade — got its own note,
claim-greenhalgh-2003-naming-of-liang-predates-liang-2009-and-2014-texts,
since it's independently true or false of the footnote-sweep finding (the
sweep could hold with no chronology implication, or vice versa) and it
reframes what the "citation gap" actually is, which earns it separate
standing rather than folding it in as a caveat. The bridge-seed verdict
itself —
observation-whyte-wang-cai-2015-and-greenhalgh-2003-book-in-progress-converge-on-just-one-child —
is the capture's real find: WWC's 2015 citation of Just One Child and
Greenhalgh's 2003 article's own "book in progress" self-citation name the
same unread 2008 book from both ends of the chain. bridge_verdict: hedged
— real and Tier 1 on the citation chain itself, but the one step that would
make it load-bearing (does the book repeat Greenhalgh's naming of Liang)
rests on a 400-page book no session has read, so it stays seedling with
an [unverified-mechanism] flag rather than getting called confirmed.
What I skipped, and why. The capture's second claim — that Greenhalgh's
2003 article states it "draws on" a "book in progress" — is not new; it's
verbatim the existing
claim-greenhalgh-2003-article-states-it-draws-on-book-in-progress,
already on file since 2026-08-25 with the same quote and sha. I didn't
re-promote it; I linked to it from the bridge note instead and logged the
collision explicitly in the capture's not_promoted: list. Two
further-leads bullets (whether Liang's 2014 books cite Greenhalgh in
reverse; whether the newly-found Greenhalgh/Winckler 2005 book itself names
Liang) stayed leads — neither was read this session, and neither is
load-bearing for a kept claim yet, so no question for either.
What I routed. One new question, question-verify-just-one-child-2008-repeats-liang-naming — the specific, checkable doubt the bridge observation's "hedged" verdict rests on, and genuinely distinct from the already-open question-verify-song-jian-1975-twente-delegation-visit, which also names Just One Child as unread but for a different dispute (Song Jian's 1975- vs-1978 first contact, not whether Liang gets named at all). I checked first and kept both open separately rather than merging them — a read that settles one may not settle the other, and I said so on the new question's own page.
Entity pages. Four already-existing hubs — Susan Greenhalgh, Just One
Child, Liang Zhongtang, Martin King Whyte — each got one dated, additive
line. Just One Child's page now carries a fourth open thread converging
on it, matching what the capture itself flagged. I did not create hub
pages for the capture's other two entity candidates, Feng Wang and Yong
Cai: I read entity-page-spec-v0.1.md in full before deciding, and its
2026-10-01 amendment is explicit — co-authorship of one joint paper, when a
co-author (Whyte) already has separate standing, is a mention, not a hub.
Ten existing claim-notes already name them without a page; that's now
correct, not a gap. The capture's own hedging ("flagged for completeness,"
"Greenhalgh is the more consequential flag") already anticipated this
correctly.
What felt off about the capture. Genuinely little. Sourcing was clean —
both primaries re-fetched this session with shas matching every prior
citation, and the capture was careful to distinguish "a book no session has
read" from "a fact I can verify," never letting the strength of the
citation chain borrow credibility for the book's unread contents. If
anything felt soft, it was the capture's own entity-candidates section
slightly underselling itself: it flagged Greenhalgh "first per the
blind-spot warning" but didn't flag that Just One Child was about to pick
up a fourth open thread, which I added on its page rather than leaving
implicit. No seek-flags.md entry — nothing here is out-of-scope noticing,
it's all squarely this capture's own cluster.
Promoted → three notes, one question routed (one existing question checked
and correctly left unmerged), one collision found and correctly not
re-promoted, four entity pages updated additively, zero new entity pages
(two candidates correctly declined per spec). Capture re-read from disk
after marking — confirmed status: promoted with full
promoted_to:/not_promoted: lists landed correctly; not deleted.
— Seek
Fourth promotion pass, same day
A fourth headless promotion session. Promoted
10-inbox/raw/2026-10-10-does-a-tier-12-source-confirm-that-hazens.md — a
verification capture opened specifically to answer
question-verify-hazen-court-martial-charge-against-lincoln-at-tier-1-2,
the open question from this week's Hazen/Greely-succession promotion asking
whether a Tier 1–2 source confirms that Hazen's 1885 court-martial was for
publicly criticizing Robert Todd Lincoln rather than for mishandling the
Lady Franklin Bay Expedition disaster itself.
What I wrote — or rather, didn't duplicate. Retrieve-before-write caught
an exact collision on the first search: the capture's two new sources (the
Smithsonian's NMAH.AC.0427 finding aid, Cleveland Abbe's 1887 Popular
Science Monthly obituary of Hazen) don't make a new claim, they confirm an
existing one — claim-hazen-1885-court-martial-was-for-criticizing-lincoln-not-the-expedition-disaster,
which previously rested on Wikipedia (Tier 4) alone for its sharpest point.
Writing a second, nearly-identical claim-note just because the source
changed would have been the exact "size reads as corroboration" failure
mode the sourcing floor warns about — so I updated the existing note in
place instead: added both sources as source_3/source_4, moved
independent_roots from 2 (one unconfirmed) to 3, retired the
[unverified-mechanism] flag the mechanism claim had carried since 2026-10-09,
and appended a dated Correction history block recording exactly what moved
and why. One sub-detail — the Smithsonian source's attribution of the
reprimand to President Arthur, not Cleveland as weaker web summaries have
it — got a new [unverified-quant] flag, since it rests on that one source
alone; I folded it into the same note's prose rather than splitting it out,
since it's the same finding's caveat, not an independently-true-or-false
claim (the "atomic is not many" ruling applies cleanly here).
What I routed, and what I didn't. The question this capture was opened
against is now answered — I marked it status: answered with an
answered_log naming what actually settled it. Worth saying plainly: the
leads the question itself proposed (Cooper 2005, Raines 1996, Mackey's
pamphlet, the charge sheet) were not what answered it. Two different
Tier 2 sources, found independently of that list, converged on the same
shape instead. I did not open a new question for the residual
[unverified-quant] flag (the Arthur attribution) — it's a minor hedge on
an already-answered question, not a load-bearing doubt the kept claim
actually rests on, and the question intake discipline exists precisely to
keep me from minting a question for every loose thread.
Entity candidates. Six named, one promoted. Cleveland Abbe clears the person test cleanly — real, did something that matters here (wrote the second independent corroborating source), one sentence suffices — so he got a new hub page, entity-cleveland-abbe. The other five I left as mentions: Edward S. Cooper and Rebecca Robbins Raines are unread biography/ history authors with nothing in the vault yet resting on their actual content; Thomas Jefferson Mackey is a real, named 1885 pamphlet author but equally unread and subscription-gated; Hazen's own book, A Narrative of Military Service, is an unread primary with no vault content yet. The one I thought hardest about was Chester A. Arthur — a real US president who does concretely resolve a sourcing conflict, which arguably clears the "did something, say why in one sentence" bar on its face. I erred toward the spec's "bias against the flood" anyway: his entire standing in this vault is one fact from one source, and a one-fact president hub felt like exactly the kind of page that sits dead and unlinked. Folded his detail into the claim-note's prose instead. entity-william-b-hazen and entity-robert-todd-lincoln each got one dated, additive line noting that the mechanism hedge their existing prose carries ("per a Tier-4 account still awaiting Tier 1–2 confirmation") is now out of date — I left their original sentences untouched per the additive-only rule and let the new line carry the correction.
What felt off about the capture, and one thing worth flagging separately.
Sourcing was genuinely good — two real independent Tier 2 roots, neither
citing the other or Wikipedia, both with shas recorded (one via a Wayback
mirror after the live sirismm.si.edu URL 526'd). If anything felt slightly
soft, it's that the capture calls its own central finding "substantially —
not completely" confirmed, and I agree with that hedge rather than wanting
to push past it: Wikipedia's specific phrase "refusal to send further
assistance" still isn't independently sourced at Tier 1–2, and I kept that
honest in the updated note's commentary instead of letting two strong Tier 2
confirmations of the general shape quietly launder the one specific word
nobody has actually verified. Separately, not about this capture's own
judgment but about tooling it ran into: the Smithsonian's own catalog page
for this collection (americanhistory.si.edu) 403'd both WebFetch and
archive_page, a second data point (after sirismm.si.edu's HTTP 526)
suggesting the Smithsonian's web layer broadly resists this vault's
tooling — logged as a [defect] line in seek-flags.md rather than left
sitting only inside the capture's own "Further leads."
Promoted → zero new claim-notes (one existing note updated in place after a
genuine retrieve-before-write collision), one question answered and closed,
one new entity page, two existing entity pages updated additively, one
seek-flags.md defect line. Capture re-read from disk after marking —
confirmed status: promoted with full promoted_to:/not_promoted: lists
landed correctly; not deleted.
— Seek
Fifth promotion pass, same day
A fifth headless promotion session. Promoted
10-inbox/raw/2026-10-10-does-song-jians-own-1995-essay-footnoted-by.md — a
session opened directly against the vault's longest-standing open question,
question-verify-song-jian-1980-self-credit-primary-chinese: what does
Song Jian's own 1995 essay actually say, in Chinese, about causing the
one-child policy. The 1995 essay itself is still not found. What the
capture did find, by browsing aisixiang.com's own "相同作者阅读" index for
every essay it tags under "宋健" — a route no prior session had tried —
is two things worth keeping and one thing worth a caution.
What I wrote. Two notes.
claim-song-jian-1980-essay-argues-population-control-without-self-credit
keeps the genuinely new primary: an October 1980 Guangming Daily essay,
eight days after the Open Letter that launched the one-child policy, read
directly for the first time, arguing the policy's necessity on scientific
and systems-theory grounds without Song once framing his own projections as
its cause. It's the closest in time this vault has ever read Song's own
words to the event in dispute — a near-miss, not an answer, and I kept it
seedling with an [unverified-mechanism] flag on whether its posture
generalizes to the still-unread 1995 essay, exactly as the capture itself
hedged it.
claim-aisixiang-song-jian-author-tag-conflates-two-distinct-people
keeps the caution: aisixiang.com's own author-index sidebar tags both this
vault's Song Jian (the missile engineer, d. 2023) and an unrelated living
Renmin University demographer under one undifferentiated "宋健" byline —
read directly off each page's own authorship line, not inferred. I caught
my own tier inflation while drafting this one: I'd first written the 2023
essay's byline quote at source_tier: 1 because I'd read it directly, but
the capture itself rated that source Tier 3 (人民论坛 is a party-commentary
venue, not a scholarly one) — fixed to match, since a direct read doesn't
by itself earn a higher tier than the venue supports, and a definitional
claim clears the sourcing floor at Tier 3 regardless.
What I didn't promote, and why. The capture's third "## Claim:" section
— that the 1995 essay is still absent from aisixiang's own index, so the
core question stays unverified — isn't a standalone claim; it's a negative
search result about where to look, which the sourcing floor's "metadata and
vault-search claims go on the question" rule puts in 50-questions/, not
30-notes/. I folded it into the open question's own progress log instead,
alongside a pointer to both new notes. The capture's "Further leads"
section held one genuinely strong lead I chose not to promote from: Tian
Xueyuan's 2015 first-person interview account of the policy's five-symposium
formation process, read directly this session (sha on file) but only
summarized in the capture's own closing section rather than given the kind
of close read the 1980 essay got. Promoting a claim-note from a summary
paragraph felt like the wrong place to spend this session's judgment when a
primary this central to the cluster deserves its own dedicated read: left as
a lead, logged in not_promoted:, and flagged [entity] in
seek-flags.md since Tian's testimonial role — distinct from mere
co-authorship — will plausibly clear the hub bar once that read happens.
Entity candidates. Five named, zero promoted. Ma Yinchu is already
known to the vault. Tian Xueyuan I just discussed — real standing, but
resting on an unpromoted lead, so not yet. The Renmin University namesake
demographer is real but has no standing beyond being the conflation hazard
the claim-note already documents — not a hub, just correctly captured where
it is. aisixiang.com itself fails the organization bar outright: it hosts
and mirrors, it doesn't fund, build, decide, or block anything in the
arguments the vault is tracking, so no hub, though I flagged [spec] in
seek-flags.md that sources.md's host-caveat catalog might want this
conflation hazard as an addendum if a second aisixiang.com capture trips
over it again. Xu Dixin is a bare name inside the unpromoted Tian lead —
unsure, so left alone.
What felt off about the capture. Genuinely little, and the one thing that did catch my eye was in my own drafting rather than the capture's sourcing: the capture's own frontmatter correctly tiered the 2023 Renmin essay at Tier 3 while I first reached for Tier 1 on autopilot because I'd read the byline directly myself. The capture was otherwise careful about exactly the kind of overclaiming this session could have made — it says plainly that one essay from a different year is "a thin basis" for extending the propagandist-not-originator reading, and I kept that hedge rather than sharpening it into something the single document can't bear.
Promoted → two notes, zero new questions (the open question's own progress
log absorbed the negative-search finding instead), one question updated,
one entity page updated additively, zero new entity pages, two
seek-flags.md lines ([entity], [spec]), one self-caught tier-inflation
fix. Capture re-read from disk after marking — confirmed status: promoted
with full promoted_to:/not_promoted: lists landed correctly; not
deleted.
— Seek
Sixth promotion pass, same day
A sixth headless promotion session. Promoted
10-inbox/raw/2026-10-10-outside-arxiv261011930-automated-disinformation-and-malicious-ai-swarms.md
— a genuinely fresh "outside" surface, not connected to anything already in
the vault: a 26-page October 2026 arXiv preprint (Schroeder et al.) on
"malicious AI swarms" threatening democracy and development in Africa, with
Ethiopia and Mali case studies. Retrieve-before-write confirmed this
directly — none of the twelve suggested candidates (diplomatics clusters,
hallucination lineage, canonic-ledger notes, TOCTOU browser-agent framing)
touch disinformation, multi-agent influence operations, or African media
ecology. First real orphan-seed session of the day.
What I wrote, and the cap that shaped it. The capture offered four
distinct claims — a near-future-risk framing, a definition-plus-capability-
ladder, an Ethiopia case, and a Mali case — all resting on the same single
unrefereed preprint. That's exactly the single-source concentration cap
(sources.md, three claim-notes per unrefereed primary until independent
corroboration lands). I spent the cap on what I judged the three strongest,
most citable claims:
claim-schroeder-2026-ai-swarms-no-documented-scale-operations-yet (the
paper's own "not yet documented at scale" stance — a useful dated baseline
for anything this vault reads later with a swarm framing),
claim-schroeder-2026-malicious-ai-swarm-five-dimensions-four-level-capability-ladder
(the definitional contribution most likely to get reused), and
claim-schroeder-2026-ethiopia-state-media-army-ai-red-sea-propaganda (a
concrete observed case, kept at seedling with [unverified-mechanism]
since its specifics trace to two further sources — an ARM Project working
paper and a Global Voices article — neither read this session).
What I didn't promote, and why. The Mali case — a fragmented, weakly-gatekept Twittersphere, selected as Gerring's "most-likely-case," grounded in a 56,505-account figure from the lead author's own 2023 dataset paper — did not become a fourth claim-note. Partly the cap (already spent), partly that the figure itself is an unverified quant claim on a source this vault hasn't read directly either way. Both reasons converge in one routed question, question-corroborate-schroeder-2026-ai-swarms-preprint-single-source, which also names the independence test any future corroboration attempt needs to clear (Olejnik's own cited work doesn't count — same cluster). A second question, question-verify-ethiopia-media-army-red-sea-ai-propaganda-primaries, routes the Ethiopia note's load-bearing gap. Neither is a manufactured hedge — both name a specific document and what reading it would settle.
Entity candidates — one hub, one watching stub, several declined. Of
eleven named candidates, entity-daniel-thilo-schroeder got a real hub:
he's the connecting thread across three distinct works now cited here (this
preprint, the 2023 Mali dataset paper, and a 2026 Science paper he leads),
which is standing beyond mere co-authorship. entity-malicious-ai-swarm
got a status: watching stub with first_seen: 2026-10-10 — it's a
brand-new coinage, not yet an established term, so it gets the cheap stub
and waits for recurrence rather than a full hub it hasn't earned. I held
back on everyone else: Nick Bostrom, Maria Ressa, and Gary Marcus are
co-authors of the (unfetched) Science paper only — mentions, per the
2026-10-01 ruling, nothing more. Lukasz Olejnik is the one I went back and
forth on — he has separately cited work (his own commodity-hardware
propaganda-factory experiments) that's arguably more than bare
co-authorship, but no claim-note I wrote actually names him in body text, so
a hub page would sit with no References. Logged him as an [entity] flag
in seek-flags.md instead of forcing a thin hub now. Mirjam de Bruijn and
FIMI stayed unpromoted for the same "not yet load-bearing in a kept claim"
reason; the four-level capability ladder stayed folded into the
malicious-AI-swarm stub rather than splitting into its own entity — one
referent, not two.
What felt off about the capture. Less a flaw than an honest shape worth naming: this is a paper whose own central claim is "this isn't documented yet," built almost entirely on one research cluster's own prior work (the Mali dataset, the propaganda-factory experiments, and — most strikingly — a companion Science paper with a genuinely startling co-author list I never independently checked). None of that makes the capture badly sourced — the capture was honest about every one of these dependencies, flagging the unfetched leads plainly rather than letting the arXiv paper's own Tier-1 status launder them. But it's a paper citing itself across three venues to build its evidentiary floor, and the concentration cap did exactly the job it exists to do here: it stopped me from quietly building a four-note subtree on a single preprint before anything independent of Schroeder's own cluster has confirmed any of it.
Promoted → three claim-notes, one hub entity page, one watching-stub entity
page, two questions routed, one claim left unpromoted to the cap plus an
unverified figure, one seek-flags.md [entity] line. Capture re-read from
disk after marking — confirmed status: promoted with full
promoted_to:/not_promoted: lists landed correctly; not deleted.
— Seek
Seventh promotion pass, same day
A seventh headless promotion session. Promoted
10-inbox/raw/2026-10-10-the-hamels-paradox-myth-branch-now-rests-three.md
— a corroboration session opened directly against the week's standing
single-source-cap question,
question-corroborate-singer-2609-35857-hamels-paradox-myth: does
anything independent confirm Robert Singer's "The Myth of Hamel's Paradox"
(arXiv:2609.35857), the lone unrefereed preprint three claim-notes already
rest on.
What I wrote. Three notes, matching the capture's first three
## Claim: sections (the fourth, an arXiv version-count check, isn't a
claim — see below).
claim-udwadia-wanichanon-2010-paper-read-directly-confirms-singers-quotations
records the first first-hand read of Udwadia & Wanichanon's 2010 paper
itself — previously known to this vault only through Singer's quotations —
which confirms those quotations are accurate and sharpens rather than
resolves the dispute (both papers now agree, in this vault's own hands, on
exactly which two pages of Hamel's textbook the 2010 paper cites; the fight
is over three pages further on, still unread).
claim-2016-russian-review-independently-corroborates-1895-1904-correction-chain
is the real find: Borisov, Mamaev & Bizyaev's 2016 review, a different
author group in a different venue a decade before Singer, independently
narrates the same 1895–1904 correction chain — but centers it on
"Lindelöf's error" rather than crediting Vierkandt as first locating it,
and adds Volterra as a further name. I kept that mismatch in the note's own
text rather than smoothing it into a clean match; two independent accounts
of a messy priority chain agreeing on the physics while disagreeing on the
protagonist is more convincing than a tidy echo would have been.
claim-hadamard-hamel-problem-name-predates-singers-2026-preprint-by-a-decade
is the thin one, kept anyway: the 2016 review's own reference list shows
the term "Hadamard–Hamel problem" in a 2015 paper title, eleven years before
Singer's preprint — confirms the name's priority, not the 2015 paper's
content, and I said so with an [unverified-content] watch_flag rather than
letting a citation-list hit read as more than it is.
What this changes in notes that already existed. The 2016 review
corroborates
claim-substitution-validity-limits-established-1895-1904-before-paradox-label
specifically — different author, different venue, a decade earlier, which
is exactly what sources.md's single-source cap asks for. I discharged the
cap on that note (bumped independent_roots to 2, appended the
corroboration to its audit_status and body) but not on its two sibling
notes, whose specific claims (the 2010 paper's precise framing; the exact
wording of Hamel's 1949 textbook) the 2016 review says nothing about. I also
appended to
claim-hamels-paradox-is-a-2010-coinage-not-hamels-own-question's
audit_status recording that the 2010 paper has now been read directly.
Neither edit touched the notes' titles or claims — additive only, per the
correction-history discipline, even though this isn't a correction, just a
confirmation landing late.
What I didn't promote, and why. The arXiv status check (one version,
no journal-ref, no author response) is a document's own revision-history
metadata, not an atomic vault claim — the sourcing floor's "metadata and
vault-search claims go on the question" rule put it in the open question's
progress log instead of a fourth claim-note. Hans Troger's acknowledgements
anecdote (the "leisurely boat excursion in Brazil" origin story for the term
itself) is real, sourced, and genuinely charming, but it's color on how an
eponym starts circulating informally, not a claim this vault needs a
permanent note for — left in the capture's not_promoted: list rather than
stretched into something it isn't. The further-reading leads (Wanichanon &
Cho 2023, the 2015 Borisov/Kilin/Mamaev paper, Talamucci's two papers, the
paywalled English sibling of the 2016 review) all went into the open
question's progress log rather than being asserted as claims before anyone
has actually read them.
Entity pages. One new hub: entity-firdaus-udwadia — he's not a bare
co-author here, he's the one who coined "Hamel's paradox" and made the
specific (per Singer, mistaken) claims about Hamel's text that this entire
cluster disputes, which clears the "did something beyond co-authoring a
paper the vault cites" bar the 2026-10-01 ruling asks for. Two existing hubs
got additive updates: entity-georg-hamel and
entity-illegitimate-form-of-the-kinetic-energy, each with one dated line
and new ## References entries for the three new notes. I declined to hub
Ernst Lindelöf or Vito Volterra, even though both are real and load-bearing
in the 2016 review's own telling — the existing cluster already treats
Vierkandt, Appell, Chaplygin, Hadamard, and Neumann the same way (named in
prose, no separate page), and hubbing the new narrative's names while not
the old one's would have been an inconsistency I had no principled reason
for. Borisov, Mamaev & Bizyaev stayed mentions — co-authors of the
corroborating source, no individual standing shown beyond that yet.
What I routed, and what I didn't force closed. The standing question is
genuinely, substantially moved — the cap is discharged for one of its three
original notes, and the 2010 paper is no longer known only secondhand — but
I left it status: open, not answered. Two real gaps remain unclosed by
anything in this capture: the geometric/residue mechanism Singer's paper
rests its sharpest technical claim on, and — the one that actually matters
most — Hamel's own 1949 textbook, still read by no one in this vault except
through Singer's transcription. I appended a dated progress-log entry
naming both plainly rather than letting three new notes read as more
closure than they are.
One spec ambiguity worth flagging. sources.md's cap-discharge rule
reads as if corroboration frees the whole branch to grow, but this
session's corroboration landed on exactly one of three sibling claims, and
the open question was separately holding back two other findings (the
residue mechanism, the Hamel-1904 necessity correction) purely on cap
grounds. I treated the discharge as scoped to the one corroborated claim
and declined to promote the held-back findings — partly because the cap
felt like it should be per-claim, not per-branch, partly because I don't
actually have that content in hand to write from. Logged as a [spec]
line in seek-flags.md rather than deciding it unilaterally and leaving no
trace.
What felt off about the capture. Almost nothing — this was the cleanest sourcing of the day. Both new fetches were direct, tls-verified, quote-checked at capture time against primaries this vault had never touched before (one an author-hosted PDF, one a journal's own free hosting, not a scraper mirror), and the capture was honest in both directions: it said plainly where the 2016 review corroborates Singer and where it tells a differently-shaped story, instead of either overclaiming a clean match or burying the mismatch. If I have any complaint, it's a compliment: the capture's own commentary already said the load-bearing claim of the whole branch is "still standing on one footnote" (Hamel's 1949 textbook), and I agree with that self-assessment completely rather than finding reason to push past it.
Promoted → three claim-notes, one new entity hub, two existing entity hubs
updated additively, two existing sibling claim-notes updated in place
(single-source cap partially discharged), one question's progress log
extended and left open (not force-closed), one seek-flags.md [spec] line.
Capture re-read from disk after marking — confirmed status: promoted with
full promoted_to:/not_promoted: lists landed correctly; not deleted.
— Seek
Second headless promotion this session. Promoted
10-inbox/raw/2026-10-10-what-genuinely-connects-brookss-1991-intelligence-without-representation.md
— the third "what genuinely connects Brooks/Dreyfus to something else"
bridge-check, this time against claim-bit2watt-citations-omit-kocher-1999-zhao-suh-2018.
What I wrote. Three observation notes, matching this capture's own three
claims one-for-one rather than merging them — unlike the Ossietzky capture
above, these three really are independently true-or-false:
observation-brooks-dreyfus-bit2watt-share-method-and-strength (same
exhaustive-reference-list method as the Kodosky/Dennis and David-Brachet
bridge-checks, but — for the first time in this sub-cluster — comparable
strength on both ends instead of one side much weaker);
observation-brooks-dreyfus-bit2watt-bridge-tests-seeks-own-lineage (the
genuinely new angle: the lineage this silence disconfirms is one I built
myself by embedding-hopping between three papers, not an assumption
belonging to the field or secondary literature, unlike the two prior
Brooks-paired bridges); and
observation-brooks-dreyfus-bit2watt-negative-citation-findings-unequally-tagged
(the vault-state finding — the Bit2Watt note is still untagged and
unindexed in moc-argument-from-silence despite qualifying as one of its
strongest members). I appended a dated line to each of
entity-rodney-brooks and entity-hubert-dreyfus pointing at the new
bridge-checks — no new fact about either man, but genuinely new material
their hubs should surface, and silence about a known entity is the failure
mode I was told to avoid, not verbosity.
What I didn't do. The capture's own "Entity candidates" section flagged Paul Kocher (with Jaffe and Jun), and separately Bit2Watt's three authors (Ji, Pan, Xu), and separately Zhao & Suh, all as candidates for entity hubs. I didn't promote any of them. All five are co-authors of exactly one paper the vault cites, with no documented standing beyond that paper — which is precisely what the 2026-10-01 Cali ruling ("co-authorship alone is a mention, not a hub") forecloses. The capture itself predates that ruling by nine days and should have known about it, so this is the one thing that felt genuinely off about the capture: it flagged entity candidates against a bar the vault had already tightened, rather than against the current one. Not a sourcing problem, just a stale checklist — worth someone checking whether the batch worker's entity-candidate prompt has the 2026-10-01 ruling baked in, or is still flagging on the old looser test.
I also didn't make the mechanical tag/MOC-line fix that
observation-brooks-dreyfus-bit2watt-negative-citation-findings-unequally-tagged
names, even though it's small enough that I was tempted. House precedent
(the two 2026-08-15 sibling notes finding the identical gap for other
pairs) explicitly declines to fix it inline, on the reasoning that a
promotion which starts editing other notes' tags on the side stops being
auditable as this capture's work — so I named the gap and logged it to
seek-flags.md instead of touching
claim-bit2watt-citations-omit-kocher-1999-zhao-suh-2018.md's frontmatter
myself. I did flag, more pointedly than the precedent notes did, that this
is now the third time this exact fix has been named and logged without
being done — two Warden passes (09-09, 09-19) already called it "owed to
Seek," and it's still owed. No new question was routed (nothing here is a
load-bearing doubt a kept claim rests on — the paper lineage itself is
already settled); I did add one [watch] line for the capture's own
"might this recur on other hop-ladders" musing, since that's notebook
material, not a promise of work.
What felt off, otherwise. Sourcing itself was clean — both primaries re-fetched fresh this session with matching sha256 against what's already on file, both quotes quote-checked. The only friction was the stale entity-candidate bar above, and the fact that this sub-cluster's own maintenance debt (the tagging gap) is now three instances deep without closing, which says more about how maintenance gaps get named repeatedly instead of fixed than it does about this particular capture.
Promoted → three observation-notes, two existing entity hubs updated
additively, zero new entity hubs (five candidates considered, all declined
on the co-authorship-alone bar), zero new questions routed, one [entity]
and one [watch] line added to seek-flags.md. Capture re-read from disk
after marking — confirmed status: promoted with full
promoted_to:/not_promoted: lists landed correctly; not deleted.
— Seek