---
date: "2026-11-18T00:00:00.000Z"
---


Headless promotion run. One capture in the queue:
`10-inbox/raw/2026-11-18-does-any-independent-source-corroborate-descamps-arxiv261121336-force.md`
— a corroboration search for Pascal Descamps's arXiv:2611.21336 ("Why Cassini's Oval Could Not Replace Kepler's Ellipse?"), the preprint that hit the single-source cap yesterday at three claim-notes.

**What I promoted.** Four distinct claim-notes, none of them restatements of Descamps himself:

- [[claim-sivardiere-1994-no-central-force-cassini-oval-via-tucci]] and [[claim-morgado-soares-2014-oval-ellipse-indistinguishable-via-tucci]] — two 1990s/2010s *European Journal of Physics* results, reached only because Pasquale Tucci's unrelated 2025 preprint quotes them in passing, that anticipated the general shape of Descamps's dynamical and kinematic claims a decade or more early. I split these into two notes rather than one, mirroring the vault's existing split of Descamps's own force-law and equant claims — each corroborates a different existing note, and the two underlying facts (no central force fits the oval; the oval and ellipse are kinematically close) are independently true-or-false.
- [[claim-moore-smith-mertens-2026-cassini-oval-fails-exact-equant]] — a different section of a paper already in the vault for an unrelated reason (the Astronomia Ova orbit family). Its §2.4 takes on the *actual historical* Cassini oval as a candidate equant and finds it fails the *exact* bar, which doesn't match or contradict Descamps's *approximate*-equant claim — just measures something adjacent, independently.
- [[claim-leibniz-1701-draft-letter-confirms-fontenelle-letter-content]] — the strongest result of the session. The Akademie-Ausgabe's own archival text of a rejected draft Leibniz wrote to Fontenelle confirms, from his own hand, the existence and content of the letter Descamps's Fontenelle claim rests on — read independently of Descamps's citation of Cohen (1962) entirely. Leibniz's draft also asks, in 1701, the same force-law question Descamps answers 325 years later.

I did **not** promote the capture's "no citing paper found" finding — that's a claim about this session's own search, not about the world, and the spec is explicit that vault-search claims go on the question's log, not into `30-notes/`. I logged it there instead, along with a progress note on the whole corroboration search: I left [[question-corroborate-descamps-2026-arxiv-2611-21336-single-source-cap]] open rather than marking it answered, because what I found is general-shape corroboration, not independent reproduction of Descamps's actual numbers (𝒟(e), the 38× figure). Partial, not full — I'm not forcing that closure.

**Entities.** Four new hub pages (Fontenelle, Jean Sivardière, Pasquale Tucci) plus updates to three existing ones (Leibniz, Giovanni Domenico Cassini, the Cassini-oval concept hub) — appended, not rewritten, per the additive-update rule. I did *not* promote Pierre Varignon, even though the draft letter names him investigating the exact force-law question in 1701 — because that specific detail arrived in the capture as unquoted paraphrase, not a verbatim-checked passage, and I have no network access here to check it myself. I flagged it `[unverified-quote]` inside the Leibniz claim-note instead of building a biography on a sentence I can't confirm was actually there. Also skipped: Morgado and Soares as individual hubs (joint co-authors, no separate standing — mention only, per the co-authorship rule) and Stefan Jenschke (the capture's own judgment that he's too minor was right, left as-is).

**What felt off.** Nothing structurally wrong with the capture — sourcing was careful, tiers were honestly assessed (Tier 2 for the Tucci-mediated quotes, correctly not inflated to Tier 1), and the capture's own "Further leads" section was honest about what it hadn't read (Cohen 1962, the EJP originals, Bussotti, Heilbron — none of these got promoted as claims, because none were actually read). The one thing I want to name plainly: this is now six claim-notes, four entity hubs, and three entity updates all orbiting a single two-day-old, unrefereed preprint's reception — a lot of vault structure for a paper that still has zero citations. Every piece of it is honestly sourced and nothing here rests on Descamps's own numbers without saying so, but it's worth someone's eye in a few weeks whether this turned into the over-grown branch the single-source cap exists to catch, or whether it's just what a well-corroborated historical footnote actually looks like once you go looking. I don't think it's the failure mode — the cap specifically governs notes resting *on the single source*, and today's notes all rest on four different, independent documents — but the shape is dense enough to flag.

No git commands run — leaving commit and push to the auto-commit agent, per today's operating mode.

---

**Second promotion, same day.** A different capture:
`10-inbox/raw/2026-11-18-what-genuinely-connects-geert-jan-olsders-own-account.md`
— another bridge-seed check from the Song Jian/Olsder/one-child-policy
cluster, this one testing a cosine-0.88 pairing between
[[claim-olsder-account-traces-song-jian-twente-contact-to-1975-delegation-visit]]
and
[[claim-whyte-wang-cai-2015-essay-reliable-on-national-dating-unreliable-on-yicheng-specific-date]].

**What I promoted.** One note:
[[observation-olsder-1975-whyte-wang-cai-reliability-split-share-no-document-clean-false-friend]].
The capture laid out three "claims," but they were one finding wearing
three hats — the no-shared-document result, the note that two prior
sessions already rejected the nearest analogous link, and the
cosine-tracks-vocabulary explanation are a result, its caveat, and the
note on how it was found. I folded all three into a single atomic note
rather than writing three, per the "atomic is not many" ruling. It carries
`bridge_verdict: false-friend` and stays `seedling`, since it rests on
Olsder's own `[unverified-historical]` 1975 account.

**Collisions checked.** I read both target claim-notes in full plus the
two already-promoted observation notes the capture itself cites as having
investigated structurally analogous candidate bridges
([[observation-song-self-report-and-whyte-wang-cai-national-reliability-notes-connect-only-through-greenhalgh-2003]],
[[observation-olsder-liang-and-song-whyte-wang-cai-meta-pairing-is-hedged-false-friend-via-song-1986-quote]]).
No existing note already covers this exact pair — confirmed genuinely new,
not a duplicate.

**Question routing.** [[question-verify-song-jian-1975-twente-delegation-visit]]
got a dated progress-log line (now its twelfth), not a new question — the
capture is explicit that this bridge-check adds no evidence toward
resolving 1975-vs-1978 itself, only another negative result on a candidate
connection. Leaving it open was the honest call; nothing here moves the
needle on the actual dating dispute.

**Entities and other leads — skipped, deliberately.** All five entity
candidates (Song Jian, Liang Zhongtang, Yan Wei & Li Zhang, Olsder,
Greenhalgh) already have hub pages or a standing non-promotion ruling, and
the capture's own text says each is "reinforced, no new standing claimed."
I checked each hub page mentally against this capture's content and found
nothing factual that isn't already there — a negative bridge-check result
about *other* notes doesn't teach an entity page anything new about the
person. So no hub edits, and I'm saying so here rather than leaving it
silent. Also skipped as not-new: the paywalled Yan Wei & Li Zhang 2014
article (already tracked) and the "count kept in six places" bookkeeping
problem (already flagged at the 2026-10-14 audit addendum) — this capture
is one more data point on both, not new signal needing a fresh flag.

**What felt off.** Nothing wrong with the sourcing — the Olsder interview
sha matched every prior session's capture, tiers were honest, and the
capture was unusually self-aware about its own place in a crowded
neighborhood. What I'll name plainly instead: this is now at least five
bridge-check notes in this sub-cluster alone concluding "false friend" or
"hedged false friend" between slight variations on the same four or five
source documents. Each one is individually correct and non-duplicative —
I checked — but the *rate* of near-identical meta-investigation this
cluster is generating relative to how rarely it produces a new primary
fact about 1975-vs-1978 itself is worth someone noticing as a shape, not
just as individually sound outputs. I didn't flag it to `seek-flags.md`
because it isn't a defect or a new structural gap — it's the same
"count kept in six places" pattern already on the ledger — but it's the
kind of thing I'd want surfaced if the bridge-seed worker keeps finding
cosine neighbors inside this same five-document pocket.

No git commands run — leaving commit and push to the auto-commit agent.

---

**Another promotion, same day, same cluster.** A capture:
`10-inbox/raw/2026-11-18-what-genuinely-connects-whyte-wang-cais-2015-essay.md`
— one more bridge-seed check landing in this same citation-genealogy cluster
today, this one testing the cosine-0.88 pairing between
[[claim-whyte-wang-cai-and-liang-disagree-on-yicheng-pilot-start-date]] (the
1980-vs-1985 Yicheng dating conflict) and
[[claim-liang-1985-letter-to-hu-qili-hao-jianxiu-preceded-yicheng-pilot]] (the
spring-1985 letter to Hu Qili and Hao Jianxiu that Sixth Tone credits with
the pilot finally happening).

**What I promoted: nothing new in `30-notes/`.** This is the thinnest of the
four bridge-seed captures today, and I mean that as the actual finding, not a
complaint. The capture's three "claims" boiled down to one real fact (both
target notes draw on the identical archived Sixth Tone article, sha
`46eabd00…` — the discrepancy note's bare "July 1985" figure and the letter
note's escalation narrative are two halves of one document) plus a caveat
(it doesn't resolve the 1980-vs-1985 conflict) plus a methodological aside
about the cosine score itself. The capture's own words: "no new document, no
new fact." I agreed, and went further than the capture did: this exact
"shared landlord, not shared argument" shape already has well past a dozen
instances in `30-notes/` (I grepped — 11 files matched before this session,
and that's before counting the three siblings I promoted earlier today). The
capture's own "fourth documented instance" count is stale by the vault's own
reckoning — an audit addendum from 2026-10-14 already had the tally past six
and called it "a count kept in six places is a count nobody owns." Writing a
thirteenth-or-so near-identical observation-note felt like exactly the
proliferation the vault has already flagged on itself twice. So instead of a
new note, I recorded the one real fact as a single dated Updates line on
[[entity-liang-zhongtang]], cross-linking the two existing claim-notes
directly, and logged the skip in the capture's own `not_promoted:` list. The
cosine-score methodological claim I dropped outright — per promote.md §5a,
claims about the vault's own search/embeddings don't belong in `30-notes/`,
and nothing here is load-bearing enough to route to a question either.

**Entities — one real gap, found and closed.** Four of six candidates
(Liang Zhongtang, Hu Qili, Whyte/Wang/Cai) already have hubs and needed
nothing beyond the Updates line above. Yan Wei and Li Zhang stay unpromoted
— co-authorship of one still-paywalled joint 2014 article is a mention, not
a hub, per Cali's 2026-10-01 ruling, and that hasn't changed. But Hao
Jianxiu — named in the exact same sentence as Hu Qili, same source, same
weight — had no page at all, even though Hu Qili got one on 2026-10-20 for
that identical sentence. That's an actual oversight, not a judgment call, so
I created [[entity-hao-jianxiu]] this session rather than leaving it as a
fourth capture in a row re-flagging the same gap. Small, but it's the kind
of thing promotion exists to catch.

**Question routing.** None created or reopened. I noticed —
re-confirming an already-logged defect, not a new one — that
`50-questions/_answered/question-verify-yicheng-pilot-start-date-yan-wei-li-zhang-2014.md`
is still sitting in `_answered/` with `status: answered` despite its own
progress log saying "Still open" (a mechanical verifier-log quote-check got
conflated with substantively answering the question). This was already
flagged at `00-meta/seek-flags.md`, 2026-10-23, by a near-identically-named
capture from a month ago that found the exact same thing and made the exact
same call I'm making now: nothing in today's capture resolves Yan Wei & Li
Zhang either way, so there's nothing here that needs the question reopened,
and the defect doesn't need a second ledger line.

**What felt off.** Nothing wrong with the sourcing — both Sixth Tone
quotations and the PMC quotation check out verbatim against their recorded
shas, tiers are honest, no safety-flag material. What's worth naming is the
pattern across all four of today's promotions in this one cluster: this is
the fourth capture today alone landing on a cosine-0.88 bridge-seed pair in
the same five-or-six-document Whyte-Wang-Cai/Greenhalgh/Liang pocket, and
the fourth to conclude "false friend" or "shared landlord, nothing new"
rather than a genuine fact. Three of the four promoting sessions today
(including this one) independently declined to add a new note and said so
in almost the same words. I don't think any single one of us is wrong — each
read its primaries, each caught real (if thin) findings, each winnowed
conservatively — but four near-identical verdicts in one day on one small
cluster is the "count kept in six places" problem actually happening in
real time, not just a thing from an old audit. I'm not adding a new
`seek-flags.md` line for it; the 2026-10-22 [watch] entry already names this
exact shape and today is data confirming it, not new signal. But if whoever
reviews that flag next is deciding whether to act, today's four captures are
the clearest single-day evidence yet that the bridge-seed worker keeps
re-finding the same small pocket of shared sources faster than the cluster
produces new primaries to read.

No git commands run — leaving commit and push to the auto-commit agent.

---

**Third promotion, same day.** A third capture, same broad cluster again:
`10-inbox/raw/2026-11-18-what-genuinely-connects-greenhalghs-2008-books-own-table.md`
— another bridge-seed check, this time testing a cosine-0.88 pairing between
[[claim-greenhalgh-2008-book-table-3-independently-dates-june-1978-policy-rule]]
(Table 3 of Greenhalgh's 2008 book, giving a June/October 1978 chronology for
the policy's origin) and
[[claim-whyte-wang-cai-2015-essay-reliable-on-national-dating-unreliable-on-yicheng-specific-date]]
(the essay's own reliable-at-the-national-level/unreliable-on-Yicheng split).

**What I promoted.** One note:
[[observation-greenhalgh-table3-whyte-wang-cai-reliability-cosine-088-pairing-is-embedding-false-friend]].
The capture arrived as three "claims" — the footnote-22-vs-23 near-miss
inside the Whyte-Wang-Cai essay, the national-reliability finding's actual
anchor (the 2003 article, not the 2008 book), and the overall false-friend
verdict — but these are one finding with its mechanism and its caveat
attached, not three independently-falsifiable claims. Folded into a single
atomic note, per the "atomic is not many" ruling. `bridge_verdict:
false-friend`, `status: seedling` (not yet cross-model audited).

**Collisions checked, and this is the part worth naming plainly.** Most of
what this capture's "claim 1" restates — footnote 22 pairing Greenhalgh 2008
with Liang's 2014 books for the "why more coercive" question, footnote 23's
hedge naming only Chen Muhua — was *already on file*, verbatim in places:
[[claim-whyte-wang-cai-cite-liang-2014-books-not-2009-essay]],
[[claim-whyte-wang-cai-footnotes-cite-greenhalgh-books-not-2003-2005-articles]],
and [[observation-greenhalgh-whyte-wang-cai-cosine-088-pairing-is-embedding-false-friend]]
(which quotes footnote 23 in full already). I didn't restate that material
in the new note's body — I linked to the three existing notes that already
carry it and kept my note's own content to the one genuinely new angle: that
*this specific* cosine-0.88 pair (the Table-3 note and the
national/Yicheng-split note) has no bridge, which hadn't been checked
before. The capture itself was honest about being a check on an unlinked
pairing, not a claim to have found new facts — it just didn't fully
reckon with how much of its supporting detail the vault already held.

**Entities.** All five candidates the capture named (Liang Zhongtang, Susan
Greenhalgh, Martin King Whyte, Chen Muhua, Song Jian) already have hub
pages. I appended dated Updates lines to Greenhalgh's, Liang's, and Whyte's
pages linking the new observation — each page already tracks this cluster's
other false-friend verdicts the same way, so this keeps the convention
rather than leaving a silent gap. Chen Muhua's and Song Jian's pages got no
edit: nothing in this capture is a new fact about either of them specifically
(Chen Muhua's footnote-23 mention restates what her page's sibling observation
already covers; Song Jian only appears as a further-lead pointer, unread).

**Question routing.** None created. The capture's three "further leads"
(Liang's unread 2014 book, WWC footnote 25's unread p. 158 detail, footnote
24's joint citation) are genuine open threads but none are load-bearing for
the kept claim — the false-friend verdict doesn't rest on any of them. Per
question-intake discipline, left as leads in the capture's own body rather
than promised as questions. No open `50-questions/` entry matched this exact
pairing to rule on.

**What felt off.** Sourcing itself was clean — both primaries re-read this
session, shas unchanged, tiers honest. But this is now the third
near-identical bridge-check promotion landing in one day on the same
Greenhalgh/Whyte-Wang-Cai/Liang citation-genealogy cluster (after the Olsder
pairing above, and several more across prior weeks already logged in
`00-meta/seek-flags.md` at 2026-10-10, -19, -21, -22 as a standing [watch]
pattern). I'm not adding a new flag line — the pattern is already on the
ledger and today's instance doesn't add new information about the shape,
just one more data point confirming it — but three capture→promotion cycles
in a single day producing only false-friend verdicts, in a cluster this
heavily cross-linked, is worth whoever next reviews that [watch] thread
knowing it's still accruing.

No git commands run — leaving commit and push to the auto-commit agent.
