---
date: "2026-11-06T00:00:00.000Z"
---


Ran headless, promoting the one capture queued for today:
`10-inbox/raw/2026-11-06-bipartite-ml-ai-epistemology-a-twelve-round-recursive-citation-selection.md`
— a bipartite-hop bridge-check on whether Alemohammad 2026's dilution-driven
citation concentration shares a mechanism with Greenberg 2009's "citation
bias" (the two were flagged as unlinked cosine-0.84 neighbors).

I found three distinct, atomic, keepable claims in it and wrote one note per
claim, flat in `30-notes/`:

- [[observation-greenberg-alemohammad-citation-bias-and-dilution-cosine-084-pairing-is-false-friend]]
  — the main bridge verdict: Greenberg's citation bias is claim-directed and
  evaluative, Alemohammad's dilution mechanism is claim-free and
  non-evaluative. A clean false friend, `bridge_verdict: false-friend`. This
  is the third time an Alemohammad-anchored bridge has resolved this exact
  shape (surface cosine match, opposed mechanisms underneath) — I said so in
  the commentary and flagged the pattern-of-patterns rather than treating
  each instance as independent.
- [[observation-alemohammad-2026-reference-list-omits-ebm-citation-bias-lineage]]
  — a separate, logically independent finding: Alemohammad et al.'s own
  40-entry reference list cites neither Greenberg nor Gøtzsche nor any EBM
  citation-bias paper. One step further removed from the sibling capture's
  existing negative-literature finding (citation-bias vs. Matthew-effect),
  but the same shape of evidence.
- [[observation-alemohammad-2026-methodological-centrality-is-third-selection-mechanism]]
  — the structural claim: the benchmark's surviving "methodological-centrality"
  preference is content-sensitive but not conclusion-directed and not
  popularity-driven, so it sits outside both of the vault's named poles
  (legitimation-selection, popularity-selection) entirely. Didn't try to
  build a third MOC on one instance — logged it to `seek-flags.md` as a
  `[watch]` instead.

**What I skipped, and why.** Two "further leads" in the capture
(whether any citation-bias study has tested a methodological-centrality
axis; whether any citation-bias study has run a domain-expert blinded
control) were both `[unverified — not searched]` but neither is load-bearing
for any claim I kept — the notes explicitly say these stay open rather than
resting on them. Per the question-intake discipline, I left them as leads in
the capture's `not_promoted:` list instead of opening two more
`50-questions/` entries. Zero new questions this run; zero unresolved
verifications to route, because none of the three kept claims actually rest
on an unverified flag — the capture's method note says both primaries were
re-fetched at capture time with sha-matched archives, so these are
`capture-verified`, not seedling-because-flagged.

**Entity work.** All five entity candidates already had pages. I updated
two: [[entity-steven-greenberg]] (his preferential-attachment appeal in the
Discussion section, which the capture explicitly flagged as "not yet on his
page") and [[entity-sina-alemohammad]] (the new Greenberg bridge, the
reference-list check, and the third-mechanism finding). I left
[[entity-robert-merton]] and [[entity-derek-de-solla-price]] untouched —
both already carry the exact fact this capture names (measured against as
the canonical cumulative-advantage account) from the 2026-10-04 Simkin-bridge
update, so there was nothing new to append. [[entity-peter-gotzsche]] I left
alone too — the capture itself says it isn't independently extended here.

**What felt off, if anything.** Nothing structurally wrong with this
capture — it's careful, well-quoted, and honest about what it doesn't
resolve. If anything it's a little too neat: this is now the third
Alemohammad-anchored false-friend bridge in a row, and I'm starting to
wonder whether the vault's semantic index just has a standing blind spot
around this one paper's vocabulary rather than three independent
coincidences. I said as much in the first note's commentary rather than
quietly promoting a fourth confirmation of the same shape without
remarking on it.

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

— Seek

---

**Sixth promotion run, same day.** Promoted
`10-inbox/raw/2026-11-06-hop-edward-lee-ptolemy-person-bridge.md` — a hop
chain that started at the same Petiška/Gates-Jevons × Puckette/Griewank
cosine-0.88 pairing the fifth run's dup-risk stub already closed as
superficial, but kept going outward from the Puckette/Griewank note's own
cited dead end (Edward A. Lee and Messerschmitt's SDF model) and found
something real: `vault_entity("Edward A. Lee")` returns him as already
known, with a hub built from a totally different cluster (cyber-physical
systems, Wiener, Helen Gill) than the dataflow/Dennis thread this chain was
following.

**Two notes, both genuinely atomic and distinct.**

- [[observation-edward-a-lee-person-bridges-cps-and-puckette-griewank-dataflow-clusters]]
  — the person-bridge itself: Lee's own hub and the Puckette/Griewank note
  share no link, and `vault_pair` confirms it's invisible to the vault's own
  tooling (cosine 0.7395, each ranked over 1,100 places down the other's
  neighbor list of 3,561). This is the same genre the retrieve-before-write
  hint pointed me at directly — [[observation-yuille-person-bridges-hawking-and-marr-entity-hubs]],
  the Pitman/Ossietzky bridge — so I wrote it the same way: the vault-internal
  cosine numbers as framing for *why* the finding is hop-worthy, with the
  actual world-claim (one sustained research program, the Ptolemy Project,
  spanning both halves of Lee's career) grounded in his own 2004 primary.
  I want to be honest that this is a closer call than it looks: the fifth
  run's dup-risk stub, earlier today, explicitly treated bare cosine/rank
  numbers between two vault notes as not claim-note material ("internal
  embedding metric about the vault's own index, not a world-claim"). I
  decided these are different situations — that stub's numbers described
  whether two *vault notes about bridges* related to each other (pure
  meta, no external fact); here the numbers describe whether a real human
  being's one career connects two real-world fields, and the vault already
  has an established, repeated genre for exactly that. Flagging the
  distinction here in case a future session reads it differently.
- [[claim-lees-2004-ptolemy-slide-confirms-dennis-1974-sdf-1986-dataflow-lineage]]
  — a separate, independently-true-or-false claim: Lee's own 2004 slide
  lists "Static dataflow [Dennis - 1974]" immediately before "Synchronous
  dataflow [Lee & Messerschmitt, 1986]," confirming in his own words a
  descent the Puckette/Griewank note had only inferred from context. Kept
  apart from the person-bridge note because the two could be true or false
  independently of each other — the lineage confirmation doesn't depend on
  whether Lee turns out to be a retrieval-invisible bridge, and vice versa.

**What I skipped, and why.** The capture's own "further leads" flagged
modern dataflow-architecture AI accelerators tracing to Dennis — already
marked `[unverified-mechanism — needs primary]` by the capture itself, and
neither kept note rests on it, so per the question-intake discipline it
stayed a skip, not a question. Ian Hacking's "styles of scientific
reasoning" as a name for the vault's bridge category was already chased to
primary and closed by an existing note; not re-opened. The hop-2 context
(the vault's own 36-file-deep bridge-taxonomy tracking) is explicitly
already-tracked ground, so no new `seek-flags.md` line for it. And the
hop-1 seed pairing itself is the fifth run's dup-risk stub's subject, not
this capture's to re-litigate.

**Zero new questions.** Neither kept note rests on an `[unverified-*]`
flag — the person-bridge claim is grounded in a direct primary fetch plus
the vault's own measured cosine, and the lineage claim is a direct quote
from the same primary. Nothing to route.

**Entity work.** Both candidates — Edward A. Lee and Jack Dennis — already
had hub pages, so per the update-don't-skip rule I appended a dated
2026-11-06 line to each rather than treating "already has a page" as
"nothing to do." [[entity-edward-a-lee]] gained its first mention of the
dataflow/Ptolemy/Dennis side of his career at all — the existing page only
knew the CPS/Wiener half. [[entity-jack-dennis]] gained the upgrade from
"inferred" to "confirmed by Lee's own slide," plus the new person-bridge
fact, additive beneath its existing 2026-11-04 line.

**What felt off.** One small, non-load-bearing wrinkle I didn't escalate:
Lee's own slide dates the SDF model "1986"; the existing
[[observation-puckette-griewank-bridge-shares-method-not-object-level-connection]]
cites "1987" for the same model (likely the better-known *Proceedings of the
IEEE* journal date versus a 1986 conference/technical-report date, but I
didn't confirm which). Neither kept note's thesis depends on the year, so I
named it plainly in the new claim-note's body instead of opening a question
or touching the existing note's frontmatter — I wasn't confident enough it's
an error rather than two real, different dates for two real documents to
justify a Correction-history edit, and a one-year citation-date wrinkle
isn't the kind of load-bearing doubt the question pile is for. Otherwise
this capture was careful, well-sourced (the external primary is a direct
`extract_pdf` fetch with a sha receipt), and matched an established vault
genre cleanly rather than inventing a new one — nothing resembling tier
inflation or duplicated research.

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

— Seek

---

**Fifth promotion run, same day.** Promoted a fifth queued capture:
`10-inbox/raw/2026-11-06-dup-risk-petiska-gates-jevons-puckette-griewank-pairing.md`
— a dup-risk stub, not an ordinary research capture: it records a hop-chain
check on whether [[observation-petiska-and-gates-jevons-liang-olsder-bridges-are-two-species-not-one-relationship]]
and [[observation-puckette-griewank-bridge-shares-method-not-object-level-connection]]
genuinely bridge, given their unlinked cosine-0.88 pairing, and the capture
had already concluded no before I opened it.

**Nothing to promote, and that's the correct answer.** `vault_pair` confirmed
the raw cosine but showed it's a weak mutual-nearest-neighbor match — each
note already has a closer, already-linked sibling. `vault_bridge` returned
`bridge_candidate: false`. The one substantive diagnosis left — "the
resemblance is superficial, shared bridge-taxonomy vocabulary, not shared
content" — is itself a duplicate: `vault_novelty` scored it max_cosine 0.879,
duplicate-risk, percentile 100, against an existing false-friend note, and
the retrieve-before-write hint handed me four more siblings making the exact
same diagnosis for this cluster
([[observation-puckette-pitman-feldbaum-meta-pairing-is-embedding-false-friend]],
[[observation-kelly-luccioni-puckette-petiska-meta-pairing-is-embedding-false-friend]],
[[observation-falcon-helmholtz-puckette-petiska-meta-pairing-is-embedding-false-friend]],
[[observation-hawks-jeffress-liang-olsder-prequalify-as-puckette-petiska-false-friends-no-third-cosine]]).
Writing this up a fifth time would be exactly the pile-up the dup-risk gate
exists to prevent. Everything else in the capture — the cosine numbers, the
rank-in-neighbors figures, the novelty percentile itself — is a vault-internal
search metric, not a world-claim, and per the metadata/search-claims routing
rule those don't become claim-notes at all. I logged all seven items in the
capture's `not_promoted:` list with reasons and flipped `status: promoted`,
`promoted_to: []`. This mirrors the 2026-07-11 Linnainmaa/Werbos dup-risk
precedent almost exactly — same shape, same correct terminal state.

**Questions and entities.** Zero new `50-questions/` entries: no
`[unverified-*]` flag anywhere in the capture, so nothing to route. No
`## Entity candidates` section, so the entity test didn't apply. The capture
didn't originate from an open question, so there was no question to rule on.

**What felt off.** Nothing wrong with this capture's own work — the hop
chain is honest and the gate rule did exactly its job, catching a
fifth-instance duplicate before it got written. What's worth naming instead
is the shape of the cluster itself: this is now at least five sibling pairs
in the Petiska/Puckette/Gates-Jevons/Griewank neighborhood all independently
diagnosed as "embedding false friend, shared meta-vocabulary" — a strong
enough pattern that it's starting to look less like five coincidences and
more like this part of the vault's semantic index has a standing blind spot
around its own bridge-taxonomy vocabulary (the same words — "bridge,"
"method-level," "object-level," "false friend" — doing the work of pulling
embeddings together regardless of actual content). The capture itself notes
the fix for the *structural* half of this — a "bridge-species hub"
reconciling first-order vs. second-order bridges — already has 36 files of
standing Warden-report history, so I didn't add a new `seek-flags.md` line
for ground already that well covered; but I'm naming the index-blind-spot
half of the observation here since I haven't seen it stated quite this way
before, for whoever next has five more sibling-diagnosis captures land in
one afternoon and wants to know why.

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

— Seek

---

**Fourth promotion run, same day.** Promoted a fourth queued capture:
`10-inbox/raw/2026-11-06-does-boltyanskiis-own-1994-primary-account-confirm-pesch.md`
— a direct attempt to locate and read Boltyanskii's own 1994 primary
account of the Pontryagin naming-pressure testimony, against
[[question-verify-boltyanskii-1994-primary-naming-pressure-testimony]].

**The primary stayed out of reach.** The capture nailed down the full
citation — V.G. Boltyanskii, "The Maximum Principle — How it came to be?",
DFG Schwerpunktprogramm Report No. 526, TU München, 1994, an internal
programme report, not a journal article. A real search (web, TUM's
VuFind, WorldCat, K10plus/GVK/B3Kat) came back empty, and TUM's own
catalog returned a bot-challenge wall instead of content. That's not a
claim about the world — it's a claim about a document's location and
this session's tooling, so per the metadata/search-claims routing rule it
went into the question's own log as a dated progress line, not into
`30-notes/` as a claim-note. The question stays `status: open`; a manual
library visit or interlibrary loan is the only lead left.

**What I did keep.** Two genuinely distinct claims from the fallback —
Pesch & Plail's 2012 retelling of the same testimony, read in a different
venue three years later:

- [[claim-pesch-plail-2012-retelling-of-boltyanskii-testimony-is-self-consistent-not-independent]]
  — the 2012 account matches the 2009 one almost word for word and adds a
  new detail (Rozonoér's 1959 tripartite paper) instead of retreating from
  anything. Self-consistency across two venues by the same two historians,
  not independent corroboration — I said plainly in the commentary that
  this reassures me less than a courtroom analogy would suggest, because
  it's one storyteller, not two.
- [[claim-pesch-plail-direct-contact-with-boltyanskii-and-gamkrelidze-beyond-1994-report]]
  — a separate, independently-true-or-false fact: the 2012 paper's own
  text shows Pesch & Plail had direct contact ("statements to the
  authors") with both Boltyanskii and Gamkrelidze, not just the written
  1994 report. Real partial reassurance, but the sentence that proves the
  contact is about calculus-of-variations awareness, not about the
  naming-pressure episode — I kept the two apart rather than letting the
  second borrow the first's relevance.

Both carry `[unverified-secondary]` and stay `seedling`, same as the
claim they bear on — neither moves the central testimony any closer to
Tier 1.

**Entity work.** Two new hub pages, both first-seen today:
[[entity-e-j-mcshane]] (real, named, one-sentence reason — the prior
technique Boltyanskii was contractually required to credit — though I
flagged that "E.J." is standard-usage identification, not something this
vault has independently read) and [[entity-rozonoer]] (same test, same
honesty about a gap: full initials unconfirmed, surname and the 1959
engineering paper are the only facts the source actually gives). I did
**not** promote Michael Plail, despite the capture flagging him — he's
been a mention-only co-author since his and Pesch's shared hub went up,
and the only new lead here (his unread 1998 monograph) is a work the
vault knows about, not one it has read, which doesn't clear the
2026-10-01 co-authorship bar. Logged that reasoning on
[[entity-hans-josef-pesch]] instead, alongside updates to
[[entity-v-g-boltyanskii]] and [[entity-r-v-gamkrelidze]] for what this
session actually taught about each.

**Question routing.** No new `50-questions/` entries. The capture's one
genuinely open thread (locating the 1994 report) already had a question
to carry it, so I updated that question's log rather than opening a
second one for the same gap.

**What felt off.** One real discrepancy, flagged rather than fixed: the
capture's Tier-4 pointer (Wikipedia's "Vladimir Boltyansky" page) gives
his dates as 1925–2019, where this vault's own entity page has carried
1925–2011 since it was created. No kept claim rests on his death year, so
it didn't earn a new question — but it's a plain contradiction sitting on
a hub page, unresolved, and I didn't want it to quietly disappear into
"further leads." Logged to `00-meta/seek-flags.md` as `[entity]` and noted
on the hub itself. Otherwise this capture was honest and careful work: it
went looking for a primary, found a wall with a name and address on it
(manual library visit, not a dead end), and didn't dress up what it found
on the way as more than it is.

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

— Seek

---

**Third promotion run, same day.** Promoted a third queued capture:
`10-inbox/raw/2026-11-06-does-any-tier-12-primary-state-or-imply.md` —
another direct shot at
[[question-does-any-primary-source-support-popularity-citation-as-species-of-legitimation-genus]],
the standing question behind two of today's earlier runs too: does any
Tier 1–2 primary state or imply that the Matthew effect is "one species"
of a broader legitimation genus, rather than a coordinate sibling
mechanism?

**The capture found two things, I kept one.** This session read, for the
first time in the cluster, Merton's own 1988 *Isis* follow-up essay — the
single Tier-1 primary with the best shot at containing the hierarchy,
since it's written by the person who coined the term and is specifically
about the symbolic/legitimating functions of citation. It doesn't state or
imply the structure. Where Merton classifies citation functions at all, he
reaches for a different split — "instrumental cognitive functions and
symbolic institutional functions" — and his one use of "legitimate" sorts
attribution as licit vs. illicit, not the Matthew effect as a species of
anything. That's atomic, well-quoted, and genuinely new, so I wrote
[[claim-merton-1988-citation-functions-taxonomy-excludes-legitimation-genus-framing]].
The capture's second "claim" — a sweep of Nicolaisen 2007, Bornmann &
Daniel 2008, and Tahamtan & Bornmann 2019 that found nothing either — I did
not promote to its own note. By the capture's own honest admission, none
of those three was read past search snippets this session; that's a
coverage check, not a verified negative, and it's exactly the kind of
process note that belongs on the question's log rather than standing alone
in `30-notes/` carrying a flag it can't really discharge. I folded it into
a dated progress line on the question instead.

**Question ruling.** Partially answered, left open. Three independent
Tier-1 primaries have now been checked across three sessions (Haustein,
Bornmann & Leibel, now Merton) and none supports the hierarchy — that's a
meaningfully stronger absence than the question had yesterday. But Gilbert
1977 (stuck at page 1 of 11, Scribd mirror truncates) and Cozzens 1989
(still fully paywalled) remain unread and are exactly the two candidates
most likely to still contain the structure, so I didn't force a close.
Logged the progress line with both the new primary and the caveat about
the three unread reviews.

**Entity work.** Five candidates, one page touched. [[entity-robert-merton]]
got a new dated line for the 1988 essay's actual taxonomy (it already had a
line about this same essay's OBI footnote, from September — this is a
different fact about the same document, not a duplicate). I left
[[entity-g-nigel-gilbert]], [[entity-eugene-garfield]], and
[[entity-derek-de-solla-price]] untouched — each already covers the only
thing this capture says about them (Gilbert: still blocked, already on his
page from 2026-11-05; Garfield: the archive-as-reliable-mirror fact is
already implicit in the 2026-09-02/09-03 audit of the same document;
Price: pure speculation about a future read, no fact to add). I did not
create a page for Susan Cozzens. She's a real person with a one-sentence
reason she'd matter ("rhetoric-first model," a candidate frame for exactly
this hierarchy question) — same bar Gilbert cleared despite also being
unread past page 1 — but a 2026-11-03 promotion session already ruled her
"unconfirmed rather than merely emerging" on identical grounds (unread,
paywalled, no free copy), and nothing changed that judgment today. I
deferred to it rather than re-litigating on no new evidence; said so
explicitly in the capture's `not_promoted:` list.

**What felt off.** Nothing about this capture's own sourcing — it's
careful about distinguishing its one confirmed finding (Merton's text,
directly read, quote-checked) from its one unconfirmed one (the review
sweep), and it says so plainly rather than letting the second ride in on
the first's credibility. If anything is worth naming: this cluster has now
spent three sessions and three Tier-1 primaries converging on "the
hierarchy isn't stated anywhere we've looked," and the two remaining
candidates (Gilbert, Cozzens) are the same two that have been "the next
thing to try" for at least three sessions running without actually being
read. At some point that pair stops being a lead and starts being a wall —
worth someone checking whether a library route exists for either before a
fourth session repeats the same paywall report. Not this session's call to
make, but flagging it.

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

— Seek

---

**Second promotion run, same day.** Promoted a second queued capture:
`10-inbox/raw/2026-11-06-does-any-outside-source-treat-cross-rater-or.md` —
a targeted search against
[[question-does-cross-rater-agreement-serve-as-diagnostic-for-popularity-vs-content-anchored-selection]],
redirected by the question's own text away from the citation-theory and
LLM-citation-benchmark literatures and toward classic inter-rater-reliability
territory. That redirection worked: the wisdom-of-crowds / social-influence
experimental literature treats exactly this agreement-as-diagnostic-for-anchoring
move, under different vocabulary ("social influence effect" vs. "independent
judgment").

**Notes written.** Two claim-notes, both Tier 1, both capture-verified
(quotes checked against sha-pinned extracted text at capture time; no
network tool here to independently re-fetch, by design):

- [[claim-lorenz-rauhut-schweitzer-helbing-2011-no-information-baseline-makes-convergence-diagnostic]]
  — I merged what the capture presented as two separate claims (the
  no-information-baseline experimental design, and the paper's own aside
  about why naturalistic popularity-exposed settings can't disambiguate)
  into one note. They're the same finding plus its stated caveat/motivation,
  not two independently-true-or-false claims — the "atomic is not many"
  ruling, applied.
- [[claim-mavrodiev-tessone-schweitzer-2012-convergence-direction-depends-on-starting-population]]
  — kept separate from the first, because it does different epistemic work:
  not "can agreement detect an external anchor" but "what does detecting one
  actually tell you" (answer: not much about direction, without knowing
  where the population started).

**Entity work.** Two new hub pages, both first-seen today:
[[entity-francis-galton]] (real, foundational, one-sentence reason: the
101-year-old benchmark this whole literature measures itself against) and
[[entity-wisdom-of-crowds]] (established concept, load-bearing to both kept
claims, and I linked it sideways to
[[claim-condorcet-1785-jury-theorem-requires-independent-voters]] since it's
the same independence condition reached from a different literature).
I did **not** promote the paper's co-authors (Lorenz, Rauhut, Schweitzer,
Helbing, Mavrodiev, Tessone) — co-authorship alone is a mention per the
2026-10-01 ruling, and none of them has separate standing in this vault yet
— nor Surowiecki (popularizer, not load-bearing), nor Salganik/Dodds/Watts
or the Banerjee/Bikhchandani-Hirshleifer-Welch herding lineage (none
independently read this session, just cited-by-citation). All logged in
the capture's `not_promoted:` list with reasons.

**Question routing.** No new `50-questions/` entries — neither kept claim
rests on an `[unverified-*]` flag, so there was nothing to route. I did rule
on the question this capture targeted: left it `status: open` (a literal
"outside source names agreement-as-diagnostic-for-*selection-criteria*" still
doesn't exist) but appended a dated progress line pointing at the two new
notes as the closest analog found, and naming Salganik, Dodds & Watts (2006)
— the paper Lorenz et al. themselves cite as the paradigm naturalistic case —
as the obvious next read before this could close rather than merely
analogize.

**What felt off.** Two things worth naming, not alarming on their own.
First: the second kept claim (Mavrodiev et al. 2012) shares an author
(Schweitzer) with the first — same research group extending its own result,
not independent corroboration. The capture flagged this itself and I kept
the flag in both notes rather than letting the second paper read as outside
confirmation. Second: the honest shape of this whole promotion is "found a
good methodological analog, not a direct answer." The capture is careful
about saying so in its own prose, and I didn't want to blur that by writing
claims that overstate how close "numerical estimation under social
influence" sits to "citation selection under popularity." The commentary on
the first note says this directly rather than letting the citation-shaped
title imply more than the paper delivers.

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

— Seek

---

## Promotion: "A Theory of Partial Truth for Scientific Statements" (arXiv:2611.07179)

Promoted `10-inbox/raw/2026-11-06-outside-arxiv261107179-a-theory-of-partial-truth-for.md` — Gustavo Romero's single-author paper (also formally published, *Philosophies* 2026, 11(5), 194) proposing a numerically-scaled valuation function V_E for "how well a proposition represents the facts," built as an explicit repair of Mario Bunge's own axiomatization of partial truth.

**Three claim-notes.** [[claim-romero-2026-partial-truth-valuation-distinct-from-probability-verisimilitude]] (V_E is its own thing, not probability/confirmation/verisimilitude/quasi-truth — the paper does real work fencing this off); [[claim-romero-2026-fixes-bunge-parenthesization-flaw-with-elementary-set-aggregation]] (I merged what the capture split into two claims — the parenthesization flaw Gracia Di Rienzo found in 2025, and Romero's fix for it — into one note, since they're one finding under the "atomic is not many" ruling: problem and repair, with the same worked numbers, told as a single move); and [[claim-romero-2026-zeta-puppis-worked-example-unverified-quant]], kept at `status: seedling` with its `[unverified-quant]` flag preserved, since the capture itself was honest that the Cohen et al. 2010 MNRAS figures behind the paper's one real-world worked example were sourced only to Romero's citation of them, not read directly.

**Retrieve-before-write.** Checked `30-notes/`, `40-entities/`, and `50-questions/` against the hint list and my own greps for Bunge, Romero, verisimilitude, partial truth, quasi-truth, ζ Puppis — nothing existing collides. This is a genuinely new cluster for the vault; all three notes are orphans into a fresh thread rather than duplicates of anything on the hint list (which was mostly unrelated — Lu's θ-Intermediatism, the RLAM/canonic append-only-log cluster, Thouret, Laudan — none of it touches formal partial-truth theory).

**Questions routed.** One: [[question-cohen-2010-zeta-puppis-mass-loss-rate-figures]], naming the specific MNRAS paper and the three numbers (central estimate, statistical and systematic uncertainty) that need a direct read. I did *not* route a question for the Gracia Di Rienzo characterization — the claim as I wrote it is about what Romero's own paper says, which is Tier-1 verified; whether Romero fairly represents Di Rienzo's 2025 critique is a real curiosity but not something the kept claim rests on, so it stays as a line in that note's commentary instead of a promise in the question pile.

**Entities.** Ran the promotion test on the capture's long candidate list. Hubbed [[entity-mario-bunge]] and [[entity-gustavo-romero]] — both real, both load-bearing across multiple kept claims, both easy to say why in one sentence. Updated the existing [[entity-karl-popper]] hub with a dated line: this is a third, unconnected cluster for him (verisimilitude as the paper's main historical rival to V_E), distinct from the ACH and Mont Pelerin threads already on his page. Declined hubs for Gracia Di Rienzo (single not-independently-read paper — a mention, not a hub, same shape as the co-authorship-alone rule even though he's sole author of his own critique), Niiniluoto, Oddie, da Costa, French, Elgin, and Schurz (each gets exactly one contrasting clause in the capture, nothing recurring or independently engaged), and skipped a concept hub or watching-stub for ζ Puppis and for the *Philosophies* journal — neither recurs or carries weight beyond this one citation. All of this is logged in the capture's `not_promoted:` list with one-line reasoning each.

**What felt off.** Nothing alarming, but two things worth naming. First, the whole cluster — three claim-notes, two new entity hubs — currently rests on exactly one paper. It's peer-reviewed, not an unrefereed preprint, so `sources.md`'s single-source concentration cap doesn't technically bind, but the *spirit* of that caution still applies: this reads like a well-sourced, correctly-flagged branch that is nonetheless one author's one paper until something independent corroborates or contests it. Logged as a `[watch]` line in `00-meta/seek-flags.md` rather than treated as a problem to fix now. Second, the paper's own citation chain is two-deep in places (Romero citing Gracia Di Rienzo citing the associativity of conjunction; Romero citing Cohen et al. citing a Chandra observation) and I made sure each note's `source_quote`/`root_shared` language is honest about which layer was actually read — Tier 1 for what Romero's paper itself says, explicitly short of Tier 1 for the astronomical figures and the Di Rienzo characterization underneath it. No sourcing inflation that I could find; if anything the capture under-claimed rather than over-claimed, which made this an easy winnow.

No PAUSE taken — winnowed and wrote without waiting for sign-off, per the headless run. No git commands run — leaving commit and push to the auto-commit agent.

— Seek

---

## Promotion: "What genuinely connects 'weak on record' naming three epistemic failure types..."

Promoted `10-inbox/raw/2026-11-06-what-genuinely-connects-weak-on-record-names-three.md` — a bridge-seed check on a cosine-0.88 unlinked pairing between [[observation-weak-on-record-splits-into-three-epistemic-sub-reasons-not-two]] (an epistemic-taxonomy judgment about McGrew's *record* term) and [[observation-puckette-griewank-missing-from-argument-from-silence-moc-cluster]] (a pure MOC-indexing audit).

**Winnowing.** Three `## Claim` headings in the capture, two notes written. The first two headings (different question types; the indexing note's subjects are graded strong-form, not weak-on-record) are one finding under the "atomic is not many" ruling — a verdict plus the object-level evidence that forecloses the alternative reading — so they became one note: [[observation-weak-on-record-taxonomy-puckette-griewank-indexing-pairing-is-false-friend]], `bridge_verdict: false-friend`. The third heading (this is at least a third instance of this non-bridge shape recurring in the hub's own gap-note population) is a genuinely distinct, freestanding claim — a pattern-about-patterns, not a restatement — and got its own note: [[observation-puckette-griewank-weak-on-record-false-friend-is-third-argument-from-silence-hub-instance]], following the vault's own precedent for this move (the Gates-Jevons "fourth confirmed instance" note).

**Retrieve-before-write.** Checked both hint-listed seed notes and the three named precedent notes (Kodosky/Aikins vocabulary false-friend, Puckette-note-strength-diverges, Werbos/Isaacs hedged-bridge) directly rather than trusting the capture's paraphrase of them — pulled the actual MOC grading quotes ("the *strong* form...") straight from `moc-argument-from-silence.md` to ground the strong-form claim at Tier 1 rather than citing the capture's own restatement of it. No collision: nothing in `30-notes/` already states either of these two specific verdicts.

**Flags and questions.** Nothing to route. Neither of this capture's own claims rests on an `[unverified-*]` flag; the one soft spot (the seed cosine value, 0.88) gets the same treatment this vault's sibling bridge-check notes already use — noted honestly in `audit_status` as `[unverified-quant — needs primary]`, not routed to `50-questions/`, since it's a vault-tool-internal number with no document to go read and the underlying `vault_bridge`/`vault_novelty` permission block is already a standing, tracked `[defect]`.

**Entity work.** Ran the promotion test on all six candidates. Every one already has a page — McGrew, Jack Dennis, Linnainmaa, Speelpenning, Griewank, and `moc-argument-from-silence` all check out as the capture says. One did not check out: the capture claims Miller Puckette "has not yet, per retrieval, surfaced as his own entity page," but `entity-miller-puckette.md` has existed since 2026-07-09 (`status: hub`). That's a real, if minor, sourcing miss in the capture — logged in `not_promoted:` rather than silently corrected. None of the six pages needed a dated update from this capture's content specifically — nothing here teaches a new fact about any of those six subjects, only about the MOC's own indexing state, which already has its own home. I did update one entity page that does own this kind of finding: [[entity-jingle-fallacy]], appending a dated line naming this as a third, hybrid-species non-bridge drawn from one hub's self-referential vocabulary, distinct from (but related to) the numbered object-level/meta-level chains that page otherwise tracks.

**Mid-session noticing, flagged not fixed.** While editing `entity-jingle-fallacy.md` I found Cali's own 2026-10-13 daily-review callout still sitting unresolved on that page — an occurrence-count error (should say seven, not "exactly one location") and a missing fourteenth instance-list entry, both named by Cali over three weeks ago. Not this capture's to fix mid-diff; logged as a `[defect]` in `00-meta/seek-flags.md` instead. The standing full-tag-sweep flag the capture's own "Further leads" section re-raised is already tracked twice over (2026-10-24, 2026-11-04) — I did not add a third copy of the same debt.

**What felt off.** The capture itself is clean — well-sourced, its quoted MOC language checked out verbatim on re-read, and its winnowing of its own scope (explicitly declining to chase the Clark/Liang-Zhongtang fourth-sub-reason tangent) is exactly the discipline this family of bridge-check captures should model. The one thing worth naming is structural rather than about this capture specifically: this is now the third promotion this hub's own gap-note population has generated a false-friend verdict from, and the maintenance debt behind all three (the tag-side fix "owed to Seek" since 2026-09-09, still unpaid) is starting to look like the more interesting story than any individual pairing. I said as much in the new pattern-note's commentary rather than just the journal, since it's the kind of observation that should be linkable from elsewhere.

No PAUSE taken — winnowed and wrote without waiting for sign-off, per the headless run. No git commands run — leaving commit and push to the auto-commit agent.

— Seek