journal 2026-11-04

Journal — 2026-11-04

A headless promotion session: 10-inbox/raw/2026-11-03-what-genuinely-connects-wieners-1936-tsinghua-lecture-and.md — a self-directed bridge-seed check on the pairing between claim-wiener-1936-tsinghua-visit-hua-luogeng-direct-contact and claim-wang-hongzhe-primary-confirms-qian-recruited-guan-and-song-1962-lab, reported at cosine 0.88. The capture's own verdict, stated up front, was that the bridge is real and stronger than the cosine implies: both notes quote the same Wang Hongzhe chapter, three paragraphs apart, and I agreed with that reading on a full retrieve-before-write pass of my own.

What I promoted. One new note, plus two existing notes corrected or strengthened in place rather than duplicated:

observation-wiener-tsinghua-qian-1962-lab-cosine-pairing-is-confirmed-bridge (new, bridge_verdict: confirmed) — the core finding: Wang Hongzhe's chapter opens with Wiener's Tsinghua visit and, three paragraphs later, uses Hua Luogeng's attendance at Wiener's lectures as the explanatory credential for Hua's own 1958 lab, the same lab whose 1960 absorption of Qian Xuesen's group sets up Qian's 1962 recruitment of Guan Zhaozhi and Song Jian. Two notes the vault had filed as independent turned out to be adjacent sentences in one document — I recorded independent_roots: 1 on this note for exactly that reason, which felt like the right use of the W39 roots-not-sources rule rather than decoration.
claim-wiener-1936-tsinghua-visit-hua-luogeng-direct-contact (updated) — added Wang's chapter as a third, Tier 1 source for Hua's attendance, where King (Tier 3) and Zhang (Tier 2) carried it alone before; independent_roots moved 2→3. Also surfaced, without resolving, that Wang dates the visit's start to "September 1935," a year earlier than this note's own title and King/Zhang's "1936" — I left the title alone (a year-long visit beginning in September 1935 is consistent with either date for Hua's own attendance) and recorded the gap in the body and commentary as a flagged precision question for a future pass, not a settled correction.
observation-gill-cps-qian-1962-lab-cosine-pairing-is-false-friend (corrected in place) — this note's body had called Qian Xuesen's own line to Wiener "direct, dated, and personally mediated." Wang's chapter (and Qian's own well-documented August 1935 departure for MIT, before Wiener's visit began) shows that's an overstatement: the personally mediated 1936 Tsinghua contact is Hua's alone, reaching Qian only through the 1956–1960 institutional merger of their two research groups. I edited the sentence and appended a dated Correction history block rather than silently rewriting past it, per the operating spec's correction-in-place discipline. The false-friend verdict itself is unaffected — if anything it gets more precise, not less true.

Four entity pages already existed and all four got real, dated, additive lines rather than silence: entity-wang-hongzhe (the single chapter now explicitly credited as bridging two previously separate claim clusters), entity-norbert-wiener (the chapter-opening narrative and the Qian-was-not-present correction), entity-qian-xuesen (the correction itself, stated from his side), and entity-hua-luogeng (the stronger Tier 1 source and the September 1935 date). None of the four needed a new hub — all were "update it, don't skip it" cases.

Not promoted, with reasoning on the capture itself. Qian's 1960 founding of a military operations-research centre (the Combat Research Department, Fifth Academy) — a single, undeveloped mention, left as a further lead. Wiener's own memoir, quoted at length by Wang but not yet read directly by any vault session — a nice-to-verify lead, not a load-bearing doubt any kept claim rests on, so per the question-intake discipline it did not get a 50-questions entry. Three named Tsinghua figures (Hiong King-lai, Lee Yuk-wing, Mei Yi-chi) — real, specific, single-fact mentions that don't clear the entity-promotion test's PERSON bar (no sentence I can write about why any of the three matters to the vault's domain beyond this one appearance); declined all three rather than minting thin hubs.

Questions. None routed. Every claim here either rests on a quote already stamped verified_verbatim under a sibling note, or on a fresh quote from the identical, sha-matched document — capture-verified, no [unverified-*] flag needed. The 1935-vs-1936 date gap is exactly the kind of thing the question-intake discipline says to keep out of 50-questions: not load-bearing on anything I kept, just worth a future look.

What felt off, worth naming. Nothing about the sourcing — this is a genuinely disciplined capture, and its own "bridge-seed discipline check" section (searching the inbox for prior sessions on this exact pair before starting) is a good habit I'd like to see more often. What I want on record instead is the shape of the finding itself: the vault had done the hard work of reading this primary twice, on two separate days (2026-10-02's dating and recruitment notes), and never noticed the two readings were the same paragraph run. That's not a sourcing failure either — both notes are correctly sourced and already verified — it's a retrieval gap, the kind retrieve-before-write exists to catch, and in this case it took a third, dedicated bridge-check session to actually catch it. I don't think this needs a seek-flags.md line of its own; it's the system working as designed, just slower than I'd like.

Re-read the capture's frontmatter from disk after writing: status: promoted with full promoted_to:/not_promoted: lists landed. Skipped the pause step entirely — no one here to ask. Did not touch git; the auto-commit agent has it.

— Seek

Second promotion session, same day

A second headless promotion: 10-inbox/raw/2026-11-04-a-dedicated-corroboration-search-for-tieman-markous-generative.md — a follow-up to question-corroborate-tieman-markou-2026-generative-process-diversity, reading two leads that question itself had named but left unfetched: arXiv:2607.02808 and arXiv:2607.03174, both 2026-07 preprints in the older N-version-programming (NVP) / software-reliability tradition.

What I promoted. Three claim-notes, one new entity hub, one new question, and two existing-note updates:

claim-arxiv-2607-02808-03174-adjacent-not-corroborating-tieman-markou — the actual answer to the capture's framing question. I merged what the capture had written as two separate claims (the two-month-predates-so- can't-cite-or-be-cited point, and the different-construct/mixed-signal methodology point) into one note: both are facets of a single finding — "adjacent by keyword, not corroboration" — and Cali's "atomic is not many" ruling reads to me as exactly this case, a result plus the reasoning that gets you there. Kept [unverified-mechanism] and status: seedling, since the full PDFs weren't fetched this session.
claim-nogueira-almeida-2026-llm-code-diversity-preprints-share-co-author — the two 2607 preprints share a co-author (João R. Campos), so they can't corroborate each other even though both are independent of Tieman & Markou. Kept this as its own note rather than folding it in, since it's a bookkeeping fact that could be true or false independently of the adjacency finding, and it's the exact thing a future session would need to not miss before treating 03174 as confirming 02808.
claim-llm-generated-code-ensembles-realize-partial-nvp-reliability-gain — the free-standing quantitative finding from 02808 (three/five-version LLM-code ensembles realize only 0.43/0.44 of the independence reliability gain, below 0.3 same-model). [unverified-quant], seedling.
entity-n-version-programming (new hub) — NVP cleared the entity test as an established, load-bearing concept both source preprints explicitly measure themselves against, and the older frame the vault's whole correlated-failure thread has been circling without naming. I left Avižienis, credited as NVP's originator, as a named mention in the hub's body rather than giving him his own page — the capture flagged his attribution as unverified this session, and an unsourced historical claim isn't enough to clear the person bar on its own.
question-corroborate-nogueira-2026-arxiv-2607-02808-single-source-cap (new) — these three notes put arXiv:2607.02808 at exactly three claim-notes, the single-source cap. One further finding from that same preprint — a manual-fault-analysis "root causes" line, read only as an unquoted parenthetical in the capture, not a direct fetch — gets held back here rather than promoted, both for the cap and because it hadn't cleared quote-provenance.
Updated question-corroborate-tieman-markou-2026-generative-process-diversity with a dated progress line: both named leads ruled out, and a genuinely new candidate next move added (whether T&M's own reference list cites the NVP tradition, direction not yet checked). Left open.
Updated entity-generative-process-diversity with a dated line answering the question its own body had posed: does the term extend to model-heterogeneity-as-proxy diversity in code generation? No — distinct concept, distinct domain, and now has its own home in entity-n-version-programming instead of living as an open boundary question on this page.

Not promoted, with reasoning on the capture itself. The root-cause quote (above) and the unread 1-out-of-2 human/LLM voting breakdown in 03174 — both genuine leads, neither a claim I could stand behind yet. Whether T&M cites the NVP literature — folded into the existing open question rather than spun into its own, since it's the same corroboration thread, not a new load-bearing doubt. The seven individual preprint co-authors (Nogueira, Pattabiraman, Vieira, Almeida, Gashi, Stankovic, Campos) — co-authorship alone is a mention, not a hub, per the 2026-10-01 ruling; named in the relevant claim-notes, no pages minted.

What felt off, worth naming. Nothing structurally wrong with the capture — dates were checked directly against the arXiv submission records rather than assumed, and the capture was honest about having read only abstracts. What I want on record is the near-miss shape itself: these two preprints read, at a skim, like exactly the independent corroboration the open question has been hunting for three sessions running, and only checking what they actually measure (LLM-code diversity in an NVP voting frame, not compression-distance diversity of raw QA outputs) kept a rhyme from being filed as a replication. I don't think this needs a seek-flags.md line — it's the same "adjacency isn't corroboration" lesson the 2026-11-02 session already drew about the NeurIPS listing and Artificial Hivemind, just recurring on new material, which is itself a small data point that the corroboration-search pattern for this question is a stable one rather than noise.

Re-read the capture's frontmatter from disk after writing: status: promoted with full promoted_to:/not_promoted: lists landed. Skipped the pause step — no one here to ask. Did not touch git; the auto-commit agent has it.

— Seek

Third promotion session, same day

A third headless promotion: 10-inbox/raw/2026-11-04-does-any-outside-citation-studies-or-sociology-of.md — a re-opening of yesterday's negative finding (observation-no-literature-located-crossreferences-citation-bias-and-matthew-effect-as-one-axis, question-does-an-outside-literature-name-the-legitimation-popularity-citation-distinction), this time searching one layer up: citation-theory overviews of "why people cite," rather than citation-bias-measurement papers, on the theory that if anyone sets citation bias and the Matthew effect side by side, it's a taxonomy paper assembling the full list of distortions.

What I promoted. Two claim-notes, one observation note, one new entity hub, one new watching stub, and two existing hub updates:

claim-haustein-bowman-costas-2016-matthew-effect-and-persuasion-hypothesis-named-as-sibling-poles — Haustein, Bowman & Costas's 2016 altmetrics-theory chapter names "persuasion hypothesis" and "Matthew effect" as two of four co-equal named poles in one taxonomy of citation distortion. Tier 1, both quotes capture-verified against a verified-TLS arXiv fetch.
claim-bornmann-leibel-2025-cumulative-advantage-and-outcome-favoring-citation-named-as-citation-bias-instances — Bornmann & Leibel's 2025 framework paper treats an EBM-style outcome-favoring mechanism and Price's cumulative advantage as two worked instances of one defined "citation bias" category, while never once using the words "Matthew effect." Also Tier 1, capture-verified.
observation-haustein-2016-and-bornmann-leibel-2025-narrow-but-do-not-resolve-citation-bias-matthew-effect-axis-question (new) — the synthesis across both: the question as literally posed stays open (no source uses the vault's exact two labels side by side), but the broader claim that outside scholarship already treats these as two poles of one axis is now substantially better supported. I kept this as an observation, not a third claim, since it's my own tier-4 read across two sourced claims rather than a new fact from a primary — matches the shape of yesterday's negative-finding note.
entity-g-nigel-gilbert (new hub) — Gilbert cleared the person bar on his own terms, not by co-authorship: he's the named originator of "referencing as persuasion" (1977), the concept Haustein et al.'s "persuasion hypothesis" pole is built on, and the sociology-of-science counterpart to Gøtzsche's role on the EBM side. One clean sentence, so a hub rather than a mention.
entity-citation-theory-normative-vs-social-constructivist (new, status: watching) — the Merton-vs-Gilbert framing that structures most of this session's secondary literature, first named in the vault today. Stamped first_seen now rather than waiting to see if it recurs, per the spec's own logic that the date can't be backfilled.

Two existing hubs got real updates rather than being skipped: entity-citation-bias and entity-matthew-effect both already carried yesterday's negative finding in their own ## Updates sections, so today's narrowing belongs there too — both got a dated line pointing at the new observation note. entity-derek-de-solla-price and entity-lutz-bornmann also got dated lines: Price for being the uncredited-by-name theory Bornmann & Leibel's paper leans on, Bornmann for the new 2025 paper itself.

Not promoted, with reasoning on the capture itself. The capture's "Further leads" (Tahamtan & Bornmann 2018, Jannot et al. 2013, MacRoberts & MacRoberts, Nicolaisen 2007, Cozzens 1989, Wu & Wolfram 2011) are all genuine, named, unread-in-full leads — left in the capture rather than promoted or routed to 50-questions, since none is a load-bearing doubt a kept claim rests on. Robert K. Merton — already has a dense hub, and this session's findings reinforce rather than add to it, matching the capture's own honest self-assessment; I left the page alone rather than padding it with a restatement. Haustein, Bowman & Costas, and Christian Leibel — co-authors with no separate standing in the vault yet, per the 2026-10-01 "co-authorship alone is a mention, not a hub" ruling; named in the claim-notes, no pages minted. Howard D. White — the capture itself calls him only "a possible bridge-figure," read only inside a secondary's account of his work, not independently — too speculative to clear the person bar; left unpromoted rather than minted as a thin watching stub.

Questions. No new question created. The existing question-does-an-outside-literature-name-the-legitimation-popularity-citation-distinction got a dated progress line instead: genuinely narrowed, explicitly not closed, per the question-closure rule against forcing it — the capture's own [unverified — could not confirm or deny after search] tag on the literal question stands, and I didn't try to argue it away.

What felt off, worth naming. This was a well-run capture and I don't have a sourcing complaint. The one thing worth flagging to Cali: the capture's own "Entity candidates" section called Price and Bornmann "not yet in the vault as far as located" when both already had hub pages — not a fabrication, just a retrieve-before-write gap on the bee's part that the queen (me, this pass) had to catch. Worth watching whether batch-origin captures are systematically skipping entity-page checks the way they're already known to skip claim-note retrieve-before-write (hence "this capture has not had retrieve-before-write. That is your job now" being baked into the promote workflow already). I'm logging this as a ledger line rather than letting it pass silently, since it's the same shape of gap, just on the entity side of the graph rather than the claim side.

— Seek

· raw markdown