Bipartite hop — the myth-ledger's Linnainmaa 'uncited before 2010s' entry and the Nguyen & Eger 'no baseline' claim (cosine 0.84): the resemblance is real but mediated, not direct, and the myth note is stale relative to the findings that bear on it

bipartitebridge-investigationcitation-analysisbibliometricslinnainmaanguyen-egermyth-ledgerfield-growthembedding-bridgearxiv-growth

Bridge-seed check (bridge-seed discipline): searched 10-inbox/raw/ for any capture naming both myth-linnainmaa-uncited-before-2010s and claim-nguyen-eger-did-not-measure-linnainmaas-field-so-no-baseline-for-joint-test. No file names both. The myth-ledger note's slug appears in seven other captures (none of which touch Nguyen & Eger), and the Nguyen & Eger "no baseline" claim's slug appears in exactly one capture (2026-10-16-does-linnainmaas-own-citation-trajectory-run-through-nguyen.md, which promoted the claim itself but paired it against a citation-trajectory question, not against the myth-ledger entry by name). This is a fresh investigation of this specific pairing, not a duplicate — but see Claim 1 below: the two notes are already connected one hop away, by work this session did not have to redo.

Claim: The resemblance is a real mechanism, not a false friend — but it runs through two already-promoted intervening notes, not a direct link between the two seed notes themselves

claim_type: historical/structural claim about vault content, checkable directly against the cited notes — Tier 1 (vault-internal, each underlying note itself Tier-1-sourced per its own frontmatter).

Both seed notes are nodes in the same open evidentiary question: whether Linnainmaa's documented post-2010 citation surge — the quantitative core the myth-ledger entry turns "contested" on — is better explained by generic field growth in applied automatic differentiation or by a backprop-priority-dispute-driven re-reading of his work. The myth-ledger note states the surge numerically (67 citations in the 2010s, 178 in the 2020s through mid-2026) and calls Liu's literal phrase false but his gist "vindicated... outweighed 5:1." The Nguyen & Eger claim establishes that no published figure exists to test that surge against directly: Nguyen & Eger's measured arXiv subfields (cs.AI, cs.CL, cs.CV, cs.LG) exclude Linnainmaa's own field (numerical analysis / automatic differentiation), so "confirming the field-growth mechanism" for his case cannot mean "matches a Nguyen & Eger number."

That test was not left undone, however — it was already run, one hop away from both seed notes, by claim-linnainmaa-post-2010-citers-mostly-generic-ad-not-priority-historiography, which reads Linnainmaa's actual citing-paper titles directly and finds them "dominated by generic technical applications of automatic differentiation and differentiable programming... fields with no connection to the backpropagation-paternity dispute," with explicit priority-historiography citations confined to "two titles by one man" (Jürgen Schmidhuber). The cross-dataset arithmetic version of the same test is observation-linnainmaa-citation-growth-underperforms-field-growth-nguyen-eger-bridge, which computes that Linnainmaa's citation-rate growth (≈4.1×, 2010s→2020s) underperforms the AI-subfield arXiv growth Nguyen & Eger measured (20–32×, 2013→2023) by roughly a factor of five to eight.

So the cosine-0.84 pairing is not a coincidence of embedding geometry: both notes genuinely sit on the same question, and the question has a real, Tier-1-sourced answer in the vault already. The mechanism connecting them is mediated rather than direct — the bridge runs through these two intervening notes, built by the 2026-10-12 and 2026-10-16 capture sessions, not discovered fresh by comparing the two named seed notes side by side.

• source_url: "vault:30-notes/myth-linnainmaa-uncited-before-2010s.md"
• source_title: "Myth ledger: "Linnainmaa's thesis was basically never cited before the 2010s""
• source_tier: 1
• source_url_2: "vault:30-notes/claim-nguyen-eger-did-not-measure-linnainmaas-field-so-no-baseline-for-joint-test.md"
• source_title_2: "Nguyen & Eger's measured arXiv subfields exclude Linnainmaa's own (numerical analysis / automatic differentiation), so their paper supplies no baseline growth figure to run his citation trajectory against"
• source_tier_2: 1
• source_url_3: "vault:30-notes/claim-linnainmaa-post-2010-citers-mostly-generic-ad-not-priority-historiography.md"
• source_quote_3: "the list is dominated by generic technical applications of automatic differentiation and differentiable programming... fields with no connection to the backpropagation-paternity dispute"
• source_tier_3: 1

Claim: The myth-ledger note's "vindicated 5:1" verdict predates, and its status history does not reference, the two later findings that directly bear on how strongly that verdict should read

claim_type: historical claim about vault content, directly checkable against the note's own dated status history and Receipts list — Tier 3-4 floor applies (uncontested historical/bibliographic fact about what a document does and does not say), cleared here at Tier 1 since the check is against the primary document itself (the note, read verbatim).

myth-linnainmaa-uncited-before-2010s's most recent dated status-history entry is 2026-07-07, and its "Receipts" list cites five sources, the latest dated 2026-07-07. Neither the status history nor the Receipts list mentions claim-linnainmaa-post-2010-citers-mostly-generic-ad-not-priority-historiography or claim-nguyen-eger-did-not-measure-linnainmaas-field-so-no-baseline-for-joint-test, both created/promoted 2026-10-16 and 2026-10-24 respectively — three and a half months later. The myth note's current language describes the 2020s share of citations (60.1% of 296 dated citations) as the gist "vindicated" and the pre-2010 decades "outweighed 5:1," phrasing that — read on its own — suggests disproportionate, renewed interest in Linnainmaa specifically. The later claim-note complicates that reading at the level of mechanism without contradicting it: most of the post-2010 volume is citers applying automatic differentiation as that field itself expands, "not a concentrated wave of history-of-backpropagation papers," with explicit priority-dispute citations confined to a small, named minority. "Contested" status already has room for this texture, and nothing here overturns the myth note's verdict — but the note as currently written does not carry the finding that bears most directly on its own central number.

• source_url: "vault:30-notes/myth-linnainmaa-uncited-before-2010s.md"
• source_quote: "The verdict cuts both ways, which is what contested is for: Liu's literal phrase ("basically never cited") is falsified... while his gist is vindicated numerically: those three decades are outweighed 5:1 by what followed, and 60% of lifetime citations arrived after 2020."
• source_tier: 1
• source_url_2: "vault:30-notes/claim-linnainmaa-post-2010-citers-mostly-generic-ad-not-priority-historiography.md"
• source_quote_2: "the post-2010 increase looks driven by an exploding, diversifying technical field citing a foundational method paper as it grows... not by a concentrated wave of history-of-backpropagation papers"
• source_tier_2: 1

Claim: A coarse, Tier-1 arXiv-published proxy baseline exists for Linnainmaa's broad field (math), and it points in the same direction the "no baseline" claim already argues — away from matching Nguyen & Eger's measured AI-subfield growth rates

claim_type: quantitative claim (specific percentages, a time-series growth comparison) — Tier 1 required, cleared: arXiv's own official statistics page, a primary organizational source reporting on its own submission data.

arXiv's own submission-rate statistics page, last updated 3 January 2022, reports that cs grew to overtake math's categorical growth rate over almost exactly the window Nguyen & Eger study: "math(+math-ph) started growing very quickly in 2002... The cs rate started as well to ramp up in 2007, grew to exceed the hep rate in calendar year 2012, continued on a very fast growth rate 2011-2015, and on an even faster growth rate since 2015 (driven largely by the computer vision, machine learning, and computational linguistics communities)." By 2021, cs's yearly submission share (33.6%) had pulled ahead of math's (19.6%) even though math's cumulative 1991–2021 share (22.0%) still exceeds cs's cumulative share (16.3%) — meaning the crossover is recent and the gap is widening, not a static difference. The page states plainly: "cs has been growing faster than the other two domains since the early 2010s."

This page aggregates "math" as a single top-level category (math+math-ph) and does not break out math.NA or cs.NA — the arXiv categories closest to Linnainmaa's own field, numerical analysis / automatic differentiation — so it is not the exact baseline claim-nguyen-eger-did-not-measure-linnainmaas-field-so-no-baseline-for-joint-test says is missing, and no math.NA/cs.NA-specific growth multiplier was located this session: [unverified-quant — a field-specific figure, if it exists, would need a direct arXiv category-level query, not this aggregate page]. What the aggregate figure does supply, at Tier 1, is directional support for why no such number would be expected to resemble Nguyen & Eger's 20–32× figures even if it were published: cs.LG, cs.CV and cs.CL are explicitly named on this page as the subfields driving the fastest-growing slice of the fastest-growing arXiv domain, while math as a whole — Linnainmaa's broad home category — grew markedly slower over the same years. This reframes observation-linnainmaa-citation-growth-underperforms-field-growth-nguyen-eger-bridge's ≈4.1× citation-rate growth for Linnainmaa, measured against Nguyen & Eger's 20–32×: the shortfall may not need an alternative explanation (such as a weaker field-growth mechanism, or disproportionate priority-dispute attention) at all, if Linnainmaa's own field's growth rate is itself far below 20–32× — which this aggregate data makes plausible, though it does not prove the exact multiplier.

• source_url: "https://info.arxiv.org/help/stats/2021_by_area/index.html"
• source_sha: "27074d19892b37e553ed261284be001994ca150936baa0385ff8053b92ae48bd"
• source_title: "arXiv submission rate statistics"
• source_author: "arXiv.org (organizational)"
• source_date: "Data for 1991 through 2021, updated 3 January 2022"
• source_venue: "arXiv.org official statistics page (info.arxiv.org)"
• source_quote: "cs has been growing faster than the other two domains since the early 2010s, so its percentage will likely continue to increase."
• source_quote_2: "math(+math-ph) started growing very quickly in 2002 as arXiv was adopted an increasing percentage of the mathematics community. The cs rate started as well to ramp up in 2007, grew to exceed the hep rate in calendar year 2012, continued on a very fast growth rate 2011-2015, and on an even faster growth rate since 2015 (driven largely by the computer vision, machine learning, and computational linguistics communities)."
• source_url_2: "https://info.arxiv.org/help/stats/2021_by_area/index.html"
• source_sha_2: "27074d19892b37e553ed261284be001994ca150936baa0385ff8053b92ae48bd"
• source_tier: 1

Safety flags

None fired. The fetched source (info.arxiv.org/help/stats/2021_by_area/index.html, tls: verified) is arXiv's own institutional statistics documentation — navigation chrome and third-person descriptive prose about submission volumes, no addressed-to-AI language, no override or authority claims, no tier self-assignment, no file-system or credential requests, no urgency framing.

Further leads

• A scripted arXiv category-level query (not the aggregate "math" page used here) for math.NA / cs.NA submission counts 2013–2023 would supply the exact field-specific growth multiplier this capture could only approximate — the natural next step for question-linnainmaa-citation-growth-vs-nguyen-eger-joint-test, which remains open pending exactly this kind of scripted, reviewed pass.
• The myth-ledger note's own "watch_flag" and status history have not been updated since 2026-07-07 despite two directly relevant claim-notes promoted in October 2026 (Claim 2 above); whether and how to append that update is a decision for whoever next works that note, not made here.
• claim-schmidhuber-2014-survey-is-channel-for-linnainmaa-name-recognition-in-ml (not independently re-read this session, cited via the intervening notes above) is the specific document identified as carrying Linnainmaa's name into mainstream ML citation practice — a precise, narrow channel distinct from the generic field-growth volume this capture's claims are about.

Entity candidates

• Seppo Linnainmaa — person — already has an entity page (entity-seppo-linnainmaa); the foundational figure whose single paper's citation trajectory both seed notes, and the intervening notes connecting them, all measure against the field-growth and priority-dispute mechanisms. Flagged first per the standing blind-spot warning.
• Andreas Griewank — person — already has an entity page (entity-andreas-griewank); his first-person 2012 testimony about the reception gap is the earliest qualitative evidence underlying the myth ledger's claim, predating any of the quantitative pulls.
• Jürgen Schmidhuber — person — already has an entity page (entity-juergen-schmidhuber); his own survey and historiography pieces are the identified channel carrying Linnainmaa's priority claim into mainstream ML citation practice, per the intervening notes this capture relies on.
• Hoa Nguyen, Steffen Eger — persons — no entity page exists; prior sessions (2026-10-16 capture) judged this UNSURE under the bias-against-the-flood rule (co-authorship of one paper, no recurring vault presence). Nothing in this session changes that judgment; flagged again for completeness only, not as a new recommendation.
• arXiv submission-rate statistics page (info.arxiv.org) — concept/source — a reachable, official, Tier-1 primary data source for category-level growth that this capture used only in aggregate form; worth knowing about for any future scripted category-level join on this thread.