---
id: "20260925-0205-bipartite-two-things-this"
title: "Linnainmaa citation histogram vs OSTI SAND-91-7095 metadata (cosine 0.84): a false friend with one narrow shared property — both are readings of an aggregator's record layer, and each aggregator's encoding limits the claim"
type: "capture"
status: "promoted"
promoted_on: "2026-10-03T00:00:00.000Z"
promoted_by: "Cali's daily review 2026-10-03 (Claude claude-opus-5-5; Cali: \"skip the recovered captures whose pair is already settled\", 00-meta/cali-checklist.md §0.10)"
promoted_to: []
not_promoted: ["All claims superseded, not promoted: the pair (claim-linnainmaa-1976-citation-histogram × claim-osti-structured-metadata-confirms-sand91-7095-january-1992) is already settled as a false friend in [[observation-osti-semantic-scholar-cosine-pairing-is-verification-genre-false-friend]] (promoted 2026-09-27, same verdict). This capture's one specific detail (date/timezone granularity of the OSTI field) is already in [[claim-osti-structured-metadata-confirms-sand91-7095-january-1992]]. The sibling 2026-09-27-recovered-capture-does-a-genuine-mechanism-connect-linnainmaas-fb616a.md stays queued: its differentiable-MPM link is not in the vault. Capture kept as written; nothing deleted."]
origin: "batch"
writer_model: "claude-sonnet-5"
date_created: "2026-09-25T00:00:00.000Z"
provenance: "batch run 2026-09-25, bipartite hop between claim-linnainmaa-1976-citation-histogram (ml-ai) and claim-osti-structured-metadata-confirms-sand91-7095-january-1992 (computing)"
derived_from: []
tags: ["bipartite","false-friend","bibliometrics","metadata","semantic-scholar","osti","embeddings"]
source_tier: 1
recovered: true
recovered_on: "2026-10-01T00:00:00.000Z"
recovered_from: "Seek session transcript 60d921fc-3a25-4b68-b883-daf8f0ac6f69.jsonl on the mini - the session's Write at 2026-09-25 09:06 UTC, replayed verbatim"
recovery_note: "Overwritten the same night: another bipartite session was told to write the same filename (2026-09-25-bipartite-two-things-this-vault-knows-in-different.md). Collision fixed in seek-code c516ae8. Restored unpromoted at Cali's instruction, 2026-10-01. Before promoting, read the same-pair captures listed below: several sessions researched this pair, and only one per night survived."
same_pair_captures: ["2026-09-26-recovered-bipartite-hop-linnainmaa-1976-citation-histogram-vs-57a59c.md","2026-09-27-recovered-capture-does-a-genuine-mechanism-connect-linnainmaas-fb616a.md"]
seek_code_commit: "f8a723c"
---


Bridge-seed check: no sibling capture in `10-inbox/raw/` mentions both slugs (the histogram note appears in three captures, the OSTI note in one; none together). Investigated fresh.

Notes compared: [[claim-linnainmaa-1976-citation-histogram]] and [[claim-osti-structured-metadata-confirms-sand91-7095-january-1992]]. Context: [[claim-sand91-7095-predates-1993-mpm-founding-report]], [[claim-ml-and-ad-communities-mutually-unaware]], and the vault's prior false-friend rulings [[observation-gates-jevons-cosine-pairing-is-embedding-false-friend]], [[observation-greenhalgh-2003-klensin-romberg-1989-cosine-pairing-is-embedding-false-friend]], [[observation-ml-ad-en-gedi-cosine-pairing-is-embedding-false-friend]].

**Verdict: false friend, with a narrow real kernel (below).** No shared subject, people, or field: one is a count of citing papers over time; the other is the date field of a single catalog record. The historical thread that could have linked them by content — [[entity-howard-schreyer]] / [[entity-deborah-sulsky]] / [[entity-zhen-chen]] (material point method) versus Linnainmaa (automatic differentiation) — is a different question already covered by [[claim-ml-and-ad-communities-mutually-unaware]].

Discriminating test (per the Gates/Jevons precedent): does the claim rest on the same evidentiary act? It does not. The histogram is an *aggregation over many records* (a count), whose weakness is corpus coverage. The OSTI claim is a *reading of one record's field*, whose weakness is field precision and rendering.

## Claim: Semantic Scholar's Linnainmaa count is a snapshot of a curated, moving index (quantitative)
claim_kind: quantitative
verifies: (none — no named question)
- source_url: https://api.semanticscholar.org/graph/v1/paper/DOI:10.1007/BF01931367?fields=title,year,citationCount,venue,authors
- source_sha: b00e4b8bfe0ffefa8b0afc9bdf620b41e6c39526aff2b6e7ffd36290212faad8
- source_author: Semantic Scholar Graph API (Allen Institute for AI); record for S. Linnainmaa
- source_date: queried 2026-09-25
- source_title: "Taylor expansion of the accumulated rounding error" (record title, verbatim from API)
- source_venue: Semantic Scholar Academic Graph API (Allen Institute for AI)
- source_quote: `"title": "Taylor expansion of the accumulated rounding error", ... "year": 1976, "citationCount": 314`
- source_tier: 1 (primary: the index's own record; says nothing about total true citations)

The live API returns `citationCount: 314` on 2026-09-25; the vault's histogram note recorded 305 records on 2026-07-07. The count rose by 9 in ~11 weeks, so the "245 since 2010" and 60%-after-2020 shares are dated snapshot figures, not fixed properties of the paper. (Arithmetic on two dated observations; the 305 figure comes from the vault note, not re-pulled here.) Also, the per-decade histogram requires the `/citations` endpoint, which was not re-run this session — I did not re-verify the 51/245 split.

## Claim: OSTI's "Dec 31 1991" timestamp is consistent with UTC-midnight first-of-month dates rendered in US Eastern time (mechanism)
claim_kind: technical-mechanism (inferred from the record's own rendering; OSTI documentation of its date storage NOT located)
- source_url: https://www.osti.gov/biblio/5900384
- source_sha: f96ee695b0188840ef1d522ddaea2e68c462a6f26b578a6a0e0ae563953a160e
- source_author: Sulsky, D; Chen, Zhen; Schreyer, H L (record authors); catalog by OSTI, U.S. DOE
- source_date: 1992 (catalog year); page read 2026-09-25
- source_title: "The application of a material-spatial numerical method to penetration"
- source_venue: OSTI.GOV, U.S. Department of Energy Office of Scientific and Technical Information; Sandia National Laboratories report SAND-91-7095
- source_quote: "Technical Report·Wed Jan 01 00:00:00 EST 1992" ; "@techreport{osti_5900384, ... year = {1992}, month = {01}}" ; similar-records: "Tue Dec 31 19:00:00 EST 1991", "Fri Apr 30 20:00:00 EDT 1993", "Mon May 31 20:00:00 EDT 1993"
- source_tier: 1 for what the record displays; the storage mechanism is [unverified-mechanism -- needs primary]

Every date in the Similar Records sidebar falls on the last day of a month at 19:00 EST / 20:00 EDT — exactly 00:00 UTC on the first of the next month (Dec 31 19:00 EST = Jan 1 00:00 UTC; Apr 30 20:00 EDT = May 1 00:00 UTC; May 31 20:00 EDT = Jun 1 00:00 UTC). That pattern across three unrelated records supports the vault's earlier "display artifact" reading and indicates month-level (not day-level) resolution. This strengthens, without contradicting, the existing OSTI note; note that the note's "five-hour offset" is precisely the EST–UTC offset, and the 1993 EDT items show a four-hour offset, which is the same artifact under daylight time. Mechanism remains inference from displayed strings, hence the flag.

## Claim: The one shared property is that each is a reading of an aggregator's encoding, so each conclusion inherits the aggregator's limits (synthesis)
claim_kind: synthesis, no new fact
- Histogram inherits index *coverage* limits (Semantic Scholar is a curated corpus; the histogram note already flags a Google Scholar 1247 figure as `[unverified-quant]`).
- OSTI inherits *precision/rendering* limits (month-level dates, UTC-to-local display).
- This is a generic property of any database-derived number or date and does not discriminate these two notes from thousands of others; it does not warrant a wikilink between them. The 0.84 cosine most plausibly reflects shared register ("independently confirms", "primary record", "structured", machine-readable fields, counts and dates), the same failure mode as the earlier false-friend observations.

> [!note] Seek's commentary:
> The useful residue is a checklist question for any metadata-based note: what does the encoding *not* resolve? For the histogram, coverage; for OSTI, day-level precision and timezone. Not a bridge, and not the same claim.

## Further leads
- OSTI documentation on how dates are stored/exported (would move the timezone mechanism above from inference to Tier 1).
- Re-pull `/citations?fields=year` for the Linnainmaa DOI to see whether the decade histogram shifted with the 305→314 change (Semantic Scholar API docs).
- Whether the 9 new citations since July are 2026 papers (would test the "resurrection curve" reading).

## Entity candidates
- Seppo Linnainmaa — person — the foundational figure whose 1976 paper the histogram measures; priority claim rests on him.
- Howard Schreyer — person — SAND-91-7095 co-author; the OSTI note's subject.
- Deborah Sulsky — person — SAND-91-7095 lead author, material point method.
- Zhen Chen — person — SAND-91-7095 co-author.
- Semantic Scholar Academic Graph API — term — curated citation index underlying the histogram; count drifted 305→314.
- OSTI.GOV — term — DOE catalog whose date rendering shows the UTC-midnight artifact.
- UTC-midnight rendering artifact — concept — explains "Dec 31 19:00 EST" entries for month-level dates.
- Embedding false friend — concept — fifth-ish instance; resemblance of register, not mechanism.