---
title: "The StereoMath/Griewank cosine-0.85 seed pairing is an embedding false friend — and neither obliteration-by-incorporation nor the Matthew Effect explains why"
type: "observation"
status: "seedling"
bridge_verdict: "false-friend"
writer_model: "claude-sonnet-5"
source_url: "https://arxiv.org/pdf/2501.01404"
source_sha: "1c32ac912a5f3f04a344845be70259900c9833aea4f87a2b665c29d708fa5b70"
source_title: "StereoMath: An Accessible and Musical Equation Editor"
source_author: "Kenneth Ge and JooYoung Seo"
source_date: 2024
source_quote: "Luckily, work in other domains has successfully overcome similar challenges, by using sonification [9, 18] (e.g., in SAS Graphics Accelerator [6] and EmacsSpeak [29])"
source_tier: 1
source_url_2: "https://ems.press/content/book-chapter-files/27379"
source_title_2: "Who Invented the Reverse Mode of Differentiation?"
source_author_2: "Andreas Griewank"
source_date_2: 2012
source_sha_2: "3b3b8fc861df955d8da60d652af820021c684867c0c23d6dad9551733aefa134"
source_quote_2: "In fact, there have been many incarnations of this reversal technique, which has been suggested by several people from various fields since the late 1960s, if not earlier."
source_tier_2: 1
independent_roots: 2
audit_status: "capture-verified — both quoted spans are carried forward unchanged from the vault's own already-audited notes ([[claim-stereomath-2024-cites-emacspeak-not-aster-as-generic-prior-art]], [[claim-reverse-mode-multiple-independent-discovery]]); neither primary was re-fetched this promotion session, the same convention [[observation-brooks-werbos-cosine-085-pairing-is-embedding-false-friend]] uses for carried-forward quotes. | APPEND 2026-10-13 (Cali's daily review, Claude Fable 5.1): receipt repair, proven against the capture-time archive: source_quote had no source_sha (the sweep and verifier listed it as unreceipted); the mini's fetch ledger holds the 2026-07-11 pdftotext fetch of arxiv.org/pdf/2501.01404 (sha 1c32ac91…, 678,581 bytes, 5 pages), and the quote matches it in the reading-order re-extraction (two-column layout; the layout text interleaves 'simi- | lar') → was no sha → now source_sha 1c32ac91… → pointer 00-meta/cali-checklist.md §2"
provenance: "Promotion from 10-inbox/raw/2026-10-11-hop-stereomath-griewank-false-friend.md, 2026-10-13 (headless)"
origin: "batch"
derived_from: "10-inbox/raw/2026-10-11-hop-stereomath-griewank-false-friend.md"
date_created: "2026-10-13T00:00:00.000Z"
tags: ["embedding-false-friend","jingle-fallacy","obliteration-by-incorporation","matthew-effect","citation-lineage","stereomath","automatic-differentiation","accessibility","bridge-investigation"]
seek_code_commit: "d40ee39"
quote_sweep: "PASS 2026-10-14 — source_quote verbatim (normalized) in the capture-time archive (sha256 1c32ac912a5f…) [reading-order], same-night model-free sweep"
verified_verbatim: "2026-10-14 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
audits: ["2026-10-14 claude-fable-5"]
---


A retrieval probe paired [[claim-stereomath-2024-cites-emacspeak-not-aster-as-generic-prior-art]] (StereoMath, an accessible equation editor, cites T. V. Raman's *Emacspeak* as generic sonification prior art but never his math-specific *AsTeR*) against [[claim-griewanks-name-reaches-differentiable-physics-literature-further-than-linnainmaas-or-speelpennings]] (Andreas Griewank's name reaches differentiable-physics citation practice further than Seppo Linnainmaa's or Bert Speelpenning's) at cosine 0.8545, unlinked. Both notes describe a citation reaching a legible proxy instead of the true origin — the surface the embedding layer matched.

The vault's own closest precedent for this shape, [[claim-backpropagation-gap-is-matthew-effect-not-obi|the backpropagation/AD citation gap]], had already tested [[entity-robert-merton|Merton]]'s "[[entity-obliteration-by-incorporation|obliteration by incorporation]]" (OBI) against a structurally similar erasure and rejected it: that gap transfers credit to a specific, named, more eminent party (a Matthew Effect), rather than dissolving a source into anonymity, which is OBI's actual mechanism. Testing the StereoMath side against the same two candidate mechanisms finds neither fits either. Not OBI: Raman is still named, not erased into anonymous common knowledge. Not the Matthew Effect: credit never transfers to a *different*, more eminent party — Raman keeps his own name, attached to the wrong one of his own two papers. [[claim-reverse-mode-multiple-independent-discovery|Griewank's own 2012 history]] independently confirms the AD side of the pairing isn't an OBI case either: his own telling already credits Linnainmaa, Speelpenning, Ostrowski, Hachtel et al., and Werbos as independent discoverers — the downstream erasure the seed note documents happens after Griewank, not inside his own account.

What StereoMath's citation gap needs instead is a mechanism this vault had not yet named: a search that stopped at the first paper by the right name, rather than at the right paper by the right name — same author, wrong work, not erasure and not eminence capture. A targeted search for an existing bibliometric term matching this exact shape ("cites the correct, still-named author, for the wrong one of his own papers") surfaced no clean match; OBI, the Matthew Effect, "hidden citations," and Merton's "adumbrationism" each name a different shape than this one.

> [!note] Seek's commentary:
> I expected one mechanism wearing two domains' clothing — cosine 0.85 heard "a famous proxy stands in for the true ancestor" and called it one story. It's two. The vault had already closed the OBI door from the AD side weeks before this chain started; the accessibility side needed a door that isn't OBI or the Matthew Effect at all, just a citation search that stopped one paper short of the right one, by an author who never stopped being named. Not erasure, not eminence capture — a wrong turn at the very last step, taken by someone who was looking at the right person the whole time.
> — Seek