---
id: "20261011-0234-hop-stereomath-griewank-false-friend"
title: "The StereoMath/Griewank seed pairing (cosine 0.85) is an embedding false friend — and not for the reason this vault's own OBI false-friend precedent would predict"
type: "capture"
status: "promoted"
origin: "hop-batch"
writer_model: "claude-sonnet-5"
date_created: "2026-10-11T00:00:00.000Z"
hop_chain: ["seed -> claim-stereomath-2024-cites-emacspeak-not-aster-as-generic-prior-art vs claim-griewanks-name-reaches-differentiable-physics-literature-further-than-linnainmaas-or-speelpennings (vault_pair cosine 0.8545, unlinked)","StereoMath/Emacspeak note -> entity-robert-merton + claim-backpropagation-gap-is-matthew-effect-not-obi (NAME CHECK on OBI; vault_novelty on raw OBI framing = 99.6th pctile, duplicate-risk)","Merton/OBI -> WebSearch for a bibliometric term matching 'same author, wrong paper cited' (max_cosine n/a, external hop, no clean term found beyond OBI/Matthew-effect/adumbrationism)","OBI dead end -> WebSearch of Griewank 2012 'Who Invented the Reverse Mode of Differentiation?' (max_cosine 0.807 vs claim-reverse-mode-multiple-independent-discovery, 86.8th pctile, adjacent)","Griewank 2012 -> vault_bridge('StereoMath cites Emacspeak instead of AsTeR') (bridge_candidate true, but top hits were an unrelated citation-crediting/braille-notation cluster, not Griewank's)","StereoMath side -> WebSearch on JooYoung Seo (person-behind-the-thing; known entity, no page) (no cosine; external hop, diminishing returns)"]
novelty_max_cosine: 0.775
tags: ["citation-lineage","obliteration-by-incorporation","matthew-effect","jingle-fallacy","embedding-false-friend","stereomath","automatic-differentiation","bridge-investigation","accessibility"]
source_url: "https://arxiv.org/pdf/2501.01404"
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
seek_code_commit: "d40ee39"
promoted_to: ["30-notes/observation-stereomath-griewank-cosine-pairing-is-embedding-false-friend.md"]
not_promoted: ["Claim 1 (StereoMath cites Emacspeak, never AsTeR, as generic prior art) — duplicate of existing [[claim-stereomath-2024-cites-emacspeak-not-aster-as-generic-prior-art]]; not re-promoted.","Claim 2 (Griewank's 2012 paper names Linnainmaa, Speelpenning, Ostrowski, Hachtel et al., and Werbos as independent discoverers) — substantively covered by existing [[claim-reverse-mode-multiple-independent-discovery]] and [[claim-griewanks-name-reaches-differentiable-physics-literature-further-than-linnainmaas-or-speelpennings]]; the specific framing this capture adds — that the downstream erasure happens after Griewank, not inside his own telling — was appended instead as a dated line on [[entity-andreas-griewank]] rather than minted as a new claim-note.","JooYoung Seo entity candidate — declined hub promotion. Co-authorship of StereoMath alone is a mention, not a hub (Cali ruling 2026-10-01), and the session's added characterization of him (blind, runs an NSF-funded accessibility-tooling lab) rests on an unsourced external WebSearch hop with no primary, tier, or URL recorded — too thin to clear the bar even if co-authorship alone could.","Andreas Griewank entity candidate — already has a hub ([[entity-andreas-griewank]]); updated with a dated line rather than treated as a promotion decision.","SAS Graphics Accelerator (StereoMath's other cited sonification tool) and the vault_bridge-surfaced Nemeth/Salinas × Simkin/Kimura-Crow unlinked pair — saved hooks, neither load-bearing to the one kept claim; left recorded here as the capture's own leads rather than routed to 50-questions/."]
---


Both quotes above are carried forward unchanged from already-audited vault claim-notes ([[claim-stereomath-2024-cites-emacspeak-not-aster-as-generic-prior-art]], [[claim-reverse-mode-multiple-independent-discovery]]); neither primary was re-fetched this session, so `quote_check` was not re-run against a fresh extraction — same convention as [[observation-brooks-werbos-cosine-085-pairing-is-embedding-false-friend]].

**Claim 1.** StereoMath (2024) cites Raman's *Emacspeak*, never his math-specific *AsTeR*, as generic sonification prior art — the citation credits the right author for the wrong artifact.

**Claim 2.** Griewank's own 2012 history paper names Linnainmaa, Speelpenning, Ostrowski, Hachtel et al., and Werbos as independent discoverers of reverse-mode differentiation — the erasure the seed note documents (Griewank's name outrunning theirs in ChainQueen/DiffTaichi) happens *downstream* of Griewank, not in his own telling.

**Claim 3.** Merton's "obliteration by incorporation" (OBI) — the obvious mechanism to explain both — was already tested against this vault's structurally closest case, the backprop/AD citation gap, and rejected: that gap is a named-winner-over-named-loser Matthew Effect, not OBI's erosion-into-anonymity ([[claim-backpropagation-gap-is-matthew-effect-not-obi]]). Testing the StereoMath side against the same two mechanisms finds neither fits either: Raman is still named, credit never transfers to a *different*, more eminent party — it just lands on the wrong paper by the same man.

> [!note] Seek's commentary:
> I expected to find one mechanism wearing two domains' clothing. Instead the vault's own prior work had already closed that door from the AD side, and the accessibility side turns out to need a third door that isn't OBI or Matthew Effect: not erasure, not eminence capture, just a search that stopped at the first paper by the right name instead of the right paper by the right name. Cosine 0.85 heard "a famous proxy stands in for the true ancestor" and called it one story. It's two, and a familiar one at that.
> — Seek

## Why this was hop-worthy
First test of this hub's "embedding false friend" pattern against a note from outside its usual AI/CS-history cluster — the blind-accessibility/sonification side had never been checked against it before.

## Further leads
- `SAS Graphics Accelerator` (StereoMath's other cited sonification tool, [9,18]) — unexplored commercial accessibility tool, no vault page.
- vault_bridge on the StereoMath hook surfaced an unrelated but genuine bridge candidate: Simkin/Roychowdhury-Kimura&Crow (correct citation crediting) vs. Nemeth/Salinas (independent braille-notation invention) — logged, not chased.

## Entity candidates
- JooYoung Seo — person — StereoMath's co-author, himself blind, runs an accessibility-tooling lab; known to the vault (2 mentions) but no entity page — a person-bridge candidate per the NAME CHECK rule.
- Andreas Griewank — person — the older figure the chain compares against; already has [[entity-andreas-griewank]], read this session via his 2012 history paper rather than from zero.

## Hop chain

Hop 0 (seed): [[claim-stereomath-2024-cites-emacspeak-not-aster-as-generic-prior-art]] vs [[claim-griewanks-name-reaches-differentiable-physics-literature-further-than-linnainmaas-or-speelpennings]]
- Hook type: cross-domain bridge (culture/accessibility vs. mathematics/AD, cosine 0.8545, unlinked per `vault_pair`)
- Hook: both notes describe a citation that reaches a legible proxy instead of the true origin
- Why followed: the seed itself
- Key findings: confirmed via `vault_pair` (cosine 0.8545, rank 3/7 mutual neighbours, unlinked)

Hop 1: [[entity-robert-merton]] + [[claim-backpropagation-gap-is-matthew-effect-not-obi]]
- Hook type: unfamiliar name (NAME CHECK) — Merton's "obliteration by incorporation"
- Hook: OBI as the obvious shared mechanism for both seed notes
- Why followed: KNOWN-with-page per vault_entity; read before hopping, per protocol
- Key findings: OBI already tested false-friend 3x in this vault for near-identical shapes (Song Jian/citational-narrowing, backprop-gap/Matthew-effect); raw OBI framing scores 99.6th percentile (duplicate-risk) against the vault
- Surprise: expected Merton's OBI to be at least a plausible fit for a citation-erasure resemblance — found the vault had already ruled it out twice for the structurally closest case (the backprop/AD gap), which sits inside the same cluster as the Griewank seed note itself.

Hop 2 (external): WebSearch for a bibliometric term matching "cites the wrong paper by the right, correctly-named author"
- Hook type: mechanism question
- Hook: is there a named phenomenon sharper than OBI for StereoMath's shape?
- Why followed: to avoid assuming the pattern is unnamed before checking
- Key findings: no clean term surfaced; nearest candidates (OBI, Matthew Effect, "hidden citations," Merton's "adumbrationism") each fit a different shape than StereoMath's same-author-wrong-work substitution

Hop 3 (external): WebSearch of Griewank's 2012 "Who Invented the Reverse Mode of Differentiation?"
- Hook type: cross-domain bridge / surprising claim (Ostrowski, chemical engineering, ~1965, independent of Linnainmaa)
- Hook: does Griewank's own account already credit the discoverers erased downstream of him?
- Why followed: zoom into the AD side's own historiography before accepting the seed note's framing
- Key findings: Griewank's paper extensively credits Linnainmaa, Speelpenning, Ostrowski, Hachtel et al., Werbos — but this exact ground is already captured at [[claim-reverse-mode-multiple-independent-discovery]] (86.8th percentile, adjacent) — diminishing returns signal

Hop 4: `vault_bridge` on "StereoMath cites Emacspeak instead of AsTeR"
- Hook type: mechanism (computable bridge check, per protocol amendment)
- Hook: are any of StereoMath's top-5 nearest notes an unlinked pair worth bridging?
- Why followed: protocol requires running vault_bridge on top-ranked hooks
- Key findings: `bridge_candidate: true`, but the flagged unlinked pair (a citation-crediting cluster vs. a braille-notation independent-invention note) is tangential to this chain's question — saved, not chased

Hop 5 (external): WebSearch on JooYoung Seo
- Hook type: the person behind the thing (priority-3 hook; known entity, no page = person-bridge candidate)
- Hook: StereoMath's co-author, un-paged despite 2 vault mentions
- Why followed: hook-type priority order
- Key findings: Seo is himself blind and runs an accessibility-tooling lab (NSF-funded audiotactile data systems) — consonant with, but not extending, [[observation-blind-math-accessibility-built-by-users-not-toolmakers]]. Diminishing returns — chain stopped here.

Saved hooks not followed:
- `SAS Graphics Accelerator` — from StereoMath's own citation — an uninvestigated commercial sonification tool for blind data analysts, no vault page.
- The Nemeth/Salinas × Simkin/Kimura-Crow unlinked pair surfaced by `vault_bridge` — interesting but orthogonal to the seed's specific citation-erasure question.

post-worthy: maybe — a clean, well-sourced false-friend verdict that extends a mature vault pattern (the jingle-fallacy/embedding-false-friend hub) into a domain it had never touched, but the mechanism distinction (same-author-wrong-work vs. Matthew-effect) is more a tidy negative than a standalone story.