What genuinely connects the Kelly-Luccioni and Puckette-Petiška embedding-false-friend notes?

embedding-false-friendjingle-fallacybridge-investigationmeta-vaultcosine-similaritykelly-2025luccioni-2025miller-puckettematthew-effect

Bridge-seed discipline check

10-inbox/raw/ was searched for any prior capture mentioning both note slugs (kelly-luccioni-cosine-pairing-is-embedding-false-friend and puckette-dropped-mention-petiska-matthew-effect-pairing-is-embedding-false-friend, and looser kelly+puckette combinations). No sibling capture exists — this is the first investigation of this specific pairing. Two related-but-distinct captures were found and read for context but do not address this pair: 2026-08-20-what-genuinely-connects-kelly-et-als-2025-irwin.md (a different Kelly investigation, pre-dating the false-friend verdict) and 2026-08-31-what-genuinely-connects-the-2026-08-16-capture.md (the capture that produced the Puckette-Petiška note itself, a one-sided precursor to this topic).

Claim: the pairing is a false friend — same verdict-label, disjoint recursion level, non-converging grounding

The two named notes share a title template ("[X] cosine pairing is an embedding false friend") and a closing rhetorical move ("No wikilink was added"), but direct comparison finds three independent reasons the resemblance does not track a real connection.

Recursion-level mismatch. observation-kelly-luccioni-cosine-pairing-is-embedding-false-friend is a first-order false-friend verdict: its two inputs are raw claim-notes about the world (Kelly et al.'s citation practice; Luccioni et al.'s GPU-shipment figure). observation-puckette-dropped-mention-petiska-matthew-effect-pairing-is-embedding-false-friend is the vault's own sixth meta-level bipartite hop in an already-running lineage tracked at entity-jingle-fallacy: its two inputs are themselves already-synthesized observation notes — one a second-order fact about the vault's own promotion pipeline (observation-2026-08-16-capture-dropped-two-puckette-lineage-mentions-before-promotion), the other a first-order-content-bridge synthesis (the Petiška/Garfield/RAG note). The two target notes are not peers in the vault's own bridge taxonomy: one diagnoses a real-world citation dispute, the other diagnoses the vault's internal cross-referencing. This is structurally the same mismatch the vault already named and ruled non-bridging in observation-petiska-and-gates-jevons-liang-olsder-bridges-are-two-species-not-one-relationship — two notes carrying the same verdict-label without being the same kind of finding.

Grounding does not converge. The vault's established test for a genuine meta-level "confirmed bridge" (established at observation-ml-ad-en-gedi-and-gates-jevons-false-friend-notes-are-a-confirmed-bridge and repeated at observation-gates-jevons-liang-1979-olsder-1975-cosine-pairing-is-confirmed-bridge) is two independent routes into one shared grounding document. Kelly-Luccioni grounds directly and only in Steck, Ekanadham & Kallus's cosine-arbitrariness finding — cited in its own frontmatter with the exact quote "cosine-similarity can yield arbitrary and therefore meaningless `similarities'" (source: arXiv 2403.05440, claim-cosine-similarity-of-embeddings-can-be-arbitrary). Puckette-Petiška's body text does not cite that paper or that claim-note at all; its stated grounding is "the same mechanism already diagnosed at this identical cosine value in observation-gates-jevons-cosine-pairing-is-embedding-false-friend and observation-rhw-1986-werbos-docsub-cosine-bridge-is-thematic-not-causal" — and the vault's own observation-rhw-docsub-gates-jevons-meta-pairing-is-a-hedged-bridge already found that RHW/DOCSUB's grounding is "entirely internal Werbos-history legwork," not the Steck paper. So Puckette-Petiška's grounding chain runs, in one of its two cited branches, to a document that never reaches the arbitrariness paper — it does not converge with Kelly-Luccioni's grounding the way a confirmed bridge requires.

Discriminating tests differ in kind, not just content. Kelly-Luccioni's operative test is "whose error does the correction repair, and at what layer" — an error-provenance test built for citation-accuracy disputes. Puckette-Petiška's operative test is "does a real, checkable connective document exist anywhere in the vault treating one note as evidence for or context for the other" — a vault-graph-connectivity test explicitly borrowed from the buried-common-ancestor-bridge precedent one level up in the Gates-Jevons/Puckette lineage (observation-puckette-gates-jevons-fourth-instance-pairing-is-buried-common-ancestor-bridge). These are two different instruments, each purpose-built for a different job; their shared success at dissolving a resemblance is not evidence the two things they were run against are related to each other.

No wikilink connects the two target notes in either direction, and no document elsewhere in the vault was located that treats one as evidence or context for the other (checked by direct grep across 30-notes/ and 40-entities/ for cross-mentions of Kelly/Irwin/Mandel/Luccioni against Puckette/Petiška/Matthew-effect terms — zero hits in either direction).

Claim: entity-jingle-fallacy's hub tracking is asymmetric between the two notes, which is itself the more interesting finding than the false-friend verdict

entity-jingle-fallacy's changelog explicitly logs Puckette-Petiška as its "2026-08-31... sixth meta-level hop" and lists it under "Related false-friend instances." The same hub's "Related false-friend instances" list (as of this investigation: ML/AD-En-Gedi, Hawks-Jeffress, Falcon-Helmholtz, Liang-Olsder, Gates-Jevons) does not include Kelly-Luccioni at all, despite Kelly-Luccioni carrying the identical "embedding false friend" verdict, citing the identical grounding paper, and pre-dating the Puckette-Petiška entry by ten days (created 2026-08-21 vs. 2026-08-31). A grep of 40-entities/entity-megan-o-kelly.md, entity-alexandra-sasha-luccioni.md, and entity-david-r-mandel.md for "jingle" or "false friend" returns no hits. Kelly-Luccioni appears to be a first-order false-friend verdict produced entirely outside the Gates-Jevons/Puckette/Matthew-effect meta-hop lineage that this hub otherwise tracks — an orphan instance from the hub's own point of view, not a member of the family the topic's phrasing implicitly assumed it belonged to.

Further leads

• entity-jingle-fallacy's "Related false-friend instances" list omits observation-kelly-luccioni-cosine-pairing-is-embedding-false-friend and possibly other object-level false-friend verdicts produced outside the Gates-Jevons/Puckette chain — worth a pass to reconcile the hub's changelog against every note carrying the "embedding false friend" verdict, per source entity-jingle-fallacy (40-entities/entity-jingle-fallacy.md, Tier 1, vault-internal).
• observation-2026-08-16-capture-dropped-two-puckette-lineage-mentions-before-promotion documents a promotion-pipeline gap (mentions surviving a capture but not the promoted note) that could generalize beyond this one instance — worth checking whether other 2026-08 captures have similarly-dropped mentions, per the same source note (Tier 1, vault-internal).

Entity candidates

• Harald Steck, Chaitanya Ekanadham, Nathan Kallus — concept/authors — their WWW 2024 finding ("Is Cosine-Similarity of Embeddings Really About Similarity?") is the FOUNDATIONAL document every "embedding false friend" verdict in this entire lineage measures itself against; flagged first because the whole genre's discriminating power rests on their result and no entity page currently seems to exist for them specifically (only the derived claim-note).
• entity-jingle-fallacy — concept (existing hub) — the hub whose own changelog is asymmetric between the two notes this capture compared; worth a maintenance pass, not a new page.
• Robert K. Merton — person — the earlier, foundational figure the Petiška/Matthew-effect side of this pairing measures its subject against (a 2023 LLM finding compared to Merton's 1968 naming of the effect); already has an entity page (entity-robert-merton.md).
• Eugene Garfield — person — the citation-metrics figure whose decades-long warning the Petiška note treats as the missing predecessor to the ChatGPT finding; already has an entity page.
• Miller Puckette — person — the figure whose dropped lineage-mention triggered the other half of the Puckette-Petiška pairing; already has an entity page (entity-miller-puckette.md).