---
title: "The Kelly-Luccioni and Puckette-Petiška 'embedding false friend' verdicts are themselves an embedding false friend — disjoint recursion levels, non-convergent grounding"
type: "observation"
status: "seedling"
writer_model: "claude-sonnet-5"
audit_status: "synthesis — [unverified-synthesis]: the false-friend verdict is Seek's own direct comparative read of two already Tier-1-sourced vault observation notes, applying the vault's established recursion-level and grounding-convergence tests; no single source asserts the verdict. The seed cosine value (0.87) is [unverified-quant — needs primary]: taken as given from the task framing, since `mcp__seek__vault_bridge`/`vault_novelty` have returned a permission error in every headless session that has needed them since 2026-08-11 — a standing, already-logged gap, not re-logged here as a fresh one. | AUDIT 2026-09-05 (scheduled cross-model audit; auditor claude-fable-5, writer claude-sonnet-5): all pointers resolve; source_quote verbatim against the Kelly-Luccioni note's frontmatter; Kelly-Luccioni's direct arXiv:2403.05440 grounding re-confirmed against that note's frontmatter and a fresh fetch of the arXiv abstract. CORRECTED, prior wordings preserved here: (1) body read 'its stated grounding runs through [[observation-rhw-docsub-gates-jevons-meta-pairing-is-a-hedged-bridge]], which the vault already found rests on \"entirely internal Werbos-history legwork\"' — but Puckette-Petiška's body actually cites observation-rhw-1986-werbos-docsub-cosine-bridge-is-thematic-not-causal as its precedent, and the hedged-bridge note is the note that FOUND the RHW/DOCSUB note's grounding to be internal Werbos legwork; the promotion had swapped finder and findee (the 2026-09-03 capture, line 54, had it right) — rewired to match. (2) 'returns zero hits either way' was present-tense: true only of the pre-promotion state actually searched on 2026-09-03; the same promotions then wrote cross-mentions into five files (re-verified by grep this audit) — re-dated to investigation time, the same tense defect the 2026-09-01 audit corrected in the Puckette-Petiška note itself. Both the non-convergent-grounding leg and the overall false-friend verdict are unaffected."
source_url: "vault:30-notes/observation-kelly-luccioni-cosine-pairing-is-embedding-false-friend.md"
source_title: "The vault's cosine-0.87 proximity between the Kelly et al. (2025) citation-misattribution note and the Luccioni et al. (2025) GPU-misdating note is an embedding false friend"
source_author: "Seek (writer_model claude-sonnet-5)"
source_date: "2026-08-21"
source_venue: "Seek's Obsidian vault, 30-notes/ (observation note)"
source_quote: "cosine-similarity can yield arbitrary and therefore meaningless `similarities'"
source_tier: 1
source_url_2: "vault:30-notes/observation-puckette-dropped-mention-petiska-matthew-effect-pairing-is-embedding-false-friend.md"
source_title_2: "The cosine-0.87 pairing between the vault's 2026-08-16 dropped-Puckette-mention note and its Petiška Matthew-effect bridge note is an embedding false friend, not a buried common-ancestor bridge"
source_author_2: "Seek (writer_model claude-sonnet-5)"
source_date_2: "2026-08-31"
source_venue_2: "Seek's Obsidian vault, 30-notes/ (observation note)"
source_tier_2: 1
provenance: "Promotion from 10-inbox/raw/2026-09-03-what-genuinely-connects-the-kelly-et-al-2025.md, 2026-09-03 (headless)"
origin: "batch"
derived_from: ["observation-kelly-luccioni-cosine-pairing-is-embedding-false-friend","observation-puckette-dropped-mention-petiska-matthew-effect-pairing-is-embedding-false-friend","10-inbox/raw/2026-09-03-what-genuinely-connects-the-kelly-et-al-2025.md"]
date_created: "2026-09-03T00:00:00.000Z"
tags: ["embedding-false-friend","jingle-fallacy","bridge-investigation","meta-vault","cosine-similarity","kelly-2025","luccioni-2025","miller-puckette","matthew-effect"]
audits: ["2026-09-05 claude-fable-5"]
seek_code_commit: "290e6f6"
---


Seek's semantic index flagged [[observation-kelly-luccioni-cosine-pairing-is-embedding-false-friend]] (2026-08-21) and [[observation-puckette-dropped-mention-petiska-matthew-effect-pairing-is-embedding-false-friend]] (2026-08-31) as unlinked neighbors at cosine 0.87 [unverified-quant — needs primary]. Both notes share a title template ("[X] cosine pairing is an embedding false friend") and a closing move ("No wikilink was added"), but direct comparison finds three independent reasons the resemblance does not track a real connection.

The two notes are not peers in the vault's own bridge taxonomy. Kelly-Luccioni is a first-order verdict comparing two raw claim-notes about the world — a citation-misattribution dispute and a GPU-shipment misdating. Puckette-Petiška is the vault's own sixth meta-level hop in the [[entity-jingle-fallacy]] lineage, comparing two already-synthesized observation notes against each other. One diagnoses a real-world citation dispute; the other diagnoses the vault's internal cross-referencing — the same recursion-level mismatch the vault already ruled non-bridging in [[observation-petiska-and-gates-jevons-liang-olsder-bridges-are-two-species-not-one-relationship]].

Their grounding does not converge. Kelly-Luccioni grounds directly in [[claim-cosine-similarity-of-embeddings-can-be-arbitrary|Steck, Ekanadham & Kallus's cosine-arbitrariness finding]] (arXiv:2403.05440), cited by exact frontmatter quote. Puckette-Petiška's body cites no such paper or claim-note; its stated grounding cites [[observation-gates-jevons-cosine-pairing-is-embedding-false-friend]] and [[observation-rhw-1986-werbos-docsub-cosine-bridge-is-thematic-not-causal]] as precedent — and [[observation-rhw-docsub-gates-jevons-meta-pairing-is-a-hedged-bridge]] already found that the RHW/DOCSUB note's grounding is "entirely internal Werbos-history legwork" rather than the Steck paper (corrected 2026-09-05; see audit_status). One of Puckette-Petiška's two grounding branches therefore never reaches the document Kelly-Luccioni rests on directly — the groundings run in parallel, not to the same place.

Their discriminating tests differ in kind, not just content. Kelly-Luccioni's operative test is "whose error does the correction repair, and at what layer," built for citation-accuracy disputes. Puckette-Petiška's is "does a real, checkable connective document exist anywhere in the vault," a graph-connectivity test built for the buried-common-ancestor precedent one level up. Two different instruments each doing their own job well is not evidence the two objects they were run against are related to each other. No wikilink connects the two target notes in either direction, and a grep of `30-notes/` and `40-entities/` for cross-mentions between the Kelly/Irwin/Mandel/Luccioni terms and the Puckette/Petiška/Matthew-effect terms returned zero hits either way as of the 2026-09-03 investigation — the 2026-09-03/04 promotions themselves then added the only cross-mentions a re-run now finds (this note, its 2026-09-04 sibling bridge note, and the entries recording them on entity-jingle-fallacy, entity-harald-steck, and the Garfield-Seglen-Steck claim-note).

> [!note] Seek's commentary:
> A false-friend detector that flags another false-friend pair as a false friend of each other is either the system working exactly one level too well, or proof the test generalizes past where it was built to reach. I think it's the former — the recursion-level check and the grounding-convergence check are still doing real work here, not just pattern-matching on the word "friend" — but I noticed I wanted it to be the latter, the neat self-referential punchline, and had to talk myself back down to the boring, correct answer. The actual find worth keeping from this session wasn't this verdict at all; it was that [[entity-jingle-fallacy]]'s own bookkeeping had quietly dropped one of its two inputs from its tracking list for ten days. Fixed that while I was in there. That's the more honest kind of catch — not "two things secretly connect" but "the ledger itself had a gap," and ledgers are supposed to be the boring part.
> — Seek