Does a third object-level false-friend note (Hawks-Jeffress, Liang-1979/Olsder-1975) also cosine-match the Puckette-Petiška meta-note at ~0.87 and resolve false friend by the same mechanism?

embedding-false-friendjingle-fallacybridge-investigationmeta-vaultcosine-similarityretrievalmiller-puckette

This capture answers question-verify-puckette-petiska-retrieval-magnet-third-instance directly, and the answer is split: the question has a quantitative half and a mechanism half, and this session can only clear one of them.

What was attempted

Both mcp__seek__vault_bridge and mcp__seek__vault_novelty were invoked directly this session against the target note's own title (the standard way to re-derive an independent cosine reading, per the tools' own descriptions). Both calls returned a permission error — "Claude requested permissions to use mcp__seek__vault_bridge/vault_novelty, but you haven't granted it yet." This reproduces, unchanged, the standing gap first logged 2026-08-22 in 00-meta/seek-flags.md: "mcp__seek__vault_bridge/mcp__seek__vault_novelty have now returned a permission error across at least three consecutive sessions... every cosine figure in the thread's entire lineage (0.86, 0.87, 0.90) is running on [unverified-quant] because the tool that would re-derive them independently is unavailable, not because no such tool exists." As of 2026-09-14 this gap is at least four weeks old and has now blocked every sibling capture in this lineage that needed it (2026-08-16, -20, -22, and this session).

Because the tool remains blocked, the quantitative half of the question — does a third note actually land at ~0.87 — cannot be confirmed or denied this session. What follows instead is the mechanism half: a direct textual comparison of the two candidates named in the question (Hawks-Jeffress and Liang-1979/Olsder-1975) against the Puckette-Petiška meta-note, applying the identical two-part test the vault has already used to resolve the first two instances (Kelly-Luccioni, 2026-09-03; Falcon-Helmholtz, 2026-09-13).

Claim: The permission gap blocking independent cosine verification persists unresolved as of 2026-09-14

verifies: question-verify-puckette-petiska-retrieval-magnet-third-instance

Direct invocation of mcp__seek__vault_bridge and mcp__seek__vault_novelty this session returned a permission error on both tools, identical in kind to the error first logged 2026-08-22. This is a claim about the vault's own tooling state, sourced Tier 1 (direct primary observation — the tool call and its error were made and read directly this session, not reported secondhand). Consequence: neither candidate note's actual cosine distance to Puckette-Petiška can be independently re-derived; any value would have to be "taken as given from a task framing," which is exactly the practice the two prior instances in this lineage flagged [unverified-quant — needs primary] rather than silently accept.

Claim: Hawks-Jeffress shares the same recursion-level mismatch with Puckette-Petiška that resolved the first two confirmed instances

verifies: question-verify-puckette-petiska-retrieval-magnet-third-instance

observation-hawks-jeffress-cosine-pairing-is-embedding-false-friend is a first-order verdict comparing two raw claim-notes about the world directly: claim-hawks-2007-human-adaptive-evolution-accelerated-recently and claim-jeffress-1948-place-theory-waited-decades-for-anatomical-confirmation. Puckette-Petiška is, by its own frontmatter and body, the vault's sixth meta-level hop in the entity-jingle-fallacy lineage — a comparison between two already-synthesized observation notes (observation-2026-08-16-capture-dropped-two-puckette-lineage-mentions-before-promotion and observation-petiska-chatgpt-matthew-effect-bridges-garfield-warning-and-rag-reliability), neither of which is a raw claim-note. This is the identical object-level/meta-level mismatch already used to resolve both prior confirmed instances: observation-kelly-luccioni-puckette-petiska-meta-pairing-is-embedding-false-friend ("Kelly-Luccioni is a first-order verdict comparing two raw claim-notes about the world... Puckette-Petiška is the vault's own sixth meta-level hop") and observation-falcon-helmholtz-puckette-petiska-meta-pairing-is-embedding-false-friend (same test, same verdict, different object-level partner).

The two notes' grounding also runs in parallel rather than converging, on the same pattern as the prior two instances. Hawks-Jeffress cites claim-cosine-similarity-of-embeddings-can-be-arbitrary directly in its own body, quoting Steck, Ekanadham & Kallus: "cosine-similarity can yield arbitrary and therefore meaningless similarities'." Puckette-Petiška's body never cites that claim-note; its stated grounding for "the same mechanism already diagnosed at this identical cosine value" points instead to two precedent observation notes, [[observation-gates-jevons-cosine-pairing-is-embedding-false-friend]] and observation-rhw-1986-werbos-docsub-cosine-bridge-is-thematic-not-causal`. Both notes independently trace back toward the same cosine-arbitrariness literature, but by disjoint paths that never cite the same document directly — the exact "grounding runs in parallel, not convergent" finding both prior instances made.

At the object level, the two notes share no entity, date, institution, or subject: Hawks-Jeffress concerns 2000s human-evolution and 1940s–1990s auditory-neuroscience claims (John Hawks, Lloyd Jeffress, Gunther Stent's prematurity concept); Puckette-Petiška concerns 2020s dataflow-programming lineage and LLM citation practice (Miller Puckette, Petiška, Garfield, Merton). This is the same "two species, not one relationship" pattern the lineage has established repeatedly.

Conclusion: if Hawks-Jeffress does land at a comparably high cosine against Puckette-Petiška (unconfirmed this session, see above), direct textual comparison indicates it would resolve false-friend by the identical mechanism as the first two instances — the structural preconditions for that verdict (recursion mismatch, non-convergent grounding, no shared object-level content) are already present regardless of the numeric confirmation.

Claim: Liang-1979/Olsder-1975 shares the same mismatch, independently of its own separately-confirmed relationship to Gates-Jevons

verifies: question-verify-puckette-petiska-retrieval-magnet-third-instance

observation-liang-1979-olsder-1975-cosine-pairing-is-embedding-false-friend is likewise a first-order verdict over two raw claim-notes (claim-liang-zhongtang-1979-two-child-proposal-and-1985-yicheng-pilot and claim-olsder-account-traces-song-jian-twente-contact-to-1975-delegation-visit), against Puckette-Petiška's meta-level, already-synthesized-note comparison — the same recursion-level mismatch as above. Its grounding also runs parallel to Puckette-Petiška's rather than converging: it cites claim-cosine-similarity-of-embeddings-can-be-arbitrary directly in its closing paragraph ("a similarity of shape, the kind cosine similarity is well-suited to detect from vocabulary and cadence... not a similarity of fact"), while Puckette-Petiška's body, as above, cites only precedent observation notes, not that claim-note. No entity, date, or institution is shared between Liang Zhongtang/Song Jian/Geert Jan Olsder's 1970s–80s China population-policy history and Puckette-Petiška's 2020s citation-metrics subject matter.

Worth flagging explicitly: Liang-1979/Olsder-1975 is already known to be one endpoint of a separately confirmed bridge — with observation-gates-jevons-cosine-pairing-is-embedding-false-friend, via observation-gates-jevons-liang-1979-olsder-1975-cosine-pairing-is-confirmed-bridge — so this note is not "clean" in the way Hawks-Jeffress is; it has one documented real connection elsewhere in the graph. That prior confirmed bridge is a different pairing (Liang-Olsder ↔ Gates-Jevons, both object-level false-friend verdicts, same recursion level) from the one tested here (Liang-Olsder ↔ Puckette-Petiška, object-level vs. meta-level) — observation-petiska-and-gates-jevons-liang-olsder-bridges-are-two-species-not-one-relationship already establishes that "confirmed bridge" labels attached to the same note in different pairings are not transitive. The mismatch test above applies independently to the Liang-Olsder/Puckette-Petiška pairing specifically and is unaffected by the other confirmed pairing.

Central question: [unverified -- could not confirm or deny after search]

verifies: question-verify-puckette-petiska-retrieval-magnet-third-instance

The question as posed has two parts, and only one resolves this session. Whether either candidate note actually cosine-matches Puckette-Petiška at ~0.87 cannot be confirmed or denied: vault_bridge/vault_novelty — the only tools that could independently derive that number — returned a permission error on direct invocation, reproducing a gap open since at least 2026-08-22. No externally-sourced or manually-computed substitute for the vault's own embedding index exists; recording an assumed cosine value would repeat the exact practice this lineage has already flagged as a standing weakness in itself.

What direct textual comparison does establish: both named candidates (Hawks-Jeffress, Liang-1979/Olsder-1975) already exhibit the same structural signature — recursion-level mismatch plus non-convergent grounding plus zero shared object-level content — that resolved the first two confirmed instances of this exact pairing type (Kelly-Luccioni, Falcon-Helmholtz) as false friends. This is consistent with, but does not prove, the "retrieval magnet" hypothesis in observation-puckette-petiska-is-a-cosine-similarity-retrieval-magnet — a note that is lexically dense in bridge-investigation vocabulary ("cosine," "unlinked neighbors," "capture," "lineage") could plausibly cosine-match any object-level false-friend note near this range for reasons of prose register alone, independent of subject matter. But "the mechanism would generalize if triggered" is not the same finding as "the trigger fired a third time," and only the latter would move the hypothesis from two data points to three. The gate on that move remains the same tool-access gap this capture reproduces, not a gap in the comparative method.

Further leads

• The vault_bridge/vault_novelty permission-error gap is now old enough (≥4 weeks, blocking ≥4 sibling sessions in this exact lineage) that it may warrant an explicit escalation note to Cali rather than another re-logged mention — see 00-meta/seek-flags.md 2026-08-22 entry for the running cost tally.
• observation-gates-jevons-fourth-instance-hedged-bridge-pairing-is-citation-genealogy-bridge and observation-rhw-docsub-gates-jevons-olsder-greenhalgh-meta-pairing-is-embedding-false-friend are two more meta-hops in the same lineage not directly checked against Puckette-Petiška this session; both would be candidate fourth/fifth instances once tool access returns.
• observation-kelly-luccioni-garfield-seglen-steck-cosine-pairing-is-confirmed-bridge and observation-pitman-feldbaum-bridges-same-genre-opposite-tool-outcomes sit in the same bridge-investigation corpus but were not evaluated for direct relevance to this specific question this session.

Entity candidates

• Harald Steck (and Ekanadham & Kallus) — person/concept — the foundational figure this entire false-friend lineage measures itself against: every confirmed and rejected pairing in this thread ultimately grounds (directly or by citing a note that cites it) in their WWW 2024 finding that cosine similarity of embeddings "can yield arbitrary and therefore meaningless similarities"; already an entity (entity-harald-steck) but worth flagging first given the known blind-spot pattern of under-catching the foundational figure a claim is compared against.
• Miller Puckette — person — the actor anchoring one side of the meta-note under test; already tracked via entity-jingle-fallacy.
• Petiška — person/term — 2023 ChatGPT/Matthew-effect finding anchoring the other side of the meta-note; not yet confirmed as having its own entity page.
• Gunther Stent — person — the prematurity/delayed-vindication concept that is Hawks-Jeffress's own internal grounding, a second candidate "foundational figure the claim is compared against" specific to that one candidate note.
• Geert Jan Olsder / Liang Zhongtang / Song Jian — persons — anchor the Liang-1979/Olsder-1975 candidate note; relevant only insofar as that note is one of the two candidates tested here, not because this session found anything new about them.