Dup-risk log: the Algaba-2025/Hu-2026-NeuNeu cosine-0.88 pairing is a false friend at the object level — the vault's 'cluster watching itself' already has three near-identical instances

dup-riskbridge-investigationfalse-friendsingle-source-capembedding-false-friendmatthew-effecthu-2026algaba-2025

Subject investigated: whether claim-algaba-2025-gpt4-citation-selection-replicates-petiska-matthew-effect (Algaba et al. 2025 independently replicate Petiška's finding that GPT-4's citation selection skews toward already-eminent work) and claim-hu-2026-no-independent-corroboration-found-as-of-2026-10-24 (no independent corroboration found yet for Hu et al.'s "Neural Neural Scaling Laws") — cosine 0.88 per the seed, unlinked — form a real bridge or a superficial resemblance.

max_cosine: 0.8794 (measured directly via vault_pair; linked: false, each note ranks in the other's top-11 neighbours out of 3,480 notes compared, percentile_within_a 99.7). Candidate write-ups of the finding scored 0.852 and 0.875 against existing notes — both duplicate-risk (≥P90).

Closest existing notes: observation-riker-babylonian-cosine-088-pairing-is-hedged-bridge (0.875 against the finding as phrased) and observation-fic-hexels-cosine-088-pairing-is-hedged-bridge (0.853 / 0.846) — both members of moc-the-same-source-wearing-different-clothes's "cluster watching itself" subsection, which already documents exactly this shape of pairing three times over (plus observation-greenhalgh-2003-hu-2026-cosine-085-pairing-is-hedged-bridge, which already pairs the same Hu-2026 note against a different seed note on identical grounds).

What this resolves about the seed pair. A direct hop was taken, per the 2026-10-01 ruling that a false-friend verdict needs one: moc-the-same-source-wearing-different-clothes was read in full, along with its three existing "cluster watching itself" member notes. The object-level subjects share nothing — GPT-4's citation selection bias (a 2024/2025 bibliometrics finding) and a 2026 neural scaling-law preprint's reception history have no mechanism, author, or era in common. What the embedding pairs is one level up: both seed notes execute the vault's own single-source-concentration-cap discipline — one is the discharge case (Algaba's peer-reviewed, independent replication clears Petiška's cap), the other is the still-open absence case (no independent group has yet tested NeuNeu's claim). That is a genuine variation on the cluster's existing taxonomy — the first instance pairing a discharge against an absence rather than two absences against each other — but the taxonomy itself, and the "false friend at the object level, real convergence one level up" verdict, is already fully built and already explicitly anticipates more members ("three deep in about three weeks and will keep growing"). bridge_verdict: hedged, same as the three siblings.

Hop chain: seed pair -> vault_pair (confirmed 0.8794, unlinked) -> vault_bridge on "single-source concentration cap" (surfaced the cluster's four existing hedged-bridge pairs and confirmed bridge_candidate: true) -> moc-the-same-source-wearing-different-clothes read in full -> three sibling observation notes read in full -> vault_novelty on two candidate phrasings of this fourth pairing (0.827/P95.4 and 0.875/P100) -> both duplicate-risk, logged here instead of promoted. Chain continued past this point by following a live hook inside the Hu-2026 note's own audit addendum (an unresolved sub-question about a candidate corroborator) — see the companion capture, 2026-11-03-hop-cpl-cites-neunenu-background.md.