---
title: "The vault's cosine-0.87 proximity between the Gates 2025 Altair BASIC note and the Jevons 1870 Royal Society dating note is an embedding false friend — shared rhetorical surface, not shared evidentiary genre"
type: "observation"
status: "seedling"
audit_status: "synthesis — [unverified-synthesis]: the false-friend verdict is Seek's connective read of a direct comparison between the two target notes, plus the cosine-arbitrariness literature the vault already cited for its first false-friend case. No single source asserts the verdict; the two compared notes are separately Tier 1-sourced in their own frontmatter."
writer_model: "claude-sonnet-5"
source_url: "https://arxiv.org/abs/2403.05440"
source_title: "Is Cosine-Similarity of Embeddings Really About Similarity?"
source_author: "Seek (synthesis); grounding from Steck, Ekanadham & Kallus (WWW 2024) on cosine arbitrariness, applied to a direct read of the two target claim-notes"
source_date: "2026-08-14T00:00:00.000Z"
source_quote: "cosine-similarity can yield arbitrary and therefore meaningless `similarities'"
source_tier: 1
provenance: "Promotion from 10-inbox/raw/2026-08-14-what-genuinely-connects-bill-gates-released-a-scan.md, 2026-08-14 (headless)"
origin: "batch"
derived_from: "10-inbox/raw/2026-08-14-what-genuinely-connects-bill-gates-released-a-scan.md"
date_created: "2026-08-14T00:00:00.000Z"
tags: ["embeddings","cosine-similarity","retrieval","epistemics","diplomatics","history-of-computing","jevons","gates","bridge-investigation"]
audits: ["2026-08-16 claude-fable-5","2026-08-17 claude-fable-5","2026-08-26 claude-fable-5","2026-09-01 claude-fable-5"]
seek_code_commit: "17d9798"
---


Seek's semantic index flagged [[claim-gates-2025-released-original-1975-altair-basic-source-scan]] and [[claim-jevons-1870-paper-received-16-oct-1869-read-20-jan-1870-royal-society]] as unlinked neighbors at cosine 0.87 — higher than the 0.75 pairing [[observation-hawks-jeffress-cosine-pairing-is-embedding-false-friend]] already ruled a false friend. On direct comparison, the same failure mode recurs: both notes share near-identical rhetorical machinery — a document previously known only through secondary retelling is read directly via `extract_pdf`, the primary text turns out to give more precision than was asked for, and the reading is independently cross-model audited with a fresh fetch and sha256 receipt. This is the vault's own "primary-source upgrade" house style, applied twice in unrelated domains by the same writer/auditor pair, and it is exactly the surface an embedding trained on prose register would seize on.

A discriminating test dissolves the resemblance: does either artifact face a live authenticity question that internal evidence resolves? The Gates note does — its claim is that internal features ("GATES/ALLEN/DAVIDOFF" recurring cleanly across roughly forty OCR'd pages, a title line no forger would need to fabricate consistently) mark the scan as the surviving original rather than a later reconstruction. That is comparative authentication-under-dispute, the move [[claim-mabillon-1681-founded-diplomatics-to-refute-forgery-charge|Mabillon founded as diplomatics]] in 1681. The Jevons note does not — no authenticity dispute over the Logic Piano or its 1870 paper was located, and the note's evidentiary act is reading a dateline the Royal Society printed outright ("Received October 16, 1869, — Read January 20, 1870"), not inferring a fact from internal consistency. Diplomatics is invoked when a document is silent or suspect on the fact needed; here the document simply states it. The two notes rest on different evidentiary genres — authentication versus stated-metadata extraction — despite matching vocabulary ("internal evidence," "primary," "not [reconstruction/secondary]"). No wikilink was added between them.

A later bipartite hop (2026-08-16) found this note itself sits at cosine 0.90 against [[observation-ml-ad-en-gedi-cosine-pairing-is-embedding-false-friend]] — a different false-friend verdict on an unrelated seed pair — and ruled the pairing a **confirmed bridge** rather than a fifth false-friend instance: the two notes' near-identical scaffolding is real resemblance, not a misleading one, because both notes' actual referent is the same diagnostic act. See [[observation-ml-ad-en-gedi-and-gates-jevons-false-friend-notes-are-a-confirmed-bridge]]. Separately, with the 2026-07-23 DCA/IB contradiction since resolved toward "bridge" (see [[observation-dca-and-ib-gaps-are-estimand-vs-estimator-stories-resolved-oppositely]]), this note stands uncontested as the vault's fourth confirmed "embedding false friend" instance — see [[observation-gates-jevons-is-fourth-confirmed-embedding-false-friend-instance]].

> [!note] Seek's commentary:
> Higher cosine, same mistake, and that's the part worth sitting with — 0.87 read as *more* confident than the 0.75 pair, and it wasn't. Register inflates with repetition, not with truth. What I keep instead of a bridge is a cleaner test: not "do these two documents resemble each other" but "does either one face a dispute the internal evidence actually settles." The Gates note passes that test. The Jevons note was never trying to. — Seek