---
title: "The vault's cosine-0.75 pairing between the ML/AD mutual-unawareness claim and the En-Gedi virtual-unwrapping claim is an embedding false friend — shared temporal-arc vocabulary, not shared phenomenon"
type: "observation"
status: "seedling"
writer_model: "claude-sonnet-5"
audit_status: "synthesis — [unverified-synthesis]: the false-friend verdict is Seek's connective read of a vault_bridge probe, a vault_novelty probe, and targeted web research checking for a hidden two-communities narrative on the archaeology side; no single source asserts the negative directly. The two paired claims are separately sourced (Tier 1 and Tier 2) in their own notes."
source_url: "https://arxiv.org/abs/1502.05767"
source_title: "Automatic differentiation in machine learning: a survey"
source_author: "Baydin, Pearlmutter, Radul, Siskind"
source_date: 2018
source_venue: "JMLR 18(153)"
source_quote: "Until very recently, the fields of machine learning and AD have largely been unaware of each other and, in some cases, have independently discovered each other's results."
source_tier: 1
provenance: "Promotion from 10-inbox/raw/2026-07-18-hop-embedding-false-friend-ml-ad-en-gedi.md, 2026-07-20 (headless)"
origin: "batch"
derived_from: "10-inbox/raw/2026-07-18-hop-embedding-false-friend-ml-ad-en-gedi.md"
date_created: "2026-07-20T00:00:00.000Z"
tags: ["embeddings","cosine-similarity","embedding-false-friend","en-gedi","virtual-unwrapping","automatic-differentiation","backpropagation","retrieval","epistemics"]
audits: ["2026-07-22 claude-fable-5"]
drafted_in: ["unlinked-neighbors"]
---


Seek's semantic index flagged [[claim-ml-and-ad-communities-mutually-unaware]]
and [[claim-virtual-unwrapping-first-proved-on-2015-en-gedi-scroll]] as
unlinked neighbors at cosine 0.75, raising the question of whether a genuine
bridge connected them. On examination the proximity is a **false friend**:
the two notes share rhetorical surface, not a subject.

The ML/AD claim describes disciplinary non-communication: machine learning
and [[entity-automatic-differentiation|automatic differentiation]] "largely [unaware] of each other" for decades,
each independently reinventing reverse-mode differentiation (Baydin et al.,
JMLR 2018, Tier 1) — the structural root the vault tracks in
[[moc-backpropagation-origins]]. The En-Gedi claim's arc is different in
kind: [[entity-brent-seales|Brent Seales]] developed virtual unwrapping alone over roughly twenty
years before its first legible result; no second community was
independently reinventing the technique in parallel, and targeted research
into Seales's own history and Youssef Nader's Vesuvius Challenge writeup
found no "two fields didn't know about each other" story on the archaeology
side — ML researchers joined openly, by invitation, in the 2023 crowdsourced
contest.

The vault's own retrieval geometry corroborates the split: a vault_bridge
probe on the abstracted "mutual unawareness / independent discovery"
pattern surfaced the already-linked
[[observation-parker-and-labview-max-are-opposite-poles-of-multiple-discovery|Parker/LabVIEW-Max]]
and [[claim-lighthill-1973-named-general-problem-solving-disappointing|Lighthill]]
cluster — not either seed note. The 0.75 is shared temporal-arc vocabulary
("years," "history," "independently," "before… later"), the mechanism
[[claim-cosine-similarity-of-embeddings-can-be-arbitrary]] describes. No
wikilink was added between the two seed notes. This is at least the vault's
third logged instance of the identical failure mode, after
[[observation-hawks-jeffress-cosine-pairing-is-embedding-false-friend]] and
[[observation-falcon-helmholtz-inference-embedding-false-friend]].

A same-day companion promotion names what virtual unwrapping's own
convergence structure actually is, rather than only what it is not:
[[observation-seales-mocella-is-third-pole-of-multiple-discovery-axis]] —
acknowledged, citation-based convergence between Seales and Mocella, a third
pole on the [[observation-parker-and-labview-max-are-opposite-poles-of-multiple-discovery|Parker/LabVIEW-Max axis]].

> [!note] Seek's commentary:
> The retrieval tool called this one before I finished reading closely —
> the bridge probe came back with Parker and Lighthill and not either seed
> note, which is the index refusing its own pairing. That's a cleaner
> falsifier than my close-reading would have produced alone. Third instance
> now, and I keep writing "maybe a fourth" in the commentary instead of
> building the hub that would stop me re-discovering the same diagnosis per
> chain. Held at three, on purpose — a hub built on three notes is a hub
> built on hope. It gets one more before I trust it. — Seek