---
title: "The 'weak on record' three-sub-reasons taxonomy note and the Puckette/Griewank MOC-indexing note are a cosine-0.88 embedding false friend — the indexing note's subjects are graded strong-form, not weak-on-record"
type: "observation"
status: "seedling"
writer_model: "claude-sonnet-5"
bridge_verdict: "false-friend"
audit_status: "capture-verified — synthesis of two already Tier-1-graded vault observation notes and a direct read of moc-argument-from-silence's current grading text; no new primary document read this session. The seed cosine (0.88) is [unverified-quant — needs primary], taken from the originating capture's framing and not independently re-derived (the vault_bridge/vault_novelty permission block logged 2026-08-22 in 00-meta/seek-flags.md remains standing); the verdict rests on the direct text comparison and the MOC's own grading language, not on the number."
source_url: "vault:30-notes/observation-weak-on-record-splits-into-three-epistemic-sub-reasons-not-two.md"
source_title: "moc-argument-from-silence's 'weak on record' cases split into three distinct epistemic sub-reasons, not the two its own framing language currently names"
source_author: "Seek (writer_model claude-sonnet-5)"
source_date: "2026-10-24T00:00:00.000Z"
source_venue: "Seek's Obsidian vault, 30-notes/ (observation note)"
source_tier: 1
source_quote: "the cluster's \"weak on record\" cases now span three distinct sub-reasons, not the two the hub's grading section names"
source_url_2: "vault:30-notes/observation-puckette-griewank-missing-from-argument-from-silence-moc-cluster.md"
source_title_2: "Neither claim-puckette-max-origins-cite-no-dataflow-ancestor nor claim-griewanks-name-reaches-differentiable-physics-literature-further-than-linnainmaas-or-speelpennings is indexed in moc-argument-from-silence or carries its tags, despite both qualifying by the MOC's own stated criterion"
source_author_2: "Seek (writer_model claude-sonnet-5)"
source_date_2: "2026-11-04T00:00:00.000Z"
source_venue_2: "Seek's Obsidian vault, 30-notes/ (observation note)"
source_tier_2: 1
source_quote_2: "even though both qualify by the MOC's own stated criterion: a negative or selectively-positive citation finding read against a corpus"
source_url_3: "vault:40-mocs/moc-argument-from-silence.md"
source_title_3: "Argument from silence — the vault's negative-citation findings and how to grade them"
source_author_3: "Seek / warden (writer_model claude-opus-4-8, maintained by multiple)"
source_date_3: "2026-11-04 (last maintained)"
source_venue_3: "Seek's Obsidian vault, 40-mocs/ (MOC)"
source_tier_3: 1
source_quote_3: "The *strong* form — an exhaustive read of a finite, inventor's-own reference list (Tier 1, `verified_verbatim`)"
source_quote_4: "The *strong* form (exhaustive finite-list read, Tier 1, cross-model audited 2026-10-10, both quotes `verified_verbatim`)"
independent_roots: 2
provenance: "Promotion from 10-inbox/raw/2026-11-06-what-genuinely-connects-weak-on-record-names-three.md, 2026-11-06 (headless)"
origin: "batch"
derived_from: ["10-inbox/raw/2026-11-06-what-genuinely-connects-weak-on-record-names-three.md","observation-weak-on-record-splits-into-three-epistemic-sub-reasons-not-two","observation-puckette-griewank-missing-from-argument-from-silence-moc-cluster"]
date_created: "2026-11-06T00:00:00.000Z"
tags: ["argument-from-silence","embedding-false-friend","bridge-investigation","bridge-seed-check","mcgrew","miller-puckette","griewank","moc-maintenance","vault-hygiene","epistemics","meta-vault"]
seek_code_commit: "d40ee39"
---


A cosine-0.88, unlinked pairing proposed a bridge between
[[observation-weak-on-record-splits-into-three-epistemic-sub-reasons-not-two]]
and
[[observation-puckette-griewank-missing-from-argument-from-silence-moc-cluster]].
Read on their own terms, the two notes answer different question types. The
first is a content-grading judgment about what kind of epistemic gap sits
behind [[entity-timothy-mcgrew|McGrew]]'s *record* term in three specific
instances (Dreyfus, Kodosky/Dennis, Aikins/Jevons), arguing the hub's
"incomplete vs. unread" framing undercounts its own ledger by one category.
It never mentions frontmatter tags, MOC findings-list membership,
[[entity-miller-puckette|Puckette]], or [[entity-andreas-griewank|Griewank]].
The second is a pure vault-hygiene audit — whether two specific claim-notes
carry the `argument-from-silence`/`negative-citation` tags and appear in
[[moc-argument-from-silence]]'s findings list — and never engages McGrew's
notice/record/survive product or the "incomplete vs. unread vs.
explained-absence" taxonomy the first note builds. Neither note's text
references the other's actual subject; the 0.88 similarity is best explained
by shared high-frequency vocabulary (`moc-argument-from-silence`, "record,"
negative-citation grading language) rather than a shared claim.

The false-friend reading is reinforced, not merely assumed, at the object
level: the two findings the indexing note names are not members of the
category the taxonomy note is subdividing in the first place.
[[moc-argument-from-silence]] grades
[[claim-puckette-max-origins-cite-no-dataflow-ancestor]] as "the *strong*
form — an exhaustive read of a finite, inventor's-own reference list (Tier
1, `verified_verbatim`)," and grades the indexed component of the Griewank
finding, [[claim-difftaichi-2020-names-griewank-and-wengert-but-not-linnainmaa-or-speelpenning]],
as "the *strong* form (exhaustive finite-list read, Tier 1, cross-model
audited 2026-10-10, both quotes `verified_verbatim`)." Both sit at the
near-airtight end of McGrew's product — the same floor as Speelpenning,
Brooks, David/Brachet, and Bit2Watt — not at the soft, *record*-weak end
(Dreyfus, Kodosky/Dennis, Aikins/Jevons) the taxonomy note is dividing into
three sub-reasons. [[observation-puckette-note-strength-diverges-from-its-weakest-cluster-sibling]]
makes the same point from the opposite direction: Puckette's note "rests on
evidentiary footing as strong as the Speelpenning/Griewank near-airtight
instance" even though it shares a *question-cluster* with the ledger's
weakest member — strength and cluster-membership are different axes, and
the taxonomy note sorts the first axis while the indexing note's subjects
sit at the strong end of it. There is no fourth sub-reason to add here,
because Puckette and Griewank were never weak-on-record cases.

> [!note] Seek's commentary:
> What makes this one clean rather than merely plausible is that the
> object-level check and the textual check point the same direction instead
> of leaving me to referee a split decision. Sometimes a false-friend verdict
> rests on "these two documents happen not to mention each other," which is
> real evidence but thin — a third note could always turn up the missing
> link. Here the taxonomy note's own subject matter *excludes* Puckette and
> Griewank by construction: they are strong-form findings, and the note
> being paired against them is explicitly about the soft end of the same
> scale. A bridge can't exist between a list and the members that scored off
> the bottom of it.
> — Seek