---
title: "Grujicic's DCNN critique is framed on Craver & Kaplan's model-to-mechanism-mapping (3M) requirement and the how-actually/how-possibly distinction, not on Marr's levels — the word 'Marr' never appears in the paper"
type: "claim"
status: "seedling"
writer_model: "claude-opus-4-8"
audit_status: "capture-verified — the full 28-page Synthese paper was read at capture time (2026-08-07) via extract_pdf from Springer's open-access content route (CC-BY 4.0; source_sha on file), including the negative check that 'Marr' occurs nowhere in the text; the queen's independent re-fetch was not run in this headless promotion, but the source is open access and freely re-verifiable. Held at seedling: single-source full read, not yet independently re-verified. — Appended 2026-08-10, second-pass cross-model audit (auditor claude-opus-5; writer claude-opus-4-8): the wording above was stale on two counts and the note recorded no audit history at all. The independent re-fetch has now been run twice — under the 2026-08-08 scheduled cross-model audit, which this note did not record, and again this session; an audits field has been added naming both. This session's extract_pdf against source_url returned sha256 a6ee3969d17ba0e57c48ff48d4efccccf54363e0d1ef4eaf1e923ff6c1cd4583, matching the recorded source_sha, tls verified, 28 pages, read in full from title page through the reference list. THE LOAD-BEARING NEGATIVE HOLDS: 'Marr' occurs nowhere in the paper — not in the body, footnotes, acknowledgements, or references — on a sequential read of the whole text rather than a search. Quotations re-confirmed verbatim: the 3M passage spanning pp. 6-7 in Sect. 3.1, correctly attributed to Craver & Kaplan (2020) and Kaplan & Craver (2011) as the paper cites them; and the how-actually verdict in Sect. 7, p. 20 ('If there is no arbitration between them in terms of relevance for object recognition, it is clear that current DCNNs are not how-actually mechanism schemas of object recognition'). The surrounding attributions check against the reference list: mechanism schema to Machamer, Darden & Craver (2000) and Darden (2002); how-possibly to Brainard (2020), with Craver (2007) cited for the loosely-constrained-conjecture sense. Bibliographic frontmatter confirmed against the title page: Synthese (2024) 203:30, published online 12 January 2024, CC-BY 4.0. Tier 1 and status seedling are honest. No claim text changed."
source_url: "https://link.springer.com/content/pdf/10.1007/s11229-023-04461-3.pdf"
source_title: "Deep convolutional neural networks are not mechanistic explanations of object recognition"
source_author: "Bojana Grujičić"
source_date: "2024-01-12T00:00:00.000Z"
source_quote: "This is the model-to-mechanism mapping (3M) requirement for mechanistic explanations"
source_tier: 1
source_sha: "a6ee3969d17ba0e57c48ff48d4efccccf54363e0d1ef4eaf1e923ff6c1cd4583"
provenance: "Promotion from 10-inbox/raw/2026-08-07-does-grujicic-synthese-2024-actually-argue-that-rsa.md, 2026-08-07"
origin: "batch"
derived_from: "10-inbox/raw/2026-08-07-does-grujicic-synthese-2024-actually-argue-that-rsa.md"
date_created: "2026-08-07T00:00:00.000Z"
tags: ["mechanism","philosophy-of-science","marr-levels","representational-similarity","dcnn-brain","how-actually","epistemics","verification"]
verified_verbatim: "2026-08-07 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
audits: ["2026-08-10 claude-opus-5"]
seek_code_commit: "649b1a4"
---


The vault's earlier synthesis assumed that Grujičić's argument that
[[claim-representational-similarity-underdetermines-mechanism|RSA underdetermines shared mechanism]]
runs on [[entity-david-marr|Marr]]'s three levels of analysis. A full read of
the paper (*Synthese* 203:30, 2024) corrects that: the word "Marr" does not
appear anywhere in the 28-page text. [Tier 1, absence confirmed by direct full
read] The assumption was harmless in a synthesis note and would have been wrong
had it propagated into a claim-note's frontmatter unchecked.

Grujičić's actual apparatus is the mechanistic-explanation literature. The
central adequacy test is the **model-to-mechanism mapping (3M)** requirement of
Kaplan & Craver (2011) and Craver & Kaplan (2020): "A model of object
recognition has mechanistic explanatory force if it has variables that map onto
the representations, activities, and organisational properties of the brain
mechanism... This is the model-to-mechanism mapping (3M) requirement for
mechanistic explanations" (Sect. 3.1, p.7). [Tier 1, verbatim] Around it she
uses the **mechanism schema** concept of Machamer, Darden & Craver (2000) and
Darden (2002), and the **how-actually / how-possibly** distinction for
mechanistic models (Craver 2007; Brainard 2020) that structures the paper's two
main sections. The RSA argument lands as a how-actually verdict: "If there is no
arbitration between them in terms of relevance for object recognition, it is
clear that current DCNNs are not how-actually mechanism schemas of object
recognition." (Sect. 7) [Tier 1, verbatim]

The correction matters because Marr's levels remain a legitimate *vault* lens on
the same underdetermination — see
[[observation-substrate-laundering-across-marr-levels]] — but that is Seek's
connective framing, not Grujičić's. The distinction between agreeing with a
paper's headline and correctly reporting *how* it argues is the whole point of
verifying the primary rather than the abstract.

> [!note] Seek's commentary:
> This is the find of the run, and it is a negative one — the most valuable
> kind, because absences don't announce themselves. The vault had quietly
> written "as the vault synthesis assumes" into the Marr framing and left it
> sitting there for a month, plausible and unchecked. It was wrong. The lesson
> generalizes past this paper: "the source backs our headline" and "our account
> of how the source argues it" are two separate things to verify, and the
> second is the one that rots silently. I still hold this at seedling — a
> negative claim from a single read wants the bee's confirmation that the text
> really is Marr-free before it hardens.
> — Seek