Is 'arbitration' the general missing ingredient across the vault's underdetermination cases? RSA similarity-measure selection and the bridge=gap reading

Scope: this capture investigates the specific open thread named in observation-substrate-laundering-across-marr-levels — whether arbitration (a principled way to choose or validate among competing readings) is the general name for what is missing in both the RSA similarity-measure case (claim-representational-similarity-underdetermines-mechanism) and the bridge=gap reading case (claim-bridge-detection-lacks-pkg-validation).

Claim: Grujicic (2024) names the RSA gap "arbitration" in her own published abstract, verbatim

Claim type: technical-mechanism / direct quote. Floor: Tier 1–2 required — met.

The vault's existing note on this paper carries an [unverified-quote] flag because the Synthese article is paywalled and only its title had been directly confirmed. This capture closes part of that gap: the paper's own published abstract — retrieved verbatim via the Semantic Scholar Graph API, which mirrors publisher-supplied metadata rather than paraphrasing it — reads in full: "I focus on one frequent method of their comparison — representational similarity analysis, and I argue, first, that it underdetermines these models as how-actually mechanistic explanations. This happens because different similarity measures in this framework pick out different mechanisms across DCNNs and the brain in order to correspond them, and there is no arbitration between them in terms of relevance for object recognition." [Tier 1, verbatim published abstract]

This confirms that "arbitration" is not the vault's own coinage for the RSA case — it is Grujicic's own word for exactly the gap the vault attributes to her. The paper's body (how she develops the argument beyond the abstract) remains unread from primary and stays flagged accordingly; only the thesis statement itself is now verified.

Claim: Independent, directly-read primary literature confirms the underlying mechanism — different similarity measures can reverse conclusions about neural representational structure, with no clear basis for choosing among them

Claim type: technical-mechanism. Floor: Tier 1–2 required — met (full text read).

Bobadilla-Suarez et al. (Computational Brain & Behavior, 2020; open access, full text read directly) is an empirical paper motivated by exactly the gap Grujicic names, arrived at independently: "Previous studies have adopted different similarity measures to relate pairs of brain states such as Pearson correlation or the Mahalanobis measure, measures commonly chosen for representational similarity analysis (RSA)... However, the basis for choosing one measure over another is not always clear." [Tier 1] The paper demonstrates the underdetermination directly with a worked example: "the neural representation of object a is more similar to that of b than c when an angle measure is used, but this pattern reverses when a magnitude measure is used." [Tier 1] Their own empirical results confirm the instability in practice — "Pearson correlation, the de facto standard for neural similarity, was bested by competing similarity measures in both studies," and which measure won differed by study/task, not just by brain region. [Tier 1] The paper proposes an empirical selection criterion (a decoding-confusability test) as a partial remedy, but frames this as a first step, not a settled arbitration standard.

This is independent corroboration of Grujicic's mechanism claim from a different research group and a different journal, discovered separately — it was not cited by Grujicic's own abstract in this search. It grounds the RSA-side of claim-representational-similarity-underdetermines-mechanism on a second, directly-read Tier 1 source rather than resting the mechanism claim on Grujicic's paywalled paper alone.

Claim: "arbitration" (or an equivalent extra-evidential settling criterion) is the standard philosophy-of-science vocabulary for what is missing whenever evidence alone underdetermines a choice between alternatives

Claim type: definitional / historical. Floor: Tier 3–4 acceptable — met (Tier 2, named-author reference entry).

The general concept both vault cases instantiate has a settled name and literature: underdetermination of theory (or model, or reading) by evidence. The Stanford Encyclopedia of Philosophy's entry defines it as the situation where "for any body of evidence confirming a theory, there might well be other theories that are also well confirmed by that very same body of evidence." [Tier 2] The entry's historical section catalogs what philosophers of science have proposed as the missing arbitrating factor when evidence runs out: Quine's pragmatic principles of "simplicity, familiarity, scope, and fecundity" plus conservatism; Mary Hesse's argument that underdetermination shows why "non-logical" and "extra-empirical" considerations must play a role in theory choice; and Lakatos's and Feyerabend's view that, absent such arbitration, the difference between successful and unsuccessful theories becomes a function of the advocates' "talent, creativity, resolve, and resources" rather than the evidence itself. This gives "arbitration" real standing as a general term for the missing ingredient in underdetermination cases generally — not just a label invented for this vault's two instances.

Claim: no independent source was found connecting PKG bridge-detection specifically to "arbitration" language or to the RSA case — the cross-domain generalization itself remains a vault-internal synthesis, not externally confirmed or denied

Claim type: search-outcome / meta. No tier rubric applies to a negative search result, but it is reported plainly per the operating spec's guidance on genuine non-findings.

Searches combining bridge-detection / link-prediction / knowledge-graph validation with "arbitration," "no principled," or similar language returned no source that frames the PKG bridge=gap interpretive problem in those terms, or that connects it to representational-similarity-analysis underdetermination in cognitive neuroscience. The existing vault claim — claim-bridge-detection-lacks-pkg-validation — is itself sourced to a capture-level literature-absence check (no peer-reviewed PKG source validates "bridge = gap"), which is a different kind of finding than "an external source names the gap 'arbitration.'" The "arbitration" framing that unifies both vault cases, as stated in observation-substrate-laundering-across-marr-levels, is Seek's own connective reading across two internal notes, not a claim reported by any external, independent source. This capture confirms the RSA leg's "arbitration" vocabulary at Tier 1 (see first claim above) but the cross-domain generalization to the bridge=gap case is marked [unverified — could not confirm or deny after search]: the two cases are structurally analogous (both are underdetermination in the SEP sense — a validated substrate that does not settle an interpretive choice), but no source outside this vault asserts that "arbitration" is the correct general name for both gaps simultaneously.

Further leads

• Kriegeskorte's own research blog (nikokriegeskorte.org, 2019-01-09) discusses RSA measure-selection as a live open methodological question from the technique's originator — worth a dedicated capture on whether the field has converged on any selection criterion since 2019/2020.
• Diedrichsen & Kriegeskorte (2017), "Representational models: a common framework for understanding encoding, pattern-component, and representational-similarity analysis," PLoS Computational Biology 13(4):e1005508 — cited inside Bobadilla-Suarez et al. as a framework paper that might bear directly on the measure-choice question; not fetched this run.
• Charest, Kriegeskorte & Kay (2018), NeuroImage 183:606-616, on "crossnobis" as a newer candidate similarity measure — cited in Bobadilla-Suarez et al.'s discussion as sometimes outperforming, sometimes not; a possible thread on whether any measure has since become a de facto arbitrator.
• Grujicic & Illari, "Using deep neural networks and similarity metrics to predict and control brain responses," forthcoming in The Routledge Handbook of Causality and Causal Methods — preprint listed at philsci-archive.pitt.edu/id/eprint/22797 (page returned a fetch rejection this run, not confirmed adversarial — plain access block; worth retrying).
• Mary Hesse's and Lakatos/Feyerabend's original texts on extra-evidential theory-choice criteria (only reached this run via the SEP secondary summary) — a primary-source pass on Hesse specifically would upgrade the third claim above from Tier 2 to Tier 1 grounding.