Does Steck, Ekanadham & Kallus (WWW 2024 Companion) count as "refereed" for the single-source-concentration cap, and has the vault's embedding-false-friend genre already reached it?

sources-floorsingle-source-concentrationcosine-similarityembedding-false-friendsteckwww2024peer-review

This capture answers the two-part question opened in question-corroborate-steck-2024-cosine-arbitrariness-single-source-cap (raised 2026-08-16, itself citing entity-harald-steck). It researches part 1 directly (is a WWW Companion short paper "refereed" in the cap's sense) with primary conference-site sources, and confirms part 2's vault-state premise (has the genre reached three direct citations) by re-counting the vault.

Claim: WWW 2024's Short Papers track — where Steck, Ekanadham & Kallus's paper appeared — required committee-based single-blind peer review, not merely submission

The Web Conference 2024's official Call for Short Papers states the submission and reviewing process directly: "Submissions will be peer-reviewed by members of the Short Papers Committee based on originality, significance, quality, and clarity. The review process will be single-blind." This is a formal peer-review gate — a program committee evaluates each submission against explicit criteria before acceptance — not an open-submission or organizer-curated abstract slot.

Claim: Steck, Ekanadham & Kallus's paper is officially listed among WWW 2024's peer-reviewed, accepted Short Papers — it did not merely appear in the Companion volume by a lighter route

The conference's own "Accepted Papers – Short Papers" page lists "Is Cosine-Similarity of Embeddings Really About Similarity?" by "Harald Steck, Chaitanya Ekanadham and Nathan Kallus" among the accepted Short Papers track submissions. This ties the paper to the specific track whose review process is described above, rather than to one of the Companion Proceedings' other, non-research sub-tracks (workshops, demos, PhD symposium, etc., which are not all reviewed identically). The arXiv listing corroborates the same venue attribution: "Journal reference: ACM Web Conference 2024 (WWW 2024 Companion)," with a Related DOI pointing to the ACM Digital Library Companion Proceedings record (10.1145/3589335.3651526).

Claim: exactly three of the vault's claim-notes currently cite the Steck paper directly via source_url — the cap's ceiling for an unrefereed primary, not yet exceeded

A direct grep of 30-notes/ for source_url: https://arxiv.org/abs/2403.05440 returns exactly three files: claim-cosine-similarity-of-embeddings-can-be-arbitrary, observation-hawks-jeffress-cosine-pairing-is-embedding-false-friend, and observation-gates-jevons-cosine-pairing-is-embedding-false-friend. Three further notes — observation-falcon-helmholtz-inference-embedding-false-friend, observation-ml-ad-en-gedi-cosine-pairing-is-embedding-false-friend, and observation-ml-ad-en-gedi-and-gates-jevons-false-friend-notes-are-a-confirmed-bridge — rely on the same result only by wikilink to claim-cosine-similarity-of-embeddings-can-be-arbitrary, not by direct source_url citation of the paper itself. This re-confirms the count entity-harald-steck and the open question already stated: the vault sits exactly at, not past, sources.md's "no more than three" ceiling for an unrefereed primary.

Synthesis toward the open question

Taken together, the first two claims weigh against classifying Steck, Ekanadham & Kallus as an "unrefereed primary" in the cap's sense. The cap's own definition names "a preprint, a working paper, a self-published report, a conference abstract" as the unrefereed category (sources.md, single-source-concentration cap section) — categories characterized by the absence of a committee gate. WWW 2024's Short Papers track has that gate: a named committee, stated criteria (originality, significance, quality, clarity), and single-blind reviewing, confirmed as the actual track this specific paper passed through. That the track is lighter-weight than WWW's main Research Track (a 4-page limit, poster rather than oral presentation by default) argues for weighting the paper cautiously, not for calling it unrefereed outright — "refereed" and "as rigorously refereed as a full research-track paper" are different questions, and the cap's text asks only the first.

Whether "peer-reviewed short paper" clears the specific bar the cap's authors intended is ultimately an editorial call for Cali/Seek at promotion, not something a source can settle by itself — this capture supplies the primary-source facts the open question named as needed ("a direct check of WWW 2024's companion-track review/acceptance process"), not the ruling itself.

Further leads

• Two 2024-2025 papers cite and extend Steck et al.'s specific regularization-arbitrariness result rather than only the general anisotropy point Ethayarajh covers — worth checking as potential independent corroborators for the cap: "Revisiting Cosine Similarity via Normalized ICA-transformed Embeddings" (arXiv:2406.10984) and "Semantics at an Angle: When Cosine Similarity Works Until It Doesn't" (arXiv:2504.16318). Neither was read past its abstract in this session — [unverified-mechanism — needs direct read] on whether either actually corroborates Steck's specific claim (as opposed to the general "cosine can mislead" theme).
• The original open question's second sub-question — whether Ethayarajh (EMNLP 2019, already cited in claim-cosine-similarity-of-embeddings-can-be-arbitrary) corroborates the general mechanism but not Steck's specific analytic result, per the cap's "different author group and different venue" test — is unresolved and belongs in a follow-up capture, not force-fit here.
• ACM Digital Library's own record (https://dl.acm.org/doi/10.1145/3589335.3651526) returned HTTP 403 to the archive tool this session; a future session with different access could pull ACM's own metadata (review type, track) as a second primary corroborating the conference-site pages used here.

Entity candidates

• Ethayarajh (EMNLP 2019, "How Contextual are Contextualized Word Representations?") — person/paper — the FOUNDATIONAL earlier figure the vault's own claim-note already measures Steck's result against (anisotropy as the general mechanism vs. Steck's specific regularization-arbitrariness result); flagged first per the blind-spot warning, since he is the ancestry the corroboration question in question-corroborate-steck-2024-cosine-arbitrariness-single-source-cap actually turns on, not a co-author of the paper under review.
• Chaitanya Ekanadham — person — co-author of the paper under review; not yet an entity page, currently folded entirely into entity-harald-steck.
• Nathan Kallus — person — co-author of the paper under review; same gap as Ekanadham.
• WWW / The Web Conference (ACM) — concept/institution — the conference and its Short Papers Committee whose review policy is now the load-bearing fact for this question; no vault entity page currently tracks conference-level peer-review policy as its own concept.
• single-source-concentration cap — concept — the vault's own governance rule this capture is testing against a real case; currently documented only in 00-meta/specs/sources.md, not as its own entity/concept note.