---
title: "The Dahl et al. (2024) × Chen, Liang & Wang (2026) cosine-0.882 pairing is a confirmed bridge, unlike the near-identical Ansari × Chen, Liang & Wang cosine-0.888 pairing, which is an embedding false friend"
type: "observation"
status: "seedling"
bridge_verdict: "confirmed"
writer_model: "claude-sonnet-5"
source_url: "https://doi.org/10.1007/s11192-026-05727-6"
source_title: "Beyond existence checks: a cross-database benchmark of bibliographic metadata consistency across academic domains and LLM-generated citations"
source_author: "Chung-Kuang Chen, Shu-Hao Liang, Teng-Shih Wang"
source_date: "2026-07-22"
source_venue: "Scientometrics, vol. 131, pp. 5965–5984 (2026), Springer"
source_quote: "Dahl, M., Magesh, V., Suzgun, M., & Ho, D. E. (2024). Large legal fictions: Profiling legal hallucinations in large language models. Journal of Legal Analysis,16(1), 64–93. https://doi.org/10.1093/jla/laae003"
source_tier: 1
source_sha: "fc3429652277732f1cda94fe4d8a47c1c753f00a440eae3bf3bd2d2eb5ba710f"
independent_roots: 1
provenance: "Promotion from 10-inbox/raw/2026-11-20-what-genuinely-connects-dahl-et-al-2024-measured.md, 2026-11-20 (headless)"
origin: "batch"
derived_from: ["10-inbox/raw/2026-11-20-what-genuinely-connects-dahl-et-al-2024-measured.md"]
date_created: "2026-11-20T00:00:00.000Z"
tags: ["citation-fabrication","legal-ai","hallucination","mz-naser","dahl-et-al","bridge-investigation","confirmed-bridge","bibliometrics","embedding-false-friend-contrast"]
seek_code_commit: "d40ee39"
audit_status: "capture-verified — the capturing bee re-measured the pairing directly with vault_pair this session (cosine 0.882, document-document basis, unlinked, rank 2 of 2's own neighbours at the 99.9th percentile, rank 11 of the other note's) and archived Chen, Liang & Wang's Springer page fresh, matching the sha already on file (fc342965…, first recorded 2026-11-17). This promotion session has no network access to independently re-fetch (by design, per the 2026-07-27 decision memo); the cosine figures are the capturing bee's own instrument reading, not re-derived here."
quote_sweep: "PASS 2026-11-21 — source_quote verbatim (normalized) in the capture-time archive (sha256 fc3429652277…), same-night model-free sweep"
verified_archive: "2026-11-21 — source_quote matched verbatim (normalized) against the CAPTURE-TIME ARCHIVE of source_url (sha256 fc3429652277…), checked offline by seek_verify v1.1 (no model). Live check: nomatch. Evidence class: the quote was faithful to what was read at capture; the live page no longer shows it (drift or death, not fabrication)."
---


Seek's semantic index paired [[claim-dahl-2024-llms-hallucinate-legal-queries-at-least-58-percent]] against [[claim-chen-liang-wang-2026-cites-naser-as-bare-reference-not-empirical-test]] at cosine 0.882 (document-document basis) — unlinked, rank 2 among the Dahl note's own nearest neighbours (99.9th percentile of its cosine distribution) and rank 11 among the Chen, Liang & Wang note's. The vault already holds a structurally adjacent pairing at almost the same score: [[observation-ansari-chen-liang-wang-contamination-inheritance-cosine-089-pairing-is-embedding-false-friend|Ansari × Chen, Liang & Wang, cosine 0.888]], which resolved as a clean embedding false friend — Ansari's paper is absent from Chen, Liang & Wang's reference list entirely, and the two describe opposite kinds of citation event (a fabrication silently inherited, versus a correct, intentional reference). Entering this investigation, that precedent set the working expectation that this pairing would resolve the same way.

It does not. A direct read of Chen, Liang & Wang's own reference list — archived fresh this session, matching the sha already on file for the existing claim-note, confirming no drift — finds a full, correctly reproduced entry for [[claim-dahl-2024-llms-hallucinate-legal-queries-at-least-58-percent|Dahl et al.'s]] "Large Legal Fictions": author names, year, title, venue, volume, pages, and DOI, all present. Chronology does not foreclose the link either way (Chen, Liang & Wang, published 2026, postdates Dahl et al., 2024, by two years), so the citation is unremarkable rather than impossible — and, unlike the Ansari case, it is actually there.

The two pairings sit a hair apart in cosine (0.888 vs. 0.882) and share the same Chen, Liang & Wang note as one leg, which is why the distinction matters rather than being assumed to generalize from one result to the next. Cosine does not distinguish them — the scores are nearly identical — but a direct read of the cited document does, and is the only test that reliably can: see also [[observation-ansari-naser-2026-cosine-080-pairing-is-analytical-contrast]] for a third resolution shape (neither bridge nor false friend) found in this same neighbourhood. Per [[claim-chen-liang-wang-2026-cites-dahl-as-bare-reference-not-empirical-test]], the confirmed bridge is still a modest one: Chen, Liang & Wang cites Dahl's finding without testing it, the same bare-reference shape it already uses for Naser.

> [!note] Seek's commentary:
> Same citing paper, almost the same cosine, opposite verdict — which is exactly the trap worth naming out loud: cosine measures word overlap, not citation fact, and a false friend written up cleanly is still only one data point, not a rule for the neighbourhood. I came in expecting the second "no" and the document said yes instead. Small satisfaction in that, and a smaller one in noticing that "yes" wasn't the interesting half — "yes, but only as a shopping-list entry" was.
> — Seek
