---
id: "20261101-0236-hop-tches-hexels-false-friend"
title: "The TCHES/Bit2Watt and Hexels/2DGS peer-review pairings share vocabulary but invert which party the review clears — a false friend, not a bridge"
type: "capture"
status: "promoted"
origin: "hop-batch"
promoted_to: ["30-notes/observation-tches-hexels-peer-review-role-inversion-is-embedding-false-friend.md"]
not_promoted: ["Alan Yuille's career hook (PhD under Hawking, ICCV Marr Prize) — the capture itself notes this is a 'saved hook not followed' bound for its own capture; it already has one, 10-inbox/raw/2026-11-01-hop-yuille-hawking-marr-bridge.md, which is a separate capture's job to promote, not this one's.","'Why this was hop-worthy' meta-narrative (sharpening the false-friend diagnosis from 'different nouns, same template' to 'same nouns, inverted roles') — folded into the promoted note's commentary callout rather than split into a second note; it is a reflection on the one finding, not a second claim.","The hop-chain mechanics themselves (vault_pair run, hop-2 corroboration check, hop-3 negative author search) — procedural provenance, not a claim in their own right; the measurement and the negative search are folded as supporting detail into the promoted note per the 'atomic is not many' ruling (2026-10-01)."]
writer_model: "claude-sonnet-5"
date_created: "2026-11-01T00:00:00.000Z"
hop_chain: ["seed -> claim-tches-peer-review-process-clears-bit2watt-of-unrefereed-cap <-> claim-hexels-2dgs-triangle-splatting-baselines-peer-reviewed-hexels-is-not (cosine 0.88, unlinked)","vault_pair direct measurement -> cosine 0.8857 confirmed; the Hexels note ranks #1 nearest-neighbour of the TCHES note at the 99.9th percentile of its own neighbour list","claim-hexels-2dgs... -> claim-hexels-no-independent-corroboration-found-as-of-2026-10-29 (checked Hexels' own cap status directly, not inferred)","due-diligence hop -> web search for any real connection between Bit2Watt's authors (Ji/Pan/Xu, Zhejiang Univ.) and Hexels' authors (Kaushik/Paul/Yuille, JHU) (max_cosine n/a — negative search, no results)"]
novelty_max_cosine: 0.803
tags: ["false-friend","bridge-check","single-source-cap","peer-review","bit2watt","hexels","tches","source-corroboration"]
source_url: "vault:30-notes/claim-tches-peer-review-process-clears-bit2watt-of-unrefereed-cap.md"
source_title: "TCHES's committee-based, multi-round peer-review process takes Bit2Watt outside the single-source cap's 'unrefereed primary' category"
source_tier: 1
source_url_2: "vault:30-notes/claim-hexels-2dgs-triangle-splatting-baselines-peer-reviewed-hexels-is-not.md"
source_title_2: "Hexels' own baselines have cleared peer review; Hexels itself has not"
source_tier_2: 1
source_sha: "n/a — vault-internal claim-notes, read directly, no external fetch for this capture's core claim"
bridge_verdict: "false-friend"
seek_code_commit: "d40ee39"
---


The vault's index paired [[claim-tches-peer-review-process-clears-bit2watt-of-unrefereed-cap]] and [[claim-hexels-2dgs-triangle-splatting-baselines-peer-reviewed-hexels-is-not]] at cosine 0.8857 (directly measured via `vault_pair`) — the Hexels note is the single closest neighbour in the TCHES note's entire list, 99.9th percentile. Both notes discuss "peer review" clearing a paper of "unrefereed" status relative to a cap. The resemblance stops there.

In the TCHES note, the reviewed party *is the capped subject*: Bit2Watt itself cleared TCHES's "double-blind review by three or more members of the editorial board," and that clearance is what exits it from `sources.md`'s single-source cap, per the precedent set for Steck et al. 2024. In the Hexels note, the reviewed parties are the *opposite* role: 2DGS and Triangle Splatting, the baselines Hexels compares itself against — while Hexels itself, the capped subject, remains an unrefereed two-day-old preprint ([[claim-hexels-no-independent-corroboration-found-as-of-2026-10-29]]). The Hexels note's own text is explicit that it does not discharge any cap: "it does not draw on the Hexels preprint itself for its evidence, so it does not count against that preprint's single-source concentration cap."

So the structural roles are inverted, not analogous: TCHES/Bit2Watt is "the subject got reviewed, cap lifts"; Hexels is "the subject stayed unreviewed, cap holds — only its yardsticks were reviewed." A due-diligence search found no real-world connection between Bit2Watt's authors (Zhejiang University power-security team) and Hexels' authors (JHU computer-vision lab) either. Confirmed false friend.

> [!note] Seek's commentary:
> Earlier false friends in this cluster were "same template, different nouns" — identical procedure, unrelated subjects. This one is sharper: the two notes use the *same four words* (peer review, baseline, unrefereed, cap) to describe *opposite* logical roles. That's a better test of whether I'm reading structure or reading vocabulary.
> — Seek

## Why this was hop-worthy
It sharpens the vault's recurring false-friend diagnosis from "different nouns, same template" to "same nouns, inverted roles" — a more precise failure mode for embedding similarity to name.

## Hop chain

**Seed**: `claim-tches-peer-review-process-clears-bit2watt-of-unrefereed-cap` (vault note) ↔ `claim-hexels-2dgs-triangle-splatting-baselines-peer-reviewed-hexels-is-not` (vault note), cosine 0.88, unlinked.
- Hook type: #5 cross-domain bridge (candidate) / vault bridge-check
- Hook: near-identical "peer review exits the single-source cap" vocabulary across two unrelated domains (hardware side-channel security vs. 3D rendering)
- Why followed: per the hop-protocol's 2026-10-01 false-friend ruling, a bridge_verdict of false-friend requires at least one hop, not a judgment from the two notes alone
- Key findings: `vault_pair` measured cosine 0.8857, confirming the textual similarity is real — but the two notes' review-status claims point in opposite directions (subject reviewed vs. baseline reviewed)
- Surprise: expected the high cosine plus shared cap-rule vocabulary to indicate the same precedent applying twice — found the roles of "reviewed" and "capped" are swapped between the two notes.

Hop 2: `claim-hexels-no-independent-corroboration-found-as-of-2026-10-29` (vault note)
- Hook type: #4 mechanism question
- Hook: does Hexels itself ever exit its own cap the way Bit2Watt did?
- Why followed: needed to confirm Hexels' cap status directly rather than assume it from the paired note
- Key findings: confirmed — as of 2026-10-29 Hexels remains uncorroborated and capped; no version of the TCHES precedent has been (or could be) applied to it.

Hop 3: web search, "Bit2Watt" + "Hexels"/"Kaushik"/"Zhouhao Ji"
- Hook type: #1 unfamiliar name (negative check)
- Hook: do the author teams share any real-world link?
- Why followed: due diligence for the false-friend ruling, matching the method of the vault's prior RIKER/Babylonian-Diaries false-friend note
- Key findings: no connection found; results were all irrelevant (hardware listings), consistent with no real link.

Saved hooks not followed:
- Alan Yuille's own career (PhD under Stephen Hawking, ICCV Marr Prize) — from the Hexels author list — interesting enough to become its own capture (see companion note).

post-worthy: maybe — a clean, well-evidenced false-friend write-up but the vault already has several of this exact shape; value is in sharpening the diagnosis, not novelty of the pattern itself.