# Claims review — 0.2.48 - base: v0.2.47 - head: 1375041 - surface: sha256:46439a314e3f75c085c3df6e910adf0ac4f8d1dce42c0c2a9b9f9f33600dd382 - verdict: pass - reviewer: Claude. The only surface that changed is the one that joined: `paper/two-ledgers-one-window.md` (229 lines), read in full against the question the gate asks. Its claim-to-source table (`paper/CLAIMS.md`, 66 rows, 61 carrying `file:line`) was re-checked at the R2 gate on 23 August 2026; this review re-read the prose, not the table. The number moved for one file: `docs/claims/surfaces.json` ships in the tarball and now lists the preprint. `paper/` itself is outside the tarball. No tool changed. ## Findings ### 1. paper/two-ledgers-one-window.md:221 — the package version was pinned to the wrong mechanism - Claim: "version `0.2.46` (declared, from the repository `package.json` at the commit that carries this preprint)." - Mechanism: `package.json` at the carrying commit. - Why it falls short: at the carrying commit `package.json` says `0.2.48`. The sentence would have been false the moment the branch merged. `0.2.46` is right as the release the preprint describes and the archived snapshot covers; the provenance given for it was wrong. - Disposition: reworded in `c555243` — "the release this preprint describes and the snapshot below covers; the repository at the commit that carries this preprint names a later number". `CLAIMS.md` row updated to the npm registry as the mechanism. ### 2. paper/two-ledgers-one-window.md:222 — the draft revision named was superseded on the day of merge - Claim: "Individual Internet-Draft `draft-dogru-scitt-disclosure-evidence-06`". - Mechanism: the datatracker. - Why it falls short: `-07` was posted on 23 August 2026 (archive copy HTTP 200). Not false — `-06` exists — but a reader following the reference lands on `-07`, and the bib entry's note said the datatracker fetch had timed out. - Disposition: `-07` in the markdown, the tex, the bib key and note, and `CLAIMS.md`, in `c555243`. The bib URL now points at the archive copy that was fetched. ### 3. paper/BUILD.md — "the PDF was not compiled here" was about to become a stale denial - Claim: "The PDF was **not compiled** here. Do not read `paper/arxiv/main.tex` as a file that has been shown to build." - Mechanism: a build. - Why it falls short: true until today; a denial left standing after the fact is the pattern `standards/README.md` carried for two days after -04 posted. The tex was compiled on 23 August 2026 in a TeX Live container (pdflatex, bibtex, pdflatex x2): exit 0, 15 pages, 0 undefined references or citations. - Disposition: `BUILD.md` records the build and the command in `c555243`. `paper/build/` stays ignored. ### 4. paper/two-ledgers-one-window.md:44 — the proposition pointed the reader the wrong way - Claim: "Under assumptions A1–A7 below" — in the statement of Proposition 1. - Mechanism: the assumptions section. - Why it falls short: A1–A7 are stated in Section 3, above the proposition. "Below" sends a reader looking for text that is not there. Found on the pre-submission read of the compiled PDF. - Disposition: "above", in the markdown and the tex. The same pass removed the ORCID placeholder from the title block, turned the bibliography's working notes ("opened", "confirmed on the abs page") into access dates, and removed every line that overflowed the margin (seven in the first build, none now). ## Second pass — the surface moved after the record was written The record above was written at `1c92b89`. Four commits then touched the surface (`1084309`, `2a2b63e`, `edf0bf7`, `b187859`) and the gate refused the release, correctly: the review described a text that was no longer the one about to ship. The head and hash above are the re-read. Nothing was published in between. ### 5. standards/ADOPTION-EVIDENCE.md:20 — a list of figures was written as a ratio - Claim: "at commit `a209864e`: 45 passed, 0 failed, 1 skipped over 75 cases." - Mechanism: the Vaara conformance page, row #3. - Why it falls short: the page writes the four figures as a list — "45 passed, 0 failed, 1 skipped, 75 cases" — and the three outcome counts sum to 46, not 75. "over 75 cases" invites a 45/75 reading the page does not state, and the reading runs in our favour on a page another project maintains. Same class as the prior-art row that was scored twice: the source's own units were replaced with a stronger shape. - Disposition: the page's phrasing restored, with the units named, in `1375041`. ### Read and not a finding - **`paper` sections renumbered** (unnumbered abstract, 1–10). The one internal cross-reference moved with them: "the artefacts of Section 9" resolves to `## 9. Artefacts`. No stale pointer left. - **"two propositions" → "one conditional impossibility result and two normative principles"**, and 2a/2b relabelled Principles throughout. The change narrows what the paper claims: 2a and 2b are MUST-shaped rules, and calling them propositions asserted a standing no proof in the paper carries. - **A5 and the proof sketch now hold `G`'s internal randomness and state equal across the two worlds.** Without it the construction leaks: a randomised signature would differ between worlds that agree on every row. The assumption is now stated rather than used silently, which is a narrowing, not a claim. - **"a conditional impossibility result under seven explicit assumptions, not an unconditional completeness theorem"** replaces "a proposition, not a theorem". Considered as a possible upgrade of standing, since the argument below it is labelled a proof sketch. Not recorded as a finding: every step of the sketch cites the assumption it uses, the label carries "conditional" and names the assumption count, and the sentence exists to deny the stronger reading rather than to invite it.