Point the arXiv kit at v1.2.0, and make the release gate say what it does

papers/peer-review/ARCHIVAL-INTEGRITY.md

@@ -128,6 +128,12 @@ the system it describes. The correct reading is that **creating a GitHub release
 here IS publishing a DOI**, and the approval gate must move earlier — to the
 release, or to disabling the integration — because there is no later gate.
 
+**Owner decision, 18 August 2026: auto-publication stays on, and the release is
+the gate.** The documentation is corrected to match the mechanism rather than
+the mechanism changed to match the documentation. Consequence, stated so no
+future reader re-derives it: every check must pass *before* `gh release create`,
+because nothing after it can be undone.
+
 The deposit itself is correct: right version, right relation to v1.1.0, open
 access. Nothing needs retracting. A Zenodo record cannot be unpublished in any
 case, which is why the gate mattered.

papers/peer-review/SUBMISSION-CHECKLIST.md

@@ -25,7 +25,7 @@ The focused preprint is publicly archived. The record below separates completed
 - [x] Prepare the arXiv upload artifact, form metadata, category recommendation, and endorsement instructions in `papers/peer-review/arxiv/`.
 - [x] ~~Publish the DOI-bearing PDF as a metadata-only `v1.1.1` archival version~~ — superseded by `v1.2.0`, a content revision adding the CE-14 scope limitation. See `ARCHIVAL-INTEGRITY.md`.
 - [x] Create a Zenodo version for `v1.2.0` and record its version DOI `10.5281/zenodo.21997434`. **Published automatically by the GitHub integration on release creation — there was no draft and no approval step.** See the process defect recorded in `ARCHIVAL-INTEGRITY.md`.
-- [ ] Decide where the approval gate now lives, given that creating a GitHub release here publishes a DOI immediately: gate the release itself, or disable auto-publication in the Zenodo integration.
+- [x] Decide where the approval gate lives. **Owner decision, 18 August 2026: auto-publication stays on.** Creating a GitHub release for this repository publishes a Zenodo DOI immediately and irreversibly. The gate is therefore the **release itself**, and there is no later one. Any wording implying a reviewable Zenodo draft has been corrected.
 - [x] ~~Decide the open item: the tracked `v1.1.0` PDF is not byte-identical to the deposited one.~~ Resolved: both v1.1.0 artifacts restored to the deposited bytes, verified by a byte-identical rebuild.
 - [ ] Submit the single PDF to arXiv from the author's own account.
 - [ ] Record venue, submission identifier, date, and manuscript status without describing the preprint as peer reviewed.
@@ -40,4 +40,4 @@ The focused preprint is publicly archived. The record below separates completed
 
 ## Blocking conditions
 
-Do not publish a release if the PDF differs from the reviewed source, repository checks fail, citation metadata disagrees with the title/author/version, or a DOI/status is represented as assigned before the external archive assigns it.
+Do not publish a release if the PDF differs from the reviewed source, repository checks fail, citation metadata disagrees with the title/author/version, or a DOI/status is represented as assigned before the external archive assigns it. **Creating the release IS the irreversible step**: Zenodo publishes automatically and a published record cannot be withdrawn. Everything to be checked must be checked before `gh release create`, not after.

papers/peer-review/arxiv/README.md

@@ -6,15 +6,15 @@ This directory prepares the author-controlled arXiv submission without pretendin
 
 Upload exactly one file:
 
-`output/pdf/minority-prophet-peer-review-v1.1.0.pdf`
+`output/pdf/minority-prophet-peer-review-v1.2.0.pdf`
 
 The manuscript is authored in Markdown and rendered directly by ReportLab; it is not generated from TeX. arXiv permits a single machine-readable PDF in this case. Do not upload this directory or combine the PDF with these metadata files.
 
 Before upload, run:
 
 ```text
 make paper-check
-pdffonts output/pdf/minority-prophet-peer-review-v1.1.0.pdf
+pdffonts output/pdf/minority-prophet-peer-review-v1.2.0.pdf
 ```
 
 Confirm that the title, author, abstract, page count, DOI, and rendered pages match the repository record.
@@ -27,6 +27,26 @@ Confirm that the title, author, abstract, page count, DOI, and rendered pages ma
 
 The author chooses the category in the arXiv form, and moderators may reclassify it.
 
+## What endorsement actually requires
+
+Checked against `info.arxiv.org/help/endorsement.html` rather than assumed.
+
+- Endorsement is per **endorsement domain**, not per paper. For this submission
+  the domain is the one containing `cs.LO`.
+- The author sends an endorser a **six-character alphanumeric code** that arXiv
+  issues; the endorser enters it on arXiv's endorsement form.
+- An eligible endorser must have submitted a paper in that domain **between
+  three months and five years ago**, and must themselves hold a positive
+  endorsement for it.
+- Endorsement **is not peer review**. The endorser checks that the paper is
+  appropriate for the subject area, not that it is correct.
+- One positive endorsement is sufficient.
+
+**Unverified and worth planning around:** arXiv auto-endorses submitters whose
+registered email is at a recognised academic institution. A personal-domain
+address is therefore likely to trigger the endorsement requirement rather than
+bypass it. Confirm at registration; do not assume an account alone is enough.
+
 ## Author-controlled steps
 
 1. Register or sign in at arXiv and start a new submission.
@@ -46,7 +66,7 @@ Hello [Name],
 
 I am preparing an arXiv submission titled "The Minority Prophet Property: Copy-Invariant Evidence Aggregation in Rooted Claim Graphs." It presents Lean-checked copy-invariance and root-margin results for aggregation over declared provenance graphs, with explicit limits around truth, independence, and root qualification.
 
-arXiv has asked me for endorsement in [category/domain]. If the paper is appropriate for that area and you are eligible and comfortable endorsing the submission, the arXiv request link is [link]. The manuscript and archival record are available at https://doi.org/10.5281/zenodo.21965713.
+arXiv has asked me for endorsement in [category/domain]. If the paper is appropriate for that area and you are eligible and comfortable endorsing the submission, the arXiv request link is [link]. The manuscript and archival record are available at https://doi.org/10.5281/zenodo.21965712.
 
 Thank you for considering it,
 James Siyuan He

papers/peer-review/arxiv/metadata.json

@@ -4,8 +4,8 @@
     "James Siyuan He"
   ],
   "abstract": "When claims can be copied at negligible cost, a large apparent majority may contain only one underlying observation. This paper defines a root-counting aggregator over finite binary claim graphs and proves that, under side consistency, its verdict is invariant to lineage changes preserving assertions and the root set and to copies with recorded parents. Tight root-margin bounds expose decision sensitivity. The Lean-checked results concern correct counting of declared provenance, not truth, authenticity, causal independence, discovery completeness, or authority.",
-  "comments": "Preprint v1.1.0; not peer reviewed. Lean 4 proofs and reproducibility artifacts are available from the linked repository. Archival DOI: 10.5281/zenodo.21965713.",
-  "doi": "10.5281/zenodo.21965713",
+  "comments": "Preprint v1.2.0; not peer reviewed. Lean 4 proofs and reproducibility artifacts are available from the linked repository. Archival DOI: 10.5281/zenodo.21997434 (all versions: 10.5281/zenodo.21965712).",
+  "doi": "10.5281/zenodo.21997434",
   "primary_category_recommendation": "cs.LO",
   "cross_list_recommendations": [
     "cs.AI"