Reduce the arXiv kit to a record, and unwire it from CI

[Silentpartnercoding]

The kit had turned into a tutorial for using arXiv — registration steps, endorsement mechanics, an email template.

None of that is a repository artifact. It is one person’s task list: not testable, read by nobody but the author, and it went stale twice in one day while living here. That is the argument against keeping it.

What remains

A record: which artifact is submitted, why a single PDF, the category choice and its rationale, why submission is pending, and where the identifier goes once it exists.

"Why has this not been submitted" is a fact about the record. "How do I fill in the form" is not.

Deleted: arxiv/metadata.json

It duplicated the title, abstract, and author metadata already canonical in papers/peer-review/metadata.json — and check_peer_review_package.py enforced the two copies against each other.

That is duplication policed by a test rather than removed. The checker no longer reads it, so the suite stops policing a scratchpad.

Kept, because it is genuinely provenance

• The endorsement reason. arXiv’s 21 Jan 2026 policy update closed the institutional-email path for authors with no prior arXiv paper. That is why this is unsubmitted, and it belongs in the record.
• The affiliation note. The manuscript asserts no institutional affiliation and should not acquire one at submission unless it is real.

No behaviour changes, no claim changes. paper-check passes; 426 passed, 8 skipped.

🤖 Generated with Claude Code