Skip to content

fix(ci): track the paper count the published version actually declares - #1584

Open
gHashTag wants to merge 1 commit into
masterfrom
fix/stale-paper-count
Open

fix(ci): track the paper count the published version actually declares#1584
gHashTag wants to merge 1 commit into
masterfrom
fix/stale-paper-count

Conversation

@gHashTag

Copy link
Copy Markdown
Owner

tools/check_catalog_count.py prints a warning on every invocation:

WARN: SSOT (83) != paper count (84). An erratum to arXiv:2606.09686 is required
      (see ERRATA_2026-06-14.md). Canonical live count is 83.

The constant behind it tracks the paper's declared count:

# The count claimed in arXiv:2606.09686 Table 1 abstract ("exactly 84").
PAPER_DECLARED_COUNT = 84

That was accurate for v1. It is not accurate now. Fetching the current entry:

$ curl -sS "https://export.arxiv.org/api/query?id_list=2606.09686"

TITLE    : An 83-Format Numeric Catalog with Bit-Exact Conformance Vectors:
           A Vendor-Neutral Reference for FP8, BF16, MXFP4, and Microscaling Formats
ID       : http://arxiv.org/abs/2606.09686v2
UPDATED  : 2026-06-22T12:28:45Z
abstract : "...a catalog of 83 numeric formats spanning 13 families..."

The v2 replacement corrected both the title and the abstract. The erratum the gate demands is the one ERRATA_2026-06-14.md recorded and the v2 replacement carried out. The gate has been asking for it ever since.

Why this is not the edit the comment forbids

The comment says:

Do NOT silently edit this to match SSOT — the whole point is to surface the divergence.

Agreed, and that is not what this does. The constant tracks what the paper declares, and the paper changed. Editing it to match the SSOT would defeat the gate; editing it to match the current published title and abstract is what keeps the gate doing its job. Nothing is silent about it — the evidence, the fetch command and the date are in the comment now, so the next reader can re-check in one line.

The alarm still works

Verified by mutation, in the spirit of wp18_selftest_gate.py. With the constant set to a divergent value:

WARN: SSOT (83) != paper count (99). An erratum to arXiv:2606.09686 is required
strict-mode exit with divergence: 3

And with the corrected value:

SSOT   (// CATALOG: lines)      = 83
regen  (codegen fresh)          = 83
paper  (arXiv:2606.09686 declared)    = 83
OK: SSOT == fresh regen == 83 (canonical).

A gate that cries wolf on every run is one nobody reads on the day the divergence is real. This silences the false alarm without touching the alarm.

Closes #1583

check_catalog_count.py printed, on every invocation, that the SSOT's 83 disagreed
with a paper count of 84 and that an erratum to arXiv:2606.09686 was required. The
erratum in question is ERRATA_2026-06-14.md, and the v2 replacement carried it out.

Fetching the current entry shows both the title and the abstract were corrected:

  TITLE   : An 83-Format Numeric Catalog with Bit-Exact Conformance Vectors: ...
  ID      : http://arxiv.org/abs/2606.09686v2
  UPDATED : 2026-06-22T12:28:45Z
  abstract: "a catalog of 83 numeric formats spanning 13 families"

The constant tracked the withdrawn v1.

The existing comment warns against silently editing this to match the SSOT, and
that warning is right -- but this is not that edit. The constant tracks what the
PAPER declares, and the paper changed. Matching it to the SSOT would defeat the
gate; matching it to the current published title and abstract is what keeps the
gate doing its job. The comment now carries the fetch command, the version and the
date, so the next reader can re-check in one line rather than trusting this one.

The alarm is intact, verified by mutation in the spirit of wp18_selftest_gate.py:
with a divergent constant the gate still warns and still exits 3 under
--strict-paper. With the corrected value it reports SSOT == fresh regen == paper ==
83 and exits 0.

A gate that cries wolf every run is one nobody reads on the day the divergence is
real. This silences the false alarm without touching the alarm.

Closes #1583
@github-actions

Copy link
Copy Markdown
Contributor

📓 NotebookLM Notebook linked to this PR

This notebook contains session context, decisions, and artifacts for this work.

@github-actions

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-07-31 17:51:25 UTC

Summary

Status Count
Total Open PRs 50
PRs with Failing Checks 41
PRs with All Checks Green 9
READY 1
FAILING 41
PENDING 0

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=d6b51a3bcc20 != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

check_catalog_count.py warns on every run about an erratum v2 already made

1 participant