PUBLISHED CORRECTIONS00No published article has required a material correction yet.
PREPUBLICATION HOLDS01A passing candidate was withheld before it could affect readers.
LEDGER VERSION1.0Last updated 2026-08-31. JSON is public and versioned.
THE POLICY
Correct the public claim, preserve the original run, and publish enough evidence to reconstruct what changed and why.
Download the ledger ↗SEVERITY CONTRACT
Different mistakes.
Different obligations.
Severity follows reader impact, not embarrassment. A central result receives a different response from a harmless typo.
01 / materialChanges the central result, decision guidance, population, uncertainty, source rights, or review status.
Freeze promotion or add an article-level correction, rerun affected gates, preserve the superseded run, and include the correction in the next available reader update.
02 / clarificationImproves scope or interpretation without changing the reported result.
Update the article with a dated note and record the changed language in this ledger.
03 / minorTypography or presentation that cannot alter the reader's interpretation.
Correct in place; ledger inclusion is optional unless the error recurs or affects an artifact.
CORRECTION WORKFLOW
The original run does not disappear.
Every material correction or hold must connect the public statement to the evidence, decision, and replacement state.
- Triage the claim and identify every affected public surface.
- Freeze publication or promotion when reader guidance may be wrong.
- Preserve the original run, hashes, reviews, and decision artifacts.
- Recompute and repeat the relevant specialist and mechanical checks.
- Publish a dated correction or withholding note with evidence links.
- Update the machine-readable ledger and article disclosure together.
PUBLIC LEDGER / 01
What changed
and why.
Prepublication holds appear here too. Catching a mistake before release is part of the operating record, not a reason to hide it.
COR-001prepublication holdmaterialresolved by withholding
Touchdown Regression 2025 candidate
Visual QA found prior-sample values in two otherwise current chart titles after the research gate passed.
Reader impact: None. The candidate was never promoted; the published 2024 launch article remained unchanged.
Resolution: The run was preserved and marked not published. Future figures are generated from a title/metric/hash contract independently checked by both the research gate and promotion command.