Paper Integrity Record · LEDGER

Teacher-Guided Policy Optimization for On-Policy Reasoning Distillation under Large Policy Divergence

As of 26 May 2026, Pith completed 4 of 4 listed checks against arXiv:2605.13230. The completed checks produced 1 public finding within their stated scope.

This is a record of named checks, not a clean-status claim or a paper verdict.

pith.integrity.v1
2605.13230 v2
pith:2026:VPZTCO4J4EG4HSUEEIFR2J7HZT

Coverage vector

completed 4 of 4 listed checks

Counted as completed checks in the claim sentence numerator.

Listed checks

ai_meta_artifact

completed

v1.0.0, observed 2026-05-26 10:42:14.942882+00:00

Scope: Paper body text scanned for literal AI-assistant artifacts.
Completed surface: Checked 1 reference.

doi_title_agreement

completed

v1.0.0, observed 2026-05-20 09:31:23.733560+00:00

Scope: Cited titles compared with titles resolved from their DOI or arXiv identifier.
Completed surface: Checked 20 references.

doi_compliance

completed

v1.0.0, observed 2026-05-20 07:36:45.383378+00:00

Scope: DOI and arXiv identifiers listed in the bibliography checked against named registries.
Completed surface: Checked 16 references. Recorded 1 finding from this run.

claim_evidence

completed

v1.0.0, observed 2026-05-19 20:01:57.039285+00:00

Scope: Recorded paper claims checked for the evidence artifacts they name.
Completed surface: Completed within the stated scope.

Observations

advisory doi_compliance · recoverable_identifier

DOI in the printed bibliography is fragmented by whitespace or line breaks. A longer candidate (10.64434/tml.20251026.https://thinkingmachines.ai/blog/on-policy-distillation.Rein) was visible in the surrounding text but could not be confirmed against doi.org as printed.

ref #13 · 2026-05-19 06:43:48.367443+00:00

Finding detail Try DOI

Methods and limits

Each listed check names the surface it examined. A skipped, failed, partial, unavailable, not collected, not requested, or withheld check says nothing about what a completed check would have returned.

Only checks with status completed enter the claim sentence numerator.
Findings appear only from completed checks and only within the scope printed for that check.
Severity labels describe one observation row. They do not roll up into a paper judgment.