Problem
The published OCR duplicate analysis attributes Luther's post-PR-130 reduction in repeated comments to a successful-path same-head prefilter. Current Luther source does not execute that lookup before a successful batch:
- the lookup helper is defined in
.github/workflows/ocr-pr-review.yml;
- the normal path posts the raw inline candidate list;
- the prior-comment lookup is invoked only after batch failure, before fallback posting.
Code and Jefe do run the prior-comment lookup before batch posting. Luther PR 130 primarily added provider concurrency, fail-closed redaction, safer notification behavior, and tracking-issue convergence. The observed reduction in long-interval reposts may be real, but the study cannot presently attribute it to a successful-path prefilter.
The configuration/workflow timeline referenced during the analysis is also absent from the public package, preventing reproducible intervention attribution.
Desired correction
- Publish an immutable intervention timeline with repository, workflow path, commit, PR, effective timestamp, acquisition SHA, and exact control behavior.
- Distinguish per-PR cancellation, successful-path prior-comment filtering, fallback-only filtering, sticky-summary deduplication, and infrastructure tracking-issue deduplication.
- Re-run or relabel the Luther before/after mechanism analysis.
- Remove causal wording unsupported by the source while preserving the observed metrics and correction history.
- Add direct source/commit links for every attributed workflow intervention.
- Preserve the small post-change cohort and mutable-comment limitations prominently.
- Rebuild the Markdown, PDF, evidence ZIP, inventories, and checksums after the correction.
Non-goals
- This issue does not claim that the observed reduction from 18 long-interval reposts to zero is false.
- It does not change repository workflows or rank providers.
- It should not silently overwrite the original result without a documented correction note.
Related implementation work
Problem
The published OCR duplicate analysis attributes Luther's post-PR-130 reduction in repeated comments to a successful-path same-head prefilter. Current Luther source does not execute that lookup before a successful batch:
.github/workflows/ocr-pr-review.yml;Code and Jefe do run the prior-comment lookup before batch posting. Luther PR 130 primarily added provider concurrency, fail-closed redaction, safer notification behavior, and tracking-issue convergence. The observed reduction in long-interval reposts may be real, but the study cannot presently attribute it to a successful-path prefilter.
The configuration/workflow timeline referenced during the analysis is also absent from the public package, preventing reproducible intervention attribution.
Desired correction
Non-goals
Related implementation work