What is wrong
docs/kit-friction-log.md's ## 2026-09-03 — Backlog migrated to GitHub Issues (#671–#672) section says:
The 2026-09-01 entry and its recurrence are already recorded on
#393
The tracker does not support that. gh api repos/topij/agentic-dev-kit/issues/393/comments
on 2026-09-06 returned comments dated 2026-08-29, 2026-09-02 and 2026-09-02. At
2026-09-03, when the marker was written, only the 2026-08-29 comment existed, and it
concerns a different occurrence. The 2026-09-02 recurrence comment does reference the
2026-09-01 entry, but only as the hypothesis that recurrence did not survive; the
2026-09-01 entry's own revision and run have never been recorded on the tracker.
Why it matters now
The 2026-09-06 sweep corrected the same claim where it recurred in its own marker, so
the file now contradicts itself a few sections apart. A reader who trusts the older
section still walks away with the overstatement the newer one exists to correct.
Mechanism
A graduation marker asserted tracker state without verifying it at write time, and the
file's convention treats dated sections below the newest marker as an immutable record
of what was believed and decided then. That convention is worth keeping — it is what
makes the log a history rather than a rolling summary — but it also preserves an
assertion that was false when made, with nothing marking it as such.
Proposed direction
Decide the convention explicitly. Either a marker block must verify any tracker
assertion at write time and a later-discovered error gets an appended correction rather
than a rewrite, or the file states plainly that dated sections record what was believed
at the time rather than verified fact — so a reader knows which claims carry weight.
Severity L. Found by the fallback review panel's correctness lens on PR #694, which
checked the claim against the live tracker API rather than reading it from the diff.
What is wrong
docs/kit-friction-log.md's## 2026-09-03 — Backlog migrated to GitHub Issues (#671–#672)section says:The tracker does not support that.
gh api repos/topij/agentic-dev-kit/issues/393/commentson 2026-09-06 returned comments dated 2026-08-29, 2026-09-02 and 2026-09-02. At
2026-09-03, when the marker was written, only the 2026-08-29 comment existed, and it
concerns a different occurrence. The 2026-09-02 recurrence comment does reference the
2026-09-01 entry, but only as the hypothesis that recurrence did not survive; the
2026-09-01 entry's own revision and run have never been recorded on the tracker.
Why it matters now
The 2026-09-06 sweep corrected the same claim where it recurred in its own marker, so
the file now contradicts itself a few sections apart. A reader who trusts the older
section still walks away with the overstatement the newer one exists to correct.
Mechanism
A graduation marker asserted tracker state without verifying it at write time, and the
file's convention treats dated sections below the newest marker as an immutable record
of what was believed and decided then. That convention is worth keeping — it is what
makes the log a history rather than a rolling summary — but it also preserves an
assertion that was false when made, with nothing marking it as such.
Proposed direction
Decide the convention explicitly. Either a marker block must verify any tracker
assertion at write time and a later-discovered error gets an appended correction rather
than a rewrite, or the file states plainly that dated sections record what was believed
at the time rather than verified fact — so a reader knows which claims carry weight.
Severity L. Found by the fallback review panel's correctness lens on PR #694, which
checked the claim against the live tracker API rather than reading it from the diff.