You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 60389a0
Browse filesBrowse the repository at this point in the historyBrowse files
Layer a parse-time skip under the content-based keys (GH-36) — the
approach ccusage took for the same bug (ccusage/ccusage#950) — with a
broader trigger: a second session_meta marks a fork-shaped rollout
(catches Desktop thread branches that carry no thread_spawn marker),
and event lines stamped in the same second as that meta are the replay
burst.
- Replayed history now contributes nothing even when the parent rollout
is outside the scan (deleted or past the mtime window), where
cross-file dedup has no original to collapse into and previously the
replay was counted once at the fork instant.
- Durations (unkeyed) and session touches from replayed lines stop
counting, closing two known remainders.
- Content-based keys stay as the safety net for burst lines that leak
past the fork second, whole-file copies, and re-emissions.
Real-data check: same-instant A/B with and without the skip yields
byte-identical Codex totals on an all-parents-in-scan corpus — the skip
only changes outcomes where originals are missing from the scan.
Co-Authored-By: Claude Fable 5 <[email protected]>
Copy file name to clipboardExpand all lines: CHANGELOG.md
+6-4Lines changed: 6 additions & 4 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -17,10 +17,12 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
17
17
(usage vectors + running cumulative; `turn_id` for effort), re-emitted
18
18
`token_count` events whose cumulative did not advance collapse too, and
19
19
keyed duplicates keep the earliest observed timestamp, so scan order
20
-
can't decide attribution when both the original and the replay are
21
-
scanned. Known remainder, scoped as follow-up: session touches and task
22
-
durations from replayed lines are still counted, and a replay whose
23
-
original falls outside the scan window keeps the fork-instant timestamp.
0 commit comments