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 cd539a8
Browse filesBrowse the repository at this point in the historyBrowse files
fix(learn): compaction-safe turn starts and exposure-scoped ambiguity
- A request right after a compaction marker may be its replay, so its turn
start is unknown; compaction markers are never request boundaries.
- Extract `turnMessages` (synthetic and compaction flags from real parts) and
test it.
- Ambiguity counts only lessons the session was shown in their current text;
a later lesson created from the correction no longer hides recurrence.
- Mark a correction message handled only after its signal is recorded, so a
failed write is retried by a later update.
- Document recurrence in `learn status` and the reflection `r:N` hint.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Copy file name to clipboardExpand all lines: docs/docs/configure/learn.md
+15-1Lines changed: 15 additions & 1 deletion
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -222,7 +222,21 @@ altimate-code learn disable
222
222
223
223
### status
224
224
225
-
Shows whether learning is on, whether automatic promotion is on with its last promotion and last held-back reason, lesson counts (approved, candidate, retired), open signals, pending recoveries, the last reflection and its summary, and the resolved limits. `--json` also includes the last reflection's token use and estimated cost.
225
+
Shows whether learning is on, whether automatic promotion is on with its last promotion and last held-back reason, lesson counts (approved, candidate, retired), open signals, pending recoveries, the last reflection and its summary, the resolved limits, and lesson recurrence. `--json` also includes the last reflection's token use and estimated cost.
226
+
227
+
**Recurrence** answers "was this lesson shown, and did the user still correct the same thing?". It reads the newest 200 session snapshots and the recorded corrections. It makes no model calls and writes nothing. For each shown lesson it counts sessions where a matching correction came:
228
+
229
+
| Count | Meaning |
230
+
|---|---|
231
+
|`recurred`| after the lesson was shown, at the start of the corrected request or earlier |
232
+
|`during`| while the agent was answering, for example a file-hook lesson added mid-turn; the lesson may have arrived too late |
233
+
|`unknown`| in an order that cannot be told, for example older corrections or a request replayed after compaction |
234
+
|`ambiguous`| when the correction matched more than one lesson the session was shown |
235
+
|`earlier_version`| to a different version of the lesson text; it never counts for the current text |
236
+
237
+
A correction matches a lesson when it names one of the lesson's backtick identifiers plus another word from the lesson, or contains most of the lesson's words. The text output lists only lessons with a matching correction and summarizes the rest as "no observed correction". That is not proof the agent followed them. Corrections are recorded only while capture is on. Status also lists up to 5 identifiers that were corrected in 2 or more sessions with no matching current lesson. They are taken from your local corrections and never leave the project. If the analysis fails, status shows `Recurrence unavailable` (`recurrence_error` in JSON), and the rest of the status still prints.
238
+
239
+
When the current text of a lesson has `recurred` in 2 or more sessions, reflection marks it `r:N` in the playbook it sends the model. The model is asked to make the lesson clearer or more specific, never weaker. The result is a candidate like any other and goes through the same review or automatic-promotion gates.
0 commit comments