Garry's Opinionated OpenClaw/Hermes Agent Brain
TypeScript
test: log the v2 pilots and the Phase 7 restore before the counted v2 run
garrytanb0071b03aa2e
diff --git a/test/fixtures/skill-evals/README.md b/test/fixtures/skill-evals/README.mdindex d20a5b3b90..b6fd7f1f49 100644--- a/test/fixtures/skill-evals/README.md+++ b/test/fixtures/skill-evals/README.md@@ -61,8 +61,11 @@ EVAL_ROOT=<scratch> GBRAIN_REPO=<gbrain checkout> test/fixtures/skill-evals/_liband changes nothing else. company-brainify: Phase 3 and the Contract say the staging copy leaves thepersonal brain's working tree untouched, and the deletion log entry is post-purge only.meeting-ingestion: the split rule says to surface the boundary to the user only when the boundary is- unclear, and every attendee gets a page, even a thin one. A three-trial pilot of the two failed tasks- on GPT-6.1 Sol runs first and is not counted. The counted v2 run is a new `after` arm (12 tasks x 3+ unclear, and Phase 7 takes back the before text of its attendee steps ("Does a people page+ exist? Only an exact match counts…", "If NO → create via the enrich skill… even a thin one"), because+ a first uncounted pilot of v2 still stopped to ask, quoting "exact match only". Uncounted three-trial pilots of the two failed+ tasks on GPT-6.1 Sol: `staging-copy` 3/3; `split-recording` 0/3 before the Phase 7 restore and 2/3+ after it. The counted v2 run is a new `after` arm (12 tasks x 3trials) on Opus 5.5, Sonnet 5.5 and GPT-6.1 Sol, judged by the same keep rule against the `before`arm above, which is reused because the tasks and the shipped skill are unchanged; one rerun of a redtask is allowed.