Repository navigation
Patch release: memex check --condense sees memos that arrive by consolidation. No schema change.
Added
ARRcolumn inmemex check --condense.NEWcompares a memo's own date with
the overview'scondensed:stamp, so a memo consolidated in from another folder —
which keeps its original, older date — was invisible to it; only theGAPcount
hinted, and a count can be explained away. On 2026-09-30 a stamp audit found 8 of 9
overviews that had been dismissed as "count mismatches" each hid exactly such a
memo.ARRcounts memos whose git first-add date (onegit log -z --diff-filter=A --no-renamesfor the whole vault, committer date, 0.04 s for 3,400 paths) is after
condensed:although the memo is dated on or before it, and names them under the
row (five, then+N more (--json)). Untracked memos count as arrived today. When
git gives no evidence — not a checkout, a shallow clone, a timeout, or a project
folder git doesn't track — the column prints-and the legend says why
(arrivals_basisin--json), never "everything arrived".memos/<sub>/set-asides are reported, not counted. A curator's
memos/archive/no longer reads as stamp drift; the row and the drift line say
+N under memos/*/.
Changed
- garden-tending and curator-practice skills: read
GAPandARR, notNEWalone;
and the fold-wave rule learned the hard way today — a live user session is a writer
the ownership map doesn't list: fold agents re-read a file immediately before
writing it and use targeted edits, and the orchestrator diffs the touched topics for
bullets dated after dispatch before accepting a wave.
Reviewed by Codex (3 rounds) and Kimi (2 rounds); 883 tests.