Commit graph

1 commit

Author SHA1 Message Date
a2019d44f7 docs(coord): measure the debt, and record why no second mechanism was built
WP5 asked for one of two candidate mechanisms, chosen by measurement. The
measurement chose neither.

coord-sweep.sh already IS the bulk-ack for pure notices, and it has never
run: no _sweep.log at the default path, and none anywhere under ~/.claude
(stated that way because --log can override the default, so an absent
default-path log alone would not prove it). Its own dry-run agrees exactly
with an independent classification of the same inboxes - 0 at 30 days, 7 at
14, 13 at 7 - so the gap is invocation, not mechanism.

The broadcast class converges on its own: reading records a broadcast as
seen, so a mailbox clears its backlog on its next session. 263 of 884 pairs
are unread, but 34 of those belong to two mailboxes no session can hold -
one a documented retired --to address, one with no checkout anywhere under
/Users/ktg (known-positive control: llm-ingestion-okf resolves). A TTL would
close those rather than reduce them.

What remains is running coord-sweep --write unattended, which closes mail in
51 other repos' inboxes unread. That is a decision on another repo's behalf
and a policy constant of the same class as the STATE.md line limit, so it is
left to the operator rather than shipped. No version bump: nothing here
changes behaviour.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-03 20:48:37 +02:00