Warming Bayes hung on a single event — `account.state_changed` reaching "live".
That covers an account added from now on and nothing else. An account that went
live before this plugin existed never gets one; neither does one whose Junk
folder was empty at the time and filled up later, or one whose owner enabled
spam-folder sync afterwards. A worker restarting at the wrong moment loses it
too. The result was a feature that quietly required someone to run it by hand
per account, which is not a feature.
An hourly sweep now walks every live account and offers each a bootstrap. That
works because the per-account job was already cheap to repeat: it returns at once
for an account already done, and declines — without marking it done — when the
corpus is still too small to learn from. So the sweep simply keeps asking until
there is something worth learning, then does it once.
`s.accounts.live()` is the new SDK surface it needs: cross-user by design,
because a sweep runs for the instance rather than a request. Ids and state only,
so reading anything still goes through the user-scoped accessors.
One failing account cannot stop the sweep — the mailbox that raises is logged and
the rest still get their turn.
docs/plugins.md now says this in general terms next to `JobSpec`: a state-change
hook is one-shot, so pair it with a sweep when the work must eventually happen
for everything.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018SqJYt7jD7oUEkoY68sBZo
docs/plugins.md described an SDK that no longer exists. It now covers
messages.raw/sample, actions.move_to_special, the envelope's account_id /
mailbox_special_use / account_state, the message.changed transition payload, the
message.actions.menu slot, Slot.title and .wm-menu-item — with the reasons, not
just the signatures. M3's acceptance is that a plugin can be written from this
file alone, and it had drifted.
Acceptance 4 (§C.8): a test that teaches 20 spam + 20 ham to a real engine and
checks the classifier generalises to held-out mail, keeps ham under the banner
threshold, and does not leak between users. It skips when no engine is reachable.
It does NOT assert the literal move_threshold — a synthetic fixture cannot reach
12.0 once the SMTP-time symbols are zeroed, which is the whole point of the
tuning. Recorded as OQ-8 rather than quietly asserting something weaker.
Bayes ignores training until min_learns, stock 200. For a personal mailbox that
means every correction appears to be ignored for months, so it is now 20.
The bootstrap would have made this filter worse. Accounts do not sync their
provider's spam folder by default, so the real corpus here is 3 junk beside 1,294
aged inbox — teaching both as found trains "everything is ham". It now refuses
below 10 spam samples (leaving the account unmarked so a later run can do it
properly) and otherwise caps ham at 3x the spam count.
Run against the live account, it correctly declined:
"only 3 spam samples (need 10); enable spam-folder sync to build a corpus"
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018SqJYt7jD7oUEkoY68sBZo
lukasandClaude Opus 5