claude 774c0660b0 router/semantic: slice 16 diagnostic — action/non-action boundary analysis ...

Experiments:
- Residual-only OOF safety metrics
- False-action decomposition by semantic family (7 families, 24 split groups)
- Paired e5 geometry (gap=0.010 between action and capability question)
- Binary action probe (PR-AUC=0.707, ROC-AUC=0.863)
- Cost-sensitive linear classification
- Expanded action-threshold sweep (7 thresholds)
- Voice-like punctuation stress evaluation
- e5 + tiny structural features
- Four-hypothesis comparison table

Verdict: e5 representation is the primary bottleneck. Action seeds and
capability questions are nearly indistinguishable in embedding space.

2026-09-07 14:02:00 +04:00

..

corpus-factory

router/semantic: corpus-factory deterministic surface generation (slice 15)

2026-09-07 13:52:12 +04:00

e2eprobe

Give reminder cancellation its own store and IPC path (V-719)

2026-08-15 17:19:13 +04:00

labelgen

Harden semantic boundaries and repair dialogue state

2026-08-13 03:00:31 +04:00

mavcaldav

Read the CalDAV password from a file, not from argv (V-644)

2026-08-07 01:19:33 +04:00

mavenclient

sweep: name voice-daemon magic numbers, drop dead keep-alive vars (V-581)

2026-08-06 03:00:21 +04:00

mavend

router: Router.Route consumes NormalizedInput directly (slice 11)

2026-09-07 01:32:45 +04:00

mavgpud

Score the boundary against wrong credentials, not just right ones (V-673)

2026-08-11 10:13:10 +04:00

mavmaild

mavmaild: own its state volume, retire aged-out UIDs, stop restart-looping

2026-08-01 14:08:32 +04:00

mavpoll

mavpoll refuses the flags that would kill it later (V-611)

2026-08-06 04:03:48 +04:00

mavseal

shutdown: close the sockets, or the database never gets sealed

2026-08-01 20:05:52 +04:00

mavsttd

sweep: name voice-daemon magic numbers, drop dead keep-alive vars (V-581)

2026-08-06 03:00:21 +04:00

mavttsd

sweep: name voice-daemon magic numbers, drop dead keep-alive vars (V-581)

2026-08-06 03:00:21 +04:00

mavupdate

update: roll back what the restart actually deploys

2026-08-01 14:06:00 +04:00

mavwaked

Pin the keyword models to one thread each and ship them (V-487)

2026-08-09 15:56:07 +04:00

mavweb

Make /reminders the complete cancellation surface (V-719)

2026-08-15 17:19:38 +04:00

merge-corpus

router/semantic: merge-corpus tool for frozen holdout preservation (slice 15)

2026-09-07 13:51:53 +04:00

semantic-router-experiment

router/semantic: slice 16 diagnostic — action/non-action boundary analysis

2026-09-07 14:02:00 +04:00