claude 8015fdbb79 Harden semantic boundaries and repair dialogue state ...

Replace nearest-neighbour personal routing with a frozen class-balanced linear head measured on historical, stratified, cross-validation, holdout, and fresh challenge gates (V-702). Close the four repair handoff holes, preserve nested clarification flows, and route Russian possession statements through structural grammar rather than lexical exceptions (V-573). Owner explicitly requested direct commits to master.

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

..

2026-07-31-model-bakeoff.md

docs: tier the tree by lifetime, so staleness shows in the path (V-446)

2026-08-02 03:28:49 +04:00

2026-07-31-phrasing.md

docs: tier the tree by lifetime, so staleness shows in the path (V-446)

2026-08-02 03:28:49 +04:00

2026-07-31-recall.md

docs: tier the tree by lifetime, so staleness shows in the path (V-446)

2026-08-02 03:28:49 +04:00

2026-07-31-routing.md

docs: tier the tree by lifetime, so staleness shows in the path (V-446)

2026-08-02 03:28:49 +04:00

2026-07-31-talk.md

docs: tier the tree by lifetime, so staleness shows in the path (V-446)

2026-08-02 03:28:49 +04:00

2026-08-02-workstation-gemma4-12b.md

docs: measure gemma-4-12b on the workstation against the resident model (V-485)

2026-08-03 10:12:21 +02:00

2026-08-03-llama-prompt-cache.md

docs: record the #499 deploy (V-499)

2026-08-03 23:28:36 +04:00

2026-08-03-personal-boundary.md

memory: score the personal boundary on seeds, not word lists (V-495)

2026-08-03 22:57:11 +04:00

2026-08-03-recall-topic-veto.md

memory: record the recall topic veto's real price (V-496)

2026-08-04 00:01:49 +04:00

2026-08-04-ecosystem-reach.md

docs: record the ecosystem reach measurement (V-405)

2026-08-04 06:22:02 +04:00

2026-08-04-recall-e5-small.md

docs: record what the e5-small swap bought (V-371)

2026-08-04 05:51:39 +04:00

2026-08-05-audio-in-routing.md

audio-in routing measured: transcribe then route (V-486)

2026-08-05 17:04:45 +04:00

2026-08-05-half-past-hours.md

half-past hours parse as the hour being entered (V-538)

2026-08-05 14:24:49 +04:00

2026-08-05-kiwix-offline-fallback.md

the ZIM fallback fires fast, and reads Russian in Russian (V-508)

2026-08-05 15:52:54 +04:00

2026-08-05-praxis-reach.md

mavend: a position resolves against the digest she last read (V-516)

2026-08-05 13:11:04 +04:00

2026-08-05-reach-llm-router.md

eval: the resident model never reaches Praxis either (V-517)

2026-08-05 21:03:05 +04:00

2026-08-05-routing-resident-model.md

eval: score and time the resident model as router (V-320)

2026-08-05 21:31:31 +04:00

2026-08-05-search-quality-signals.md

the query source that claimed a turn is readable on /chat (V-539)

2026-08-05 15:28:26 +04:00

2026-08-05-talk-fixture-resident.md

talk fixture on the resident model: 2/36 to 25/36 (V-44)

2026-08-05 14:03:45 +04:00

2026-08-05-temperature-sweep.md

eval: the temperature sweep, and what it found instead (V-402)

2026-08-05 02:49:44 +04:00

2026-08-06-alarm-verbs-reach-stage-0.md

an alarm verb reaches stage 0, and the reminder grammar reads the lexicon (V-627)

2026-08-06 15:04:08 +04:00

2026-08-06-claimant-interleave.md

evals: nine claimants interrupting a parked reminder (V-577)

2026-08-06 01:30:16 +04:00

2026-08-06-fact-parser.md

the fact parser gets a corpus, and the LLM arm gets run (V-586)

2026-08-06 12:21:50 +04:00

2026-08-06-reminder-completeness-on-the-box.md

eval: the reminder completeness rule measured on the box (V-579)

2026-08-06 02:37:33 +04:00

2026-08-06-routing-trajectory.md

the audit write-ups, so every agent starts with them (V-638)

2026-08-06 22:41:03 +04:00

2026-08-06-russian-acts-reach-tools.md

a Russian act reaches a tool, and the seeds stop disagreeing (V-633)

2026-08-06 18:09:11 +04:00

2026-08-06-seed-labels-vs-router-prompt.md

label seeds with the stage 0 grammars and gemma, and measure both (V-546)

2026-08-06 13:21:06 +04:00

2026-08-06-seeds-to-prompt-boundary.md

move the seed files onto the router prompt's intent boundaries (V-626)

2026-08-06 13:26:23 +04:00

2026-08-07-store-connection-cap.md

Keep the store's one connection, delete the seam it cannot survive (V-642)

2026-08-07 01:01:27 +04:00

2026-08-07-week-of-usage-transcript.md

Record the week of usage that found V-654 and its siblings (V-654)

2026-08-07 11:59:40 +04:00

2026-08-07-week-of-usage.md

Record the week of usage that found V-654 and its siblings (V-654)

2026-08-07 11:59:40 +04:00

2026-08-08-clarify-head-four-head.md

Record the clarify head and the confidence it replaces (V-661)

2026-08-08 20:59:44 +04:00

2026-08-08-destination-fixture.md

Fix the floor corpus and re-measure the destination head (V-661)

2026-08-08 20:10:14 +04:00

2026-08-08-destination-model-arm.md

Record the model arm, and the stage 0 trade it exposed (V-660)

2026-08-08 18:23:22 +04:00

2026-08-08-massive-warm-start.md

Record the destination number and two stuck measurements (V-659)

2026-08-08 18:07:41 +04:00

2026-08-08-parked-clarify-ride.md

Measure what the two clarify bounds bought (V-663)

2026-08-08 21:49:32 +04:00

2026-08-08-routing-heads-in-go.md

Record what the routing heads score in Go (V-664)

2026-08-08 22:32:48 +04:00

2026-08-08-routing-heads-two-head.md

Fix the floor corpus and re-measure the destination head (V-661)

2026-08-08 20:10:14 +04:00

2026-08-08-slot-head-three-head.md

Record a fortnight of usage as a re-runnable baseline (V-661)

2026-08-08 20:27:27 +04:00

2026-08-08-two-weeks-after-merge-transcript.md

Re-run the fortnight against merged master (V-661)

2026-08-08 21:14:52 +04:00

2026-08-08-two-weeks-after-merge.md

Read the source column off the badge, not off the wording (V-662)

2026-08-08 21:31:22 +04:00

2026-08-08-two-weeks-transcript.md

Record a fortnight of usage as a re-runnable baseline (V-661)

2026-08-08 20:27:27 +04:00

2026-08-08-two-weeks.md

Re-run the fortnight against merged master (V-661)

2026-08-08 21:14:52 +04:00

2026-08-09-crisperwhisper2-russian-wer.md

Measure CrisperWhisper 2.0 in Russian against the deployed floor (V-665)

2026-08-09 00:28:58 +04:00

2026-08-09-e4b-phrasing.md

Measure E4B on phrasing, the half nobody had scored (V-668)

2026-08-09 12:07:13 +04:00

2026-08-09-e4b-vs-12b-routing.md

Swap the workstation model to gemma-4-E4B (V-486)

2026-08-09 01:15:00 +04:00

2026-08-09-kiwix-topic-retrieval.md

Try the capitalized title too, and reach Париж (V-668)

2026-08-09 10:52:24 +04:00

2026-08-09-silero-vad.md

Hear speech instead of loudness in mavwaked (V-487)

2026-08-09 01:43:32 +04:00

2026-08-09-wake-word.md

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

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

2026-08-10-repo-audit.md

Give the audit's open findings a home and a trigger (V-674)

2026-08-11 10:41:54 +04:00

2026-08-11-classifier-baseline-after-tokenizer-fix.md

Correct the classifier baseline after tokenizer repair (V-704)

2026-08-13 01:27:13 +04:00

2026-08-13-conversation-continuity.md

Preserve context across conversation intents (V-542)

2026-08-13 02:14:46 +04:00

2026-08-13-personal-boundary-linear-head.md

Harden semantic boundaries and repair dialogue state

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

2026-08-13-personal-boundary-neighbourhood.md

Harden semantic boundaries and repair dialogue state

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

CLAUDE.md

Harden semantic boundaries and repair dialogue state

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