kami 89d83c0b11 Record the example-led nudge prompt experiment (#393) — it made things worse ...

Tried rewriting the nudge prompt to lead with five on-topic examples instead
of rules. Three eval runs each side: before 12/13/14 of 15, after 11/12/11.
The loss is all in the address check — formal "вы" and plural imperatives came
back once the "говоришь на ты" rule stopped being its own sentence, and the
on-topic examples leaked their wording into the wrong cases.

Prompt reverted. Only the finding is committed.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ

2026-07-31 15:40:18 +04:00

cmd

Tell the phraser to speak to him informally, singular

2026-07-31 14:52:23 +04:00

deploy

Show clarify_max_attempts in the deployed config

2026-07-31 12:34:28 +04:00

docs/plans

feat(mavweb): /ecosystem page consuming Nexus/Praxis/Hexis + shell fixes

2026-07-19 22:04:23 +04:00

internal

Don't let a verb count as the person being talked about

2026-07-31 14:54:20 +04:00

models/seeds

feat: event store, pattern inference, and routine proposals

2026-07-10 15:49:03 +04:00

scripts

scripts: add maven-backup.sh for encrypted DB backup/restore

2026-07-05 11:48:46 +04:00

vendor

Add version/correlation headers to Nexus and Praxis ecosystem clients

2026-07-20 11:12:01 +04:00

.dockerignore

fix: mavwaked is a client binary, not a docker daemon

2026-07-06 14:12:33 +04:00

.gitignore

Ignore the .claude directory

2026-07-31 02:32:44 +04:00

20-07-2026-BACKLOG.md

Record the 2026-07-20 session and the 2026-07-30 senior review

2026-07-30 23:49:44 +04:00

AGENTS.md

Swap the embedder to multilingual-e5-small (Vikunja #371, #372)

2026-07-31 11:38:18 +04:00

CLAUDE.md

Pin the resident model to Qwen3.5-0.8B and name Qwen3-1.7B as the target

2026-07-30 23:40:33 +04:00

DESIGN.md

Fold SPEC/maven/ROADMAP into DESIGN.md and drop the stale session logs

2026-07-30 23:39:56 +04:00

docker-compose.yml

Bind mavweb to loopback and make the step-up fail-open loud and overridable

2026-07-30 23:39:07 +04:00

Dockerfile

feat(ecosystem): compliant Praxis/Hexis integration + vendored build

2026-07-19 20:24:33 +04:00

go.mod

feat(ecosystem): compliant Praxis/Hexis integration + vendored build

2026-07-19 20:24:33 +04:00

go.sum

feat(ecosystem): compliant Praxis/Hexis integration + vendored build

2026-07-19 20:24:33 +04:00

kill-maven.sh

Don't fail when docker confirms nothing is running

2026-07-31 14:32:45 +04:00

Makefile

Merge branch 'worktree-agent-ad5da57e47b822152' into overnight-jul31

2026-07-31 11:39:35 +04:00

MAVEN_ECOSYSTEM_ARCHITECTURE.md

Pin the resident model to Qwen3.5-0.8B and name Qwen3-1.7B as the target

2026-07-30 23:40:33 +04:00

maven-feature-ranking.md

Fold SPEC/maven/ROADMAP into DESIGN.md and drop the stale session logs

2026-07-30 23:39:56 +04:00

MODEL-BAKEOFF-31-07-2026.md

Bake off LFM2.5 against Qwen3.5-0.8B on the RU routing fixture

2026-07-31 12:07:06 +04:00

PHRASING-EVAL-31-07-2026.md

Record the example-led nudge prompt experiment (#393) — it made things worse

2026-07-31 15:40:18 +04:00

PROGRESS.md

Record the 2026-07-20 session and the 2026-07-30 senior review

2026-07-30 23:49:44 +04:00

PROTOCOL.md

PROTOCOL.md + away-channel fallthrough

2026-07-03 13:18:38 +02:00

REARCH.md

feat(mavweb): /ecosystem page consuming Nexus/Praxis/Hexis + shell fixes

2026-07-19 22:04:23 +04:00

RECALL-EVAL-31-07-2026.md

Let a note win a recall, not only a fact (#373)

2026-07-31 13:30:38 +04:00

REVIEW-30-07-2026.md

Record the 2026-07-20 session and the 2026-07-30 senior review

2026-07-30 23:49:44 +04:00

ROUTING-EVAL-31-07-2026.md

Give the router prompt a rule for clock and calendar questions (#374)

2026-07-31 13:52:00 +04:00

start-maven.sh

initial commit

2026-07-03 00:32:48 +02:00

START.md

Fold SPEC/maven/ROADMAP into DESIGN.md and drop the stale session logs

2026-07-30 23:39:56 +04:00

S

Description

No description provided

70 MiB

Languages

Go 97.1%

HTML 0.9%

Shell 0.6%

CSS 0.5%

Makefile 0.3%

Other 0.6%