Tonight's work as one branch: all 46 reviewed PRs, verified green #47

Merged
kami merged 135 commits from integration/jul31 into master 2026-07-31 19:41:52 +02:00
Owner

Everything from tonight, on one branch, verified as a unit. Replaces the 46 stacked PRs — you reviewed the content already, this is the same content with the topology problems taken out.

135 commits. make build produces all 8 binaries. make test exits 0: 38 packages, no failures, no data races, gofmt and vet clean.

Why one PR

42 of the 46 PRs were already ancestors of overnight/resident-1.7b. Only 7, 41, 43, 44 sat outside it, and 44 already carried 41. So the whole tower was three merges, not a 46-way restructure.

Contains all 46: verified by ancestry check, every PR head is an ancestor of this branch.

What had to be fixed to make it build as one thing

  • internal/store/migrations.go — merging PR7 conflicted. Migration #8 (idx_nudges_snoozed) was already present byte-identical, so PR7's content was fully absorbed; resolved to the existing version, nothing lost.
  • gofmt — PR41 and PR44 landed internal/kiwix/client.go and rewrite_test.go unformatted. The gate PR12 added to make test failed the moment both were on one branch. Struct-tag and comment alignment only.
  • MODEL-BAKEOFF-31-07-2026.md — the file led with "Recommendation: keep Qwen3.5-0.8B" while ending with the switch to Qwen3-1.7B, and contradicted CLAUDE.md. Both sweeps are accurate, so nothing was rewritten: the lede now states the outcome and the first sweep's verdict is scoped to what it actually tested (it rejects LFM2.5-1.2B, which still holds).

Still open, deliberately not done here

Your review comments on PR4 (ipc/server.go at 947 lines with 43 case arms; mavweb/main.go at 1296 lines of HTML-in-Go) are pre-existing debt these changes sit on top of, not something they introduced. Refactoring it mid-merge would have made this unreviewable. It wants its own task.

Also unaddressed: the Russian wording notes on 6/21/23 ("На какое время" / plain "Когда?", "тепло", the persona hand-leak, ~5 timeout phrasings instead of one) and your questions on 3 and 15. Say the word and I'll do them as a follow-up on master.

PR1's six comments are answered — review #37 on that PR. All six are genuinely fixed by PR22, which was written to answer them; they were a pair.

Note PR44's own body says its score shows the Kiwix rewrite does not work yet. It is included because external search is off unless configured, but it is not a working feature.

Everything from tonight, on one branch, verified as a unit. Replaces the 46 stacked PRs — you reviewed the content already, this is the same content with the topology problems taken out. **135 commits. `make build` produces all 8 binaries. `make test` exits 0: 38 packages, no failures, no data races, gofmt and vet clean.** ## Why one PR 42 of the 46 PRs were already ancestors of `overnight/resident-1.7b`. Only **7, 41, 43, 44** sat outside it, and 44 already carried 41. So the whole tower was three merges, not a 46-way restructure. Contains all 46: verified by ancestry check, every PR head is an ancestor of this branch. ## What had to be fixed to make it build as one thing - **`internal/store/migrations.go`** — merging PR7 conflicted. Migration `#8` (`idx_nudges_snoozed`) was already present byte-identical, so PR7's content was fully absorbed; resolved to the existing version, nothing lost. - **`gofmt`** — PR41 and PR44 landed `internal/kiwix/client.go` and `rewrite_test.go` unformatted. The gate PR12 added to `make test` failed the moment both were on one branch. Struct-tag and comment alignment only. - **`MODEL-BAKEOFF-31-07-2026.md`** — the file led with *"Recommendation: keep Qwen3.5-0.8B"* while ending with the switch to Qwen3-1.7B, and contradicted `CLAUDE.md`. Both sweeps are accurate, so nothing was rewritten: the lede now states the outcome and the first sweep's verdict is scoped to what it actually tested (it rejects LFM2.5-1.2B, which still holds). ## Still open, deliberately not done here Your review comments on **PR4** (`ipc/server.go` at 947 lines with 43 `case` arms; `mavweb/main.go` at 1296 lines of HTML-in-Go) are pre-existing debt these changes sit on top of, not something they introduced. Refactoring it mid-merge would have made this unreviewable. It wants its own task. Also unaddressed: the Russian wording notes on **6/21/23** (`"На какое время"` / plain `"Когда?"`, `"тепло"`, the persona hand-leak, ~5 timeout phrasings instead of one) and your questions on **3** and **15**. Say the word and I'll do them as a follow-up on master. **PR1's six comments are answered** — review #37 on that PR. All six are genuinely fixed by PR22, which was written to answer them; they were a pair. Note **PR44's own body says its score shows the Kiwix rewrite does not work yet**. It is included because external search is off unless configured, but it is not a working feature.
kami added 135 commits 2026-07-31 19:37:41 +02:00
The router prompt tested "reports current state -> fact" before "wants
information -> query", so a question naming a fact key was written as a fact.
Query now comes first, plus an explicit question test.
Reviewers: the prompt block in llmrouter.go, and the note about the
training-side copy of the prompt that needs the same edit.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Route now sets RepeatPenalty on the request, and the grammar's string rule is
capped at 120 characters. Two of 76 fixture cases looped one sentence inside
the text field until MaxTokens, which cut the JSON in half.
Reviewers: the new constant and the grammar string rule.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Table-driven tests for water, meal, break, service_down and netdata_critical,
straight against the predicate with a fake State. Reviewers: the no-data rows
(every rule must stay quiet when its key is missing) and the ops forgery rows,
where a fact with the right value but the wrong source must be refused.
No rule fired on missing data, so no fix was needed.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
router.Slots already carries the fact payload; the dialogue copy did not, so a clarifying answer had nowhere to put it. InheritSlots carries it like Key.
Reviewer: check the new inherit block does not overwrite a filled value.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Pins the conservative side of Gate(): quiet hours, away, calendar-busy,
cooldown and snooze, plus one nudge per tick at max severity. Reviewers: the
two skipped tests at the bottom are real gaps, not flakes. Reminders ignore
snooze (loop.go:120) and the Gatherer never fills SnoozeUntil (gather.go:153),
so snooze does nothing at runtime. No behaviour was changed.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
PendingQuestion plus ClarifyStore: same shape, locking and expiry as SessionStore. Answer fills only the missing slots and never overwrites a filled one. No wiring yet — TODOs mark the daemon hooks.
Reviewer: MaxAttempts is 1 on purpose (Maven asks once, she is not a nag).

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Wires cmd/mavend/voice.go to build the LLM router when the operator asks
for it. Default false, so nothing changes on the deploy box.
Look at pickLLMRouter: the flag on with no llama-server logs one line and
keeps the classifier, it never fails a turn.
The default stays off until the router can refuse (#359) and the extractor
runs on LLM decisions — both noted as TODOs in config.go.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The MODEL default was LFM2, but the deploy runs Qwen3.5-0.8B, so the
pkill pattern matched nothing and the server survived every kill.
Now matches any llama-server serving a .gguf, so changing the model in
deploy/mavend.json cannot break the script again. MODEL still narrows it.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Look at internal/store/proposed_routines.go: status flips in place with an
`AND status = 'proposed'` guard, not append-only like facts/voids_id — a
proposal is a question with one answer, same shape as tools.status. The
UNIQUE(action, object) key is what stops a dismissed routine coming back.
New tests cover re-propose-after-dismiss and listing by status.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
.gitignore had deps/ and /models/llm/ with trailing slashes. A trailing slash
only matches a real directory, so a *symlink* with the same name is not ignored
and git add -A commits it as a symlink blob.

That bites anyone working in a git worktree, where deps/ and models/ do not
exist and have to be linked in from the main checkout. It already happened once
tonight.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The bake-off in #278/#250 needs two models' scores side by side, and the
report names only carried the config, so the rows were indistinguishable.
ModelID reads /v1/models instead of taking a string that goes stale.
New target: make eval-models MAVEN_LLM_URL=...

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Measures whether Maven can find the right note again from a paraphrased
question. Review internal/memory/recalleval/recalleval.go's Score for how
rank, gate and false recall are kept as three separate numbers, and the
fixture's filler list for why recall@3 is not free.
Fixture JSON is generated data and does not count toward the diff limit.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The router can already say "I am not sure" (Decision.Clarify) but the daemon
had nowhere to keep the request while it asked. PendingQuestion holds the
original slots, ClarifyStore parks one per dialogue id with a 90s TTL, and
Answer fills only the slots that were missing so an answer can never rewrite
what she already understood. Logic that uses this comes next.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
A table per intent (reminder needs a time, fact needs a key, act needs a fn)
plus one fixed Russian question per slot. Templates, not model output: a 0.8B
would wander and a question that rewords itself is harder to answer. Note,
query, chat and system get no question — for those a clarify decision keeps
the canned reply rather than inventing a question for noise.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Accepting a routine gives the trigger loop a new standing reason to speak to
the human, so it is the same authority tier as enabling a tool and shares the
stepUpOK gate; dismiss only ever makes maven quieter, so it is ungated.
Look at handleRoutines and acceptRoutine in cmd/mavweb/main.go: accept creates
the recurring reminder, then links it via the new ipc AcceptProposedRoutine.
The page now says what maven noticed in her own words (pattern.PhraseRoutine).

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Table-driven tests for all four severity bands crossed with present and away,
both as the pure table and end to end through the dispatcher. Three tests are
written to DESIGN.md and skipped because the code does not keep the claim: the
care-away drop is recorded nowhere, and the minimal body is enforced per-sink
rather than by the dispatcher. Also gofmt'd dispatcher_test.go.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
On a clarify decision with one identifiable gap she now asks instead of saying
"не поняла", and parks the request. The next utterance is parsed as the answer
with the router's own extractor and the completed decision runs through
applyAction like any other — so a clarified act still needs the allowlist and
still hits the destructive confirm gate. An answer that does not fill the gap
drops the request; she never asks twice. Also pulls the session-store block
that HandlePushToTalk and handleText both had into rememberTurn, since the
clarify path needed a third copy.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Fallthrough is checked per severity through the outbox trail, so sev3/sev4
reroute and sev1/sev2 still drop. Durability uses a real store on a temp file:
a crash between Begin and Complete becomes unknown, is not resent, is not
dropped, and a late Complete cannot overwrite it. One skipped test marks a real
gap: a panic mid-send leaves a permanent pending row.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Covers: a reminder with no time is asked about and completes on the answer; the
same for a fact; an answer past the TTL falls through as a fresh utterance; a
second unclear answer drops the request with no second question; a clarified act
off the allowlist neither runs nor gets enabled; a clarified destructive act
still parks a confirm; noise keeps the canned reply. No model, no network.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The gate honours State.SnoozeUntil but nothing ever filled it. New
store.SnoozedUntil returns, per rule, when the newest snooze runs out.
Reviewer: the fixed 2h SnoozeDuration and its reasoning in nudges.go —
nothing upstream can supply a per-nudge length, so no new column.
Expired snoozes are dropped in SQL, so silence can never be permanent.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Review internal/phraser/eval/checks.go -- it IS the measurement. Each check
names in a comment which DESIGN.md line it defends: length, feminine
self-reference (windowed around "я" so the operator's own masculine
second-person forms are not flagged), the cringe list (pet names, emoji,
"!!", fake concern, apology, emotional support, asking how he feels,
praise), on-topic, mood enum. No send/veto signal anywhere, per
DESIGN.md § "Rules decide, LLM phrases".
Fixture (158 lines) and tests (252) do not count toward the diff ceiling;
the scorer itself is still ~650. Splitting eval.go from checks.go would
give two commits neither of which measures anything.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
# Conflicts:
#	internal/dialogue/clarify.go
#	internal/dialogue/clarify_test.go
A table per intent (reminder needs a time, fact needs a key, act needs a fn)
plus one fixed Russian question per slot. Templates, not model output: a 0.8B
would wander and a question that rewords itself is harder to answer. Note,
query, chat and system get no question — for those a clarify decision keeps
the canned reply rather than inventing a question for noise.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
On a clarify decision with one identifiable gap she now asks instead of saying
"не поняла", and parks the request. The next utterance is parsed as the answer
with the router's own extractor and the completed decision runs through
applyAction like any other — so a clarified act still needs the allowlist and
still hits the destructive confirm gate. An answer that does not fill the gap
drops the request; she never asks twice. Also pulls the session-store block
that HandlePushToTalk and handleText both had into rememberTurn, since the
clarify path needed a third copy.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Covers: a reminder with no time is asked about and completes on the answer; the
same for a fact; an answer past the TTL falls through as a fresh utterance; a
second unclear answer drops the request with no second question; a clarified act
off the allowlist neither runs nor gets enabled; a clarified destructive act
still parks a confirm; noise keeps the canned reply. No model, no network.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The Gatherer now fills State.SnoozeUntil from store.SnoozedUntil instead
of nil, so a snooze finally reaches the gate. RemindDecisions gains the
one restraint check that applies to a reminder — quiet hours, presence
and cooldown are still bypassed, so "wake me 7" is unchanged. Reviewer:
the two tests in internal/loop/gate_test.go are the contract.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Agent worktrees land in .claude/worktrees, so the directory shows up as
untracked noise in every git status.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
# Conflicts:
#	internal/loop/gate_test.go
The gate honours State.SnoozeUntil but nothing ever filled it. New
store.SnoozedUntil returns, per rule, when the newest snooze runs out.
Reviewer: the fixed 2h SnoozeDuration and its reasoning in nudges.go —
nothing upstream can supply a per-nudge length, so no new column.
Expired snoozes are dropped in SQL, so silence can never be permanent.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The Gatherer now fills State.SnoozeUntil from store.SnoozedUntil instead
of nil, so a snooze finally reaches the gate. RemindDecisions gains the
one restraint check that applies to a reminder — quiet hours, presence
and cooldown are still bypassed, so "wake me 7" is unchanged. Reviewer:
the two tests in internal/loop/gate_test.go are the contract.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Real recall is 48% after the gate, and one must-be-silent query gets an
answer anyway. Review finding 2 (the score distributions overlap, so no
gate separates a real recall from a false one) and finding 4 (the memStore
branch at voice.go:776 is unreachable for notes). Adds an embedder cache
so the gate sweep does not re-embed the fixture nine times.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Away channels (ntfy, telegram) leave the box, so an empty Summary now sends
a fixed generic line plus the rule name instead of the full Body. The
dispatcher strips detail before any sink sees it, so a sink added later
cannot leak by reading the wrong field. Voice is local and unchanged.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
A panic in Send used to unwind past completeOutbox and leave the
delivery_attempts row pending forever, since reconciliation only runs at
startup. safeSend turns the panic into an error, logs it loudly, records the
attempt failed, and lets the other channels for the same nudge still go out.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The integration branch names one test TestAwayFallsBackToFullBodyWhenSummaryEmpty,
which describes the old bug; it is here as TestAwaySendsGenericLineWhenSummaryEmpty
and asserts the generic line instead of the body. Also covers: a normal summary
goes out unchanged, voice keeps the full body, and one panicking sink does not
eat the other channel for the same nudge. Reformatted one pre-existing struct.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Data layer only. Migration #8 adds accepted_ts and last_fired_ts to
proposed_routines, plus ListAcceptedRoutines and MarkRoutineFired so the
tick loop can own the schedule. Accepting no longer links a reminder id.
Look at the TODO(vikunja#366) in cmd/mavend/tick.go for the next commit.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Two agents wrote the same three test helpers and names for the same two
bugs. Kept the real assertions from dispatcher_test.go and removed the
skipped placeholders they replace. panicSink stays in durability_test.go
since both files use it.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The tick loop now reads accepted routines from the store and nudges when
their interval has passed; accepting no longer builds a one-shot reminder.
Look at routine.DueAccepted for the schedule rule (no catch-up backlog) and
at fireAcceptedRoutines for the restraint gate — routines do not bypass it.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Two merge fixes on top of the branch:

- migrations: keep both new steps, snooze stays #8, the routine columns
  become #9. Both agents had numbered theirs #8.
- accepting no longer takes a reminder id, on the web surface too. The
  web accept path had the same one-shot-reminder bug the voice path did,
  so both now just flip the status and let the tick loop schedule.

The test that asserted "accept creates a reminder and links it" asserted
the bug. It now asserts that accepting creates no reminder.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Formatting only: import order, and statements that were packed onto one
line split out. `git diff -w` shows nothing but that.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
DESIGN.md has always said `make test` is "gofmt + vet + -race, no
exceptions". It only ever ran the tests, which is how nine files drifted
out of format without anyone noticing.

`test` now depends on `fmt-check` and `vet`. Checked that fmt-check does
fail when a file is unformatted, so the gate is real and not decorative.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
# Conflicts:
#	Makefile
# Conflicts:
#	Makefile
The earlier before/after was taken while another eval shared
llama-server. This run had the box to itself.

Intent accuracy 61.8% llm-only, 63.2% cascade, 67.1% with thinking off.
The prompt fix holds. note→fact shows up here too, so it is real.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Chose an 8th enum value over a confidence number: the model already picks
one enum token, so it costs nothing in the grammar, while a score from a
0.8B model would be uncalibrated noise. A refusal returns "no decision"
with no error, which is the fall-through the caller already uses for a
bad parse, so the classifier and its clarify gate take the turn.

Reviewers: the prompt's counter-examples matter most — a small model will
over-use any easy escape hatch. The training workspace copy of the prompt
still needs the same edit (Vikunja #362).

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Note recall is asymmetric: a short question goes in, a longer note comes
out. Adds EmbedQuery/EmbedPassage helpers and the e5 prefixes, and points
the note/fact write path at the passage side and the query path at the
query side. Reviewers: the three call sites in voice.go.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The old model was a symmetric paraphrase model, so it scored "do these
look alike" instead of "does this note answer this question". Also fixes
the file mismatch: the Makefile, the deploy config and both evals now all
name the same quantized file, and the quantized one is what gets measured.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
recall@1 60% to 72%, answered 48% to 72%, latency 3x better. But false
recall went 1/5 to 5/5: e5 packs every score into a narrow high band, so
the 0.55 gate now admits everything. Left the gate alone as instructed.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
An LLM-routed reminder came back with no parsed time and an act with no
fn, because only the classifier path ran the extractor. Now the router
runs the same extraction after an LLM decision and fills only the empty
slots. No time in the utterance still means no time.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The two things that made this unsafe are fixed: the router can now
refuse, and slot extraction runs on its decisions.

On the held-out fixture it gets 63.2% of intents right against the
classifier's 50.0%, with no route errors. It costs about a second a
turn instead of 30ms.

The flag is a pointer now, so leaving it out of the config means on
and only writing false turns it off.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The e5 embedder puts every cosine in one narrow band (0.79-0.89), so the
absolute query_min_score gate cannot tell a real hit from a made-up
question: any value under the band answers everything, any value above it
answers nothing. False recall was 5/5.

New gate asks whether one note is clearly the best instead: top1 - top2 >
delta. New query_min_margin config knob, default 0.008, read off the sweep
in the recall harness. The absolute floor stays as a second check.

On the recall fixture with e5: answered 72% -> 68%, false recall 5/5 -> 1/5.

Vikunja #359

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Third section: why the absolute gate could not separate the two
distributions, the delta sweep, and the before/after. Marks next-steps
item 3 done.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
# Conflicts:
#	deploy/mavend.json
#	internal/config/config.go
Vikunja #278 / #250. Keep Qwen: LFM2.5-1.2B loses 8 points of intent
accuracy, all of it Russian, and runs 2.4x slower.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The system prompt showed the JSON contract as {"response": "..."} and the
user prompt repeated it. A 0.8B copies whatever sits in the response slot, so
7 of 15 nudges came back as literally "...".

Changes, all prompt-side — the {"response","mood"} contract is unchanged:
- nudge system prompt is Russian, feminine self-reference, with filled-in
  examples on topics that never appear as rules, so copying them is visible
- rule names get a Russian gloss and a required keyword, named last in the
  prompt where a small model weights it hardest
- durations render in Russian, not English
- the no-parse fallback says something Russian instead of "water — care",
  which was going straight to a Russian piper voice
- same "..." placeholder removed from replier_llm.go

Scored on internal/phraser/eval: 0/15 -> 13/15.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
MaxAttempts was 1, justified as "not a nag". Wrong reading: "not a nag" is about
interrupting unprompted, and a clarifying question is part of a conversation he
started. Now three, configurable via voice.clarify_max_attempts (default 3).
Three, because after that the likely problem is she misheard the whole request,
not one slot.

Answer used to keep the parked value, so "в три" then "нет, в пять" threw the
five away. Now a value the answer carries wins for the slot she asked about.
Only for the clarify answer — a correction in a fresh turn is followUpMerge.

The eight-field chained assertion in the Answer test is one DeepEqual now, so a
new field in Slots is covered without touching the test.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
An unclear answer used to end the request on the spot. Now she re-asks the same
question while attempts remain, and when they run out she says
"Прости, я не поняла. Скажи, пожалуйста, по-другому." — silence would leave him
thinking it was handled. Same reply when the missing slot has no question to
ask, and as a floor in finishClarified so an empty reply can never ship.

Tests: three questions allowed, the fourth gives up out loud, the cap is
configurable, and a restated time is the one that lands.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
# Conflicts:
#	cmd/mavend/clarify.go
#	cmd/mavend/clarify_test.go
#	cmd/mavend/voice.go
#	internal/config/config.go
The default is 3 either way. Writing it out means you can see the knob
without reading the Go.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Vikunja #382. A parked clarifying question past its TTL was discarded
silently on read; now she says the old request is gone and the newly
spoken words are still routed as a fresh utterance.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Vikunja #363. The follow-up session was a plain in-memory map, so any
mavend restart dropped the thread. It now mirrors to a small TTL-pruned
sqlite table and is loaded on startup; expired sessions are deleted on
load, not revived. Clarify's pending question is untouched.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The 67.1% "thinking off" column in ROUTING-EVAL-31-07-2026.md was an
artefact. It came from a hand-rolled HTTP client in the eval test that
did not send repeat_penalty, so it differed from the reference run on two
axes and the penalty was the one that mattered.

Re-scored back to back on an idle box with everything else held equal:
thinking off is identical to thinking on, case for case, same confusion
matrix, same three unparseable replies. A direct probe of the running
llama-server shows enable_thinking, thinking and reasoning_budget are all
ignored for this model on this build, so there was nothing to turn off.

No defaults changed. The misleading third configuration is removed from
internal/router/eval so its table cannot be quoted again.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The memory pass ran only after the notes-only gate had already rejected
the same note at the same score. Notes and facts share one vector index,
so a note that failed there failed again — the branch could only ever
return a fact.

Now the memory pass runs first: one search over everything Maven
remembers, one gate, and the memory that clearly matches best answers
(a note gets phrased, a fact is read back). The notes-only pass stays
behind it for notes the vector index does not hold. No threshold moved,
so the set of questions answered is unchanged — only which memory
answers them.

Fixture gained two mixed note+fact cases, so the answerable count goes
25 -> 27: hash recall@1 36.0% -> 37.0% (ratchet 0.32 unchanged, comment
updated), e5 recall@1 72.0% -> 70.4%, false recall still 1/5.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The embedder moved from paraphrase-multilingual-MiniLM-L12-v2 to
multilingual-e5-small. Both are 384-dimensional, so nothing in the code
noticed: cosine between an old stored vector and a new query vector is
noise, and recall degrades silently.

So the DB now records the embedder that wrote its vectors. One value for
the whole DB (migration #11, a small `meta` key/value table) rather than a
column on every vector row: the backfill re-embeds every note and fact in
one pass, so a per-row marker would hold the same string in every row and
cost a column on two tables for nothing.

The identity comes from the embedder itself via a new optional ID() method
("multilingual-e5-small@384", model file name plus dimension), so pointing
the config at another model changes the string without anyone editing a
constant. mavend logs a loud WARNING at startup naming both the stored and
the configured embedder when they differ.

Detection only — recall behaviour is unchanged. TODO(#378) in
store.CheckEmbedder marks where the backfill will hook in.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The embedder swap left every stored vector in the old model's space, so cosine against a new query vector is noise. Add the one-shot backfill: store.ReembedAll re-embeds every note and fact text with the currently configured embedder (the passage side, which is the side stored text was written with) and rewrites both places a vector lives — the notes table embedding column and the memory_vectors rows.

All of it plus the embedder marker happens in one transaction, so a failure partway changes nothing and writes no marker: re-run it. A run against a DB whose marker already names the current embedder does nothing.

Triggered explicitly with `mavend -reembed`, not automatically on mismatch: ONNX on the laptop CPU makes this minutes of work, and a silent multi-minute stall on boot would look like a hang. The mismatch warning now tells the user to run it.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The prompt named seven intents but never said which one a clock or date
question belongs to, so the model guessed: system->query x4 in every eval
run. The rule now says the clock and the calendar date themselves are
system, what is written in the calendar or in memory stays query, and a
time named inside a request is just part of the request.

That split follows what the daemon can answer. Only replySystem owns the
clock and the date formatter, while the agenda is answered from
CalendarEvents inside the query branch.

Also adds one calendar-agenda fixture case so an over-broad system rule
cannot pass unnoticed, and writes up the before/after numbers. The
targeted confusion is gone; the headline accuracy did not move.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
replySystem had two arms that PR 30 made reachable, and both answered confidently wrong: the date arm keyword-matched "числ" and always answered today, so "какое число завтра" answered today; the clock arm ignored a named city and answered local time. The date arm now reads the day word through router.ParseCalendarDate (which grew послезавтра/вчера and now cuts the day boundary in the local zone instead of UTC). The clock arm answers the named zone when it resolves offline from the tz database embedded in the binary, and otherwise says plainly that she only knows local time.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The user only ever asks the time in his own zone, so answering other
cities was code kept in step with the weather city list for no gain.
Any named place now gets the honest "local time only" answer that was
already there for unknown cities.

Removes the 22-entry table, the lookup and the embedded tz database.
Closes Vikunja #389 — there is only one city list again.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The away sinks fell back to the whole Body when Summary was empty. ntfy and
telegram leave the box, and the 0.8B phraser drops fields regularly, so that
fallback could push full detail off the machine.

The dispatcher already strips detail from away sendables. This exports that
one rule as delivery.AwayMessage and has both sinks use it, so a sink can't
leak the body on its own either: empty Summary means a generic line plus the
rule name, never the body.

The two sink tests named TestSendFallsBackToBodyWhenSummaryEmpty asserted the
old, wrong behaviour, so they are rewritten to assert the generic line.
TestSendRejectsEmptyMessage is likewise replaced: an away message can no
longer be empty, so the sink has nothing left to reject.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The phrasing run produced two persona breaks that scored clean:
"Приходите… Жду вас" (formal plural) and "Он не ел 11 дней" (talks
about him instead of to him). She is feminine, he is male, and she
speaks to him informally, one to one.

The new `address` check flags the "вы" family, plural imperative
endings, and a third-person "он" with no other subject named earlier in
the message. Like `hisgender` it is a keyword/suffix heuristic, not a
parser, and it prints the word it tripped on so a false alarm is easy to
dismiss. Limits are written out in the comment.

Both recorded strings are pinned as unit tests.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The phrasing eval printed "llm (0.8B, ...)" no matter which gguf
llama-server had loaded, so two runs of two different models came out
named the same and were easy to mix up when comparing.

It now asks llama-server over /v1/models, same as the router eval
already did. The helper moved to internal/llm so both share it, and it
now errors instead of returning a blank name when the id field is
missing — an unreachable server gets labelled "unknown-model", never a
plausible-looking guess.

Both eval paths stay opt-in behind MAVEN_LLM_URL; no server needed for
go test.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The check asks whether anyone else was named before "он". Time words
were not stoplisted, so "сегодня он не ел" read "сегодня" as the person
being talked about and passed — which is the recorded break with a word
in front of it, and nudges open with those words constantly.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Dropping a sev1-2 care nudge while you're away is right and still happens.
But it was a bare `continue`: no row, no log, so "she dropped it", "the gate
suppressed it" and "the rule never fired" all looked identical afterwards.

Adds a 'dropped' delivery status (migration #12 widens the CHECK constraint;
sqlite can't do that in place, so the table is rebuilt) and records the drop
as one delivery_attempts row plus a log line.

No nudges row for a drop: that table feeds the ignored_rate signal, and a
nudge nobody could see must not count as ignored.

TestVoiceNoSessionFallthroughLeavesOutboxTrail expected exactly one row for
sev1-2 when voice had no session. It now expects the voice failure plus the
drop, which is the point of the change.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
docker-compose.yml has no 'pid: host', so each container has its own PID
namespace and pkill on the host matches nothing inside them. The script
then printed "All services gracefully stopped" while mavend, its
llama-server and the rest were still running.

Now it checks for running compose containers first and stops them with
docker compose. If it cannot ask docker and finds nothing to kill on the
host, or anything survives the kill, it says so and exits non-zero
instead of claiming success. The bare-metal path is unchanged apart from
verifying the SIGKILL actually worked, and no longer risks killing the
shell it was launched from.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The migration copied rows with SELECT *, which matches columns by
position. It is correct today, but if the old table's order ever
differed it would shuffle every row instead of failing.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
"Nothing on the host" meant two different things and the script treated
them the same. If docker answers and names no running containers, Maven
really is down and the script should say so and exit 0. Only when docker
cannot be asked is the answer unknown, and that is the case that must
fail loudly.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The prompts stated the feminine self-reference rule but never said whom she is
speaking to, so the model produced formal plural ("Жду вас") and talked about
him in third person ("Он не ел 11 дней"). Adds the address rule right next to
the feminine one, in the nudge prompt and the confirmation prompt.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The third-person check asks whether anyone else was named before "он".
A nudge is mostly verbs, and they were counted as possible people, so
"попробуй встать и отдохнуть — у него есть перерыв" passed. Infinitives
and imperatives now join past tense as words that cannot be a person.

A plain noun before the pronoun still blinds it. That needs a parser,
and the comment says so.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
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
The "address him as ты" rule had only reached two of the five system
prompts. Instead of pasting it into the other three (five copies drift —
that is how this happened), there is now one block, in internal/persona,
prepended to all five: nudges, action replies, chat, note queries and
general knowledge.

The block says who he is and how to address him (a man, always "ты",
never "вы", never "он" about him; Maven stays feminine), plus the
current local date and time. It is rendered fresh each turn because the
time changes, and it is correct with an empty config — the address and
gender rules are defaults in code. Config only adds optional facts:
owner_name, city, and the existing free-text `persona` string, which is
now the static half of the block.

Russian even in front of the English prompts: the rules are Russian
grammar, so they read best stated in Russian, and there is one copy.

Vikunja #394.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The context block now lists her real capabilities: reminders, notes and
facts (write and recall), and the calendar — all three are code paths in
mavend today. Weather, telegram and shell acts are listed only when the
config actually has them, because offering something she cannot do is
worse than staying quiet about it.

Also drops the pronouns from the optional name/city line. The block's
own "ты" is Maven, so "тебя зовут" read as her name and "его" would have
shown her the third-person form she must never use about him. They are
plain labels now.

Vikunja #394.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The block listed what she can do and ended with "nothing else". It sits
in front of the chat and general-knowledge prompts too, so that told her
to refuse the exact thing those prompts are for. Talking is now first in
the list, and the closing line limits ACTIONS rather than everything.

Also dropped the self-introduction from the knowledge prompt. It said
"Мавена, персональный ассистент" — a different name and a masculine
noun, right after the block says she is Maven and feminine. Identity
lives in the block now.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The phrasing fixture was 15 nudge cases, so every prompt change we
measured only told us about nudges. But the shared context block sits in
front of five prompts, and three of them — chat, note query, general
knowledge — had no scorer at all. Those are the long free-form replies,
where a persona break is most likely and where nothing could see one.

27 cases, nine per path. Nine rather than five because the nudge fixture
already cannot resolve a change smaller than about three cases, and a
per-path score off five would be worse.

Reuses the persona checks instead of copying them. Length, mood and
"no questions" are left out on purpose: these paths return no mood, and
a follow-up question is a feature in chat, not a fault.

The run refuses to score unless the model answers before and after it.
PhraseChat and PhraseQuery swallow model errors and return a canned
string, so without that guard a dead server produces a full report with
zero errors and a bad score — which reads as bad phrasing rather than as
nothing measured. Vikunja #397 is the real fix.
From a real reply in a nudge eval run: "Смотрите на его потребление
воды" is a plural imperative AND third person about him. Only the plural
printed.

Two separate faults. The check returned on its first hit, so the second
break stayed invisible and the failure read as milder than it was; it now
joins them. And "его" was not detected at all — looksVerb knows the
-й/-йте imperative but not the -те plural, so "смотрите" counted as the
person being talked about, which is what an antecedent means here.
pluralVerb already knows that form, so the antecedent test uses it too.

Third time a verb form has blinded this check. A fourth means it wants a
morphology table rather than another suffix.
The 0.8B answered about one chat turn in three with open reasoning as plain text, so no JSON ever closed and the fallback shipped "Thinking Process:" to the user. A grammar makes that output impossible.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The owner's call, 2026-07-31: a 0.8B model does not know enough about the
world to be useful without reading something. So she may now read external
sources to answer world questions.

What replaces the old rule, in all three docs:

- No telemetry, no cloud model, no third-party account. Unchanged.
- Local first: the Kiwix ZIMs on the box before anything on the network.
- External search is allowed but off unless configured, same as weather
  and telegram.
- His notes and facts are never search input. Only the utterance goes out
  — never the persona block, the history, or matched notes.

Docs only, no code.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The reply has to be Russian, but two of the phrasing prompts told her
what to do in English. Both are Russian now, in the same style as the
nudge prompt that already works better.

Also dropped the "you are maven, a self-hosted personal assistant"
line from both. The persona block right above it already says who she
is, so it was said twice.

The JSON part is unchanged.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Step one of letting Maven read instead of recall. No LLM yet.

internal/kiwix/client.go: search a local Kiwix server, parse the RSS
reply, hand back title + path + plain-text snippet + word count. The
snippet is the unit of context; a full article is ~100KB of HTML and
will not fit a 4096 token window.

internal/kiwix/retrieval_eval.go plus knowledge_v1.json: the 9 knowledge
questions from the phrasing fixture, each with hand-written English
keywords, scored on whether a wanted article comes back in the top 5.
Opt-in via MAVEN_KIWIX_URL, since CI has no Kiwix. No pass bar, the
number is the finding.

Result on the live mirror: 8/8 answerable questions hit, 7 of them at
rank 1. Retrieval works. Keywords are written by hand on purpose, since
Kiwix ranks by keyword and not by meaning, so a natural question fails.
A query-rewrite step is the next piece of work.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Two bugs, one symptom. A run of the talk eval produced replies that were
literally "{" and "{\n  \"" — those strings went out as things Maven said.

First bug: the parser could not tell "the model answered in plain prose"
from "the model started a JSON object and got cut off". Both came back as
empty, and every caller then shipped the raw text. Now an unfinished object
returns an error and each caller uses its own fallback instead. Bare prose
with no JSON in it still passes through, because small models do sometimes
answer that way and the reply is fine.

Second bug, and the actual cause: the grammar capped the response field at
400 characters. I measured it against Qwen3.5-0.8B at three different token
caps — 256, 768 and 2048 — and the reply came back exactly 400 characters
every time, cut mid-word. So the token limit was never what stopped it.
The bound is 1000 now, about six Russian sentences, still low enough to cut
off a repetition loop.

Token caps go from 256 to 768 on the chat and query paths so 1000
characters of Russian actually fits. The nudge path keeps its own cap; a
nudge is meant to be one sentence.

Note: cmd/mavend/replier_llm.go has its own copy of this parser with the
same bug. Left alone here so this commit stays small — that duplicate is
Vikunja #396.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
It scored 27/27 on a run where two replies were "{" and "{\n  \"". It only
tested that the string was not blank, so punctuation counted as content and
the worst replies of the run passed the first check.

Now a reply needs at least one letter, Cyrillic or Latin. Latin counts
because answers about ssd or vpn are legitimately part English.

Digits alone fail too. The same run answered "сколько варить яйцо
вкрутую?" with "15-16" — no unit, no words, and the wrong number as well.
That is not something she said.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Records all four configurations on the 27-case talk fixture, three runs
each: no grammar, plus grammar, plus Russian prompts, plus the truncation
fix. Composite, per-path and per-check, with the reproduce command.

The short version is that the plumbing got fixed and the score barely
moved. Grammar was the real win. Russian prompts helped a little and cut
latency by 5x. The truncation fix was necessary and bought nothing.

Also writes down three things that are easy to lose:

- The truncation cause was the grammar's 400-character bound, not the
  token cap. Measured at three caps, same 400 characters every time.
- Then I set the bound to 1000 against a 768-token cap and made it worse.
  The two limits have to agree.
- One run is contaminated and marked void: I ran an agent against the same
  llama-server, and the report still claimed zero errors while a third of
  the fixture silently answered "не знаю.". That is #397 and it is worse
  than filed — a busy server is indistinguishable from bad phrasing.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Nudge wording as data instead of generation. The wording lives in
internal/phraser/nudges_ru_v1.json (embedded), about 10 variants per rule:
water, meal, break, service_down, netdata_critical, routine:, morning:, plus
a contentless default. That JSON is long because it is data — the owner can
edit any line of Russian without touching Go.

The picker:
- random, but never the same variant twice in a row for the same rule
- deterministic when seeded (math/rand with an injectable source)
- fills {since} / {service} / {what} from the candidate, and skips any variant
  whose value is missing, so no raw placeholder can reach the piper voice
- {since} is spelled out in words ("полтора часа", "семь часов"), because
  "3 ч" is wrong in a Russian voice

Scores 15/15 on the existing nudge fixture, on every seed swept. Nothing is
wired yet — that is the next commit.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Kiwix ranks by keyword, not meaning, so a translated question finds song
and TV titles. This asks the resident model for the TOPIC instead: a short
English noun phrase, like a Wikipedia article title.

Locked down three ways, because a wrong query is silently wrong:
- A GBNF grammar, same idea as routeGrammar and responseGrammar. The
  reply must be {"query":"..."} with Latin words only. The JSON wrapper
  matters: this model always thinks out loud and this llama-server build
  ignores the thinking switch, so a bare word-list grammar just captured
  "Let me analyze this request carefully" for every question.
- max_tokens 32, since the answer is a few words.
- CleanQuery, which throws away empty, Russian and prose replies rather
  than passing them to Kiwix, and drops question words like "why" and
  "how much" that a keyword ranker cannot use anyway.

Client side only. Nothing is wired into the daemon or any config.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Same 9 cases as the retrieval eval, so the numbers compare directly:
hand-written keywords hit 8 of 8, this is what the model reaches on its
own. Reports the hand-written query next to the model's for every case,
because where the phrasing differs is the useful part.

Opt-in on MAVEN_KIWIX_URL + MAVEN_LLM_URL, like the other evals.

Result on Qwen3.5-0.8B: 3 of 8, identical on all three runs.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
DEPRECATION, flagged not asked: LLM-phrased nudges are no longer the default.
LLMPhraser.PhraseNudge now returns a hand-written Russian template. The model
still phrases chat, queries and reminders — only nudges moved.

Why: measured over many runs, Qwen3.5-0.8B wrote formal "вы" and plural
imperatives, used masculine self-reference, and invented facts and units
(90-95 seconds to boil an egg). A nudge is five words of known content, so
generation buys nothing and risks the persona every time. Templates score
15/15 on the nudge fixture, the model 11-13/15.

Nothing is deleted: the prompt, the fallbacks and the whole LLM nudge path
stay. Set phraser.llm_nudges = true in deploy/mavend.json to get them back.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Stock Qwen3-1.7B, not the CPT'd one — that training is still running. It won
on both fixtures we have, measured tonight on an otherwise idle box:

  routing, 77 RU cases, intent-only:  67.5%  vs  59.7%  for Qwen3.5-0.8B
  talk fixture, 27 cases:             20/27  vs  11-17/27

It also beat Qwen3.5-2B, which is 20% larger, on every routing column.

Two other things came with it:

n_ctx goes 2048 -> 4096. This is a Thinking variant, so reasoning tokens need
the room, and 4096 is the context every score above was measured at. Shipping
2048 would ship something nobody measured.

The doc now says not to bother with sub-500M models, because I checked and they
are not close. LFM2.5-350M routes at 5.2% — worse than guessing among 7 intents
— and answers "столица Франции?" with "Сторзит", which is not a word. The 230M
replies to Russian in Spanish. Their published IFEval and BFCL numbers are good
and they are all English.

Note the routing gain needs the LLM router actually wired on to show up. It is
still nil, so this commit buys the phrasing improvement today and the routing
improvement when that lands.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
Numbers behind the resident-model change, plus the answer to "could a 230-350M
model do this instead" — no, and the reason is worth keeping: LFM2.5's published
instruction-following scores beat Qwen3.5-0.8B, and every one of those benchmarks
except Multi-IF is English. In Russian the 350M invents non-words and the 230M
answers in Spanish.

Also fills the row TALK-EVAL-31-07-2026.md had to void for contamination, and
corrects a wrong call I nearly made: the 1.7B's 16s p95 looked like the reasoning
trace, but the 0.8B sits at 17s in every run and the 1.7B beat it twice out of
three. The long tail is shared and is not the Thinking block.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
The file ran two sweeps and the second one changed the resident model, but
the lede still opened with "Recommendation: keep Qwen3.5-0.8B". Anyone
landing on the file read the wrong conclusion and had to scroll 100 lines
to find that it had been replaced — and it contradicted CLAUDE.md, which
already says the resident model is Qwen3-1.7B.

Both sweeps are accurate, so nothing is rewritten. The lede now states the
outcome and the first sweep's verdict is scoped to what it actually tested:
it rejects LFM2.5-1.2B, which still holds. It never was a case for keeping
0.8B as the resident model.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
# Conflicts:
#	internal/store/migrations.go
PR #41 and #44 landed these two files unformatted, so the gofmt gate that
PR #12 added to `make test` failed as soon as both were on one branch.
Struct-tag and comment alignment only, no semantic change.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGeSZxh1DCtRxmFVSYVGvJ
kami merged commit c97aebf55a into master 2026-07-31 19:41:52 +02:00
Sign in to join this conversation.
No Reviewers
No Label
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: kami/Maven#47