The three callers now read their sentences out of summary_ru_v1.json: the
plan lines in morning.Plan.FormatRU, the list and reason words in
tasks.FormatRU, and the habit readouts in memory.Profile.
Two behaviour_test assertions moved from substring to say.IsS, because the
habit gaps have variants now and a substring pins one of them. The
"по {day} у тебя обычно" variant was dropped on sight: the activities are
verbs, so it read "у тебя обычно тренируешься".
The persona scorer covers the family, and a new test asserts every gap
variant still says she has not seen enough rather than that he has nothing.
#496 asked to skip the veto when the question and the hit are in
different scripts, so an English question stops losing a Russian note.
Measured first: the fixture has no cross-language case, and en-hard-024
is an English question against an English note. Both proposed fixes are
no-ops.
What the veto actually does on the fixture, with the real embedder: it
costs en-hard-024 and buys ru-silent-029. Pass count is 22/32 either
way; false recall is 0/5 with it and 1/5 without. The two cases are one
lexical class, so no rule cheap enough for RecallAllowed separates them.
Accepts the loss and pins both sides in a test.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
queryMemory and queryNotes both gate on score alone, so both needed it. The
eval keeps its own copy of bestRecall — package main is not importable — and a
fixture that measures a weaker gate than the daemon runs flatters it, so the copy
moves in step and its test pins the new rule.
Measured on the held-out recall fixture with the real embedder: 17/32 cases pass
→ 22/32, false recall 1/5 → 0/5, answered after gate 18/27 → 17/27. The one true
recall lost is en-hard-024, an English question against a Russian note, where no
lexical test can help.
The score gate cannot separate the right note from an unrelated one: the
held-out fixture puts the right note at 0.791-0.890 and the must-be-silent cases
at 0.795-0.835, so a note about his slow network answered 'почему небо синее?'.
RecallAllowed adds a topic veto, and applies it only to a question that mentions
nothing of his. That restriction is the whole design: demanding a shared word of
every recall silenced four true recalls on the fixture to kill one false one,
because recall exists to find the note whose words he no longer remembers. A
question about his own life keeps the embedder as its only judge.
Seventeen markdown files at the repo root, twelve of them dated one-shot
reports sitting next to CLAUDE.md. That is why stale docs read as
current: nothing in the path said which was which.
Root now keeps CLAUDE.md and AGENTS.md. Living docs move under docs/
and carry a Last verified line. Dated measurements move to docs/evals/
ISO-prefixed, and are never edited after the day, so a newer number is
a new file. The senior review moves to docs/archive/.
Every reference was rewritten across markdown, Go comments, the Makefile
and the recall fixture. The touched Go packages still build.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Speaker profiles share the vector table with notes and facts. The doc
comment said reading them through Catalog is what keeps recall from
ranking a voiceprint. It is not. Catalog controls how speaker code reads
its own rows and says nothing about Search, which scanned every row.
What actually hid them was cosine returning 0 on a width mismatch, so a
192-dim ECAPA row scored 0 against a 384-dim query. Some x-vector
exports are 384-dim, and one of those would have surfaced speaker:kami
as a recall hit carrying the name of a person.
Both backends now skip the prefix in Search, and the prefix is one
constant in internal/memory so the store layer can filter on it without
importing internal/speaker.
Two more differences between the backends closed here. ByPrefix on the
in-memory store returned the stored metadata map by reference, so a
caller editing a returned Record edited the row, while the persistent
one unmarshals fresh. And the append to upsert change in Insert is a fix
in its own right, not only a speaker concern: any re-indexed id used to
leave a second stale copy searchable.
Found in review of #74.
IsDayPlanQuery fires on the token "планы" and its other-day list does not know
weekday names, so "какие у меня обычно планы по вторникам?" was claimed by the
day plan, which answered today's calendar stamped with today's date. The habit
source never ran. The matcher now declines any utterance ParseHabitQuery
claims, which keeps the decision out of the source table's ordering.
Two gaps in the same matcher. Sunday had only its dative plural listed, so "в
воскресенье" found no weekday. "по выходным" named days that no weekday word
matches, so it was answered with the whole-week profile. Both are recognised
now, and the weekend is read back as two days rather than pooled.
Found in review of #59.
Three defects in how the counted profile is read back.
The counting unit was the key the LLM invented. There is no allowlist and no
normalization behind it, so "я выпил воду" and "попил воды" landed as different
keys, split one habit into two, and dropped both below the two-day threshold.
Keys now go through an alias table in behavior_ru.json before they are counted.
An unglossed key is quoted rather than recited as a verb, because "обычно ты
выпил_воды около 09:00" is not a sentence.
The typical time was a median of minutes since midnight, which is wrong for
anything that straddles midnight. Bedtimes of 23:40, 23:50, 00:10 and 00:20
gave 12:00, on the one activity most likely to cross the boundary. It is now a
circular median, and when the observations span more than half the clock she
names the habit without a time instead of inventing one.
The rest is wording. Profile.Since was computed and never spoken, so "обычно"
was an unfalsifiable claim; the overall read-back now says over how many days
of records it holds. The no-data weekday answer said "у меня пока нет ничего
постоянного" about a question concerning him. And "quiet" was a bare prefix in
the non-behavioural list, so any future self-fact key starting with those five
letters would have been dropped.
Found in review of #59.
Seven fixes, each answering a line comment on the stack.
**Weather no longer invents Moscow** (PR 50). extractWeatherLocation returned
the string "Moscow" when he named no city and voice.weather.default_location
was unset — a made-up answer presented as fact, which is the one thing maven
must never do. It returns "" now and the query path says it does not know.
**Digest statuses are a defined type** (PR 50). DigestStatus string plus the
three constants, so a rule name cannot reach the status column.
**Quiet-mode negation is not adjacency** (PR 53). The OFF list carried
{"не","тих"}, an adjacency pattern, so "не надо тихий режим" missed OFF, hit
the ON pattern {"тих","режим"}, and asking for quiet mode to stop turned it
on. Negators are scanned over the whole utterance now, with the two ON phrases
that are themselves built on "не" excluded. "тихий режим выключи" works too,
which it did not before.
**Pattern stability uses a median band** (PR 54). max/min over the extremes
asked whether every gap resembles every other gap, so 7,7,7,7,20 — four clean
weeks and one holiday — was thrown away at a ratio of 2.9. Each interval is
now tested against the median and 70% must be in band, and the reported
interval is the median of the in-band ones, so a holiday no longer drags a
weekly habit to "every 9.6 days". The reviewer's 5,8,10,3 is still rejected.
**The weekday profile stops reciting everyday habits** (PR 59). "What do I do
on Saturdays?" answered "you drink water" — true, and useless, because it is
equally true of every other day. Activities that are habits on six or more
weekdays move to Profile.Everyday and are read back as daily habits instead of
as an answer about that day.
**Russian phrase tables move out of Go** (PR 59, PR 61). The behaviour glosses
and weekday names, and the task capture/urgency/list vocabulary, are now
behavior_ru.json and task_phrases.json, embedded with go:embed. Single-binary
deploy is unchanged; wording edits are no longer source diffs.
**nginx template stops taking nginx down** (PR 52). Two host-side failure
modes, both plausible causes of today's crash. The $connection_upgrade map is
fatal when duplicated, so it moved to its own nginx-upgrade-map.conf with a
grep-first note. And `listen 10.42.0.1:80` fails with EADDRNOTAVAIL when wg0
is not up yet, so nginx exits on a reboot that beats WireGuard — the header
now documents net.ipv4.ip_nonlocal_bind.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TrVSBKe3RFDF4fGYKWYQnX
Maven can now be told who someone is. She cannot yet tell who is speaking,
and this commit is careful to say so rather than pretend otherwise.
What works: profiles are enrolled from several deliberately recorded samples,
listed, and deleted. They live in the existing memory_vectors table under a
"speaker:" id prefix, so there is no migration; what that needed was a wider
interface than memory.Store, hence memory.Catalog with ByPrefix and Delete.
Delete is the load-bearing half — a voiceprint someone asked to be rid of has
to actually go, and a search-only store cannot do that. InMemoryStore.Insert
became an upsert by id to match what the persistent store already did.
What does not work, and why it is not faked: there is no speaker-embedding
model on this box. Sixteen ggufs in /mnt/hdd1/llms, all text; no ECAPA, no
x-vector, no titanet, no wespeaker, no .onnx anywhere under /mnt/hdd1. So
newSpeakerEmbedder returns nil, internal/speaker falls back to
speaker.Disabled, Identify answers ErrDisabled, and the daemon logs which
half is off at startup. The plan's "simple MFCC + GMM" floor is refused in
the package comment: MFCC cosine distance detects channel and loudness as
much as voice, and a biometric that is confidently wrong writes false claims
about named people into his memory. A bad floor is worse than none here.
Refused as well, and the reason is in enroll.go's doc comment: the plan asked
for unknown speakers to be enrolled on first interaction with a TTS "кто
это?". There is no request shape in the protocol that could express that.
Taking a biometric of whoever walks past the microphone does it to guests who
are not party to the exchange, and a synthesised question into a room is not
consent from whoever answers.
Authority: enrolment is AuthStepUp, because it is a deliberate sit-down act
that writes a biometric of a named person and never something done by voice
mid-conversation. Deletion is one rung lower at AuthWrite, deliberately
inverting the usual pattern — getting rid of a biometric must never be the
harder half. Listing is AuthRead and never returns the vectors themselves.
Off unless configured: no speaker block means the three methods answer
ErrUnknownMethod, so a default box has no wire path that takes a voiceprint.
make build and make test pass.
Vikunja #255
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TrVSBKe3RFDF4fGYKWYQnX
Behavioural memory, narrowed on purpose. internal/memory/behavior.go builds a
profile out of self-facts — distinct days per weekday, median time of day — and
reads it back in RU; router.ParseHabitQuery finds the weekday deterministically;
a `habits` query source answers the question.
Three things the plan doc asks for are deliberately absent, and the doc now
records why:
- The profile is COUNTED, not LLM-generated. A 1.7B asked to summarise a year of
habits writes fluent claims about the owner's life that no row supports, and a
wrong claim about him is the most expensive kind of wrong maven can be.
- No cached profile fact, so no "update on fact write" machinery. It is
recomputed on the question; a cache that can disagree with its own rows is two
truths.
- No proactive daily plan nudge. A dispatcher proposal at 08:00 every day is the
definition of a nag. The path from "she noticed a pattern" to "she acts on it"
already exists in internal/pattern with the proposal queue on /routines, and it
goes through him.
A one-off is not a habit: an activity needs two distinct days before she will
call it usual, and until then she says she does not know yet. Only self-facts
count — env rows are the world, config rows are her own tuning state. The typical
time is a median so one 03:00 outlier cannot move a morning habit into the night.
An unrecognised fact key is read back verbatim rather than glossed into something
she made up.
The source sits before "calendar" in querySources, and its matcher requires a
habit marker, so "что я делаю в среду?" still reaches the calendar — answering a
question about this coming Wednesday with a statistical average would be
answering a different question.
Verified: make build and make test both exit 0.
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 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
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
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
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
claude
kami