Build the memory block for the providers the turn sends to, send it on
both paths, keep the facts it recalled in the owner's trail, and offer a
health fact the background found as the answer's one note. Every turn
enqueues the memory refresh; the job's debounce and singleton key decide
when it runs. A remembered fact points back to the stored user message.
Unit tests for the turn budget and the no-progress brake, the budgeted
loop (every stop reason, the round-by-round ledger, repeated calls,
questions, comparisons, reasoning and provider state), the trail recorder
and its screen, clarifications through the tool and the brake, the
comparison table and chart, the trail column and its owner-only read, the
reply frames, and the turn's frames end to end through the route. An
integration test proves the trail against a real database. Eval
scenarios cover a multi-round why question, a period comparison with its
chart, a question asked through the tool, and keep the charts the
existing table scenarios already showed.
Take the effective reasoning level from the route rather than resolving
it in the turn; the turn only caps it when the operator pays.
Give the Coach one statement of how an answer is shaped, in the German and
English bodies and on every other locale: the first sentence answers, one
carrying number with its unit and window, the person's own baseline, an
assumption in one line, at most one next step and no habitual closing
question, 40 to 120 words, and a closed list of filler it never writes.
- Brief is 30 to 70 words in one paragraph, detailed up to 220 in three.
- Motivational-interviewing moves only when the person opens a change topic,
at most one, never as a closing question; the confidence ruler only on an
explicit plan request.
- The few-shot examples are rewritten to model the voice.
- Result tables: show the numbers whenever they carry the answer and span
more than one value, a table for exact values or up to twelve like rows,
one figure alone only for a moment.
- Follow-up chips are the answer's only "what next".
- A voice grader for the eval (length, filler, emoji, next steps, closing
question, warm-up opening) plugs into the golden set as criteria; five
voice cases in German and English.
- A guard holds both prompt bodies to the rules and every example answer to
the grader, each check broken once on purpose.
What a person told the Coach never reached a tool-mode turn: facts and plans
rode the snapshot's memory block, which tool mode does not send, and the
extraction ran only past twenty turns and failed to parse on every
non-Codex chain, because the status chain runs in JSON-object mode while the
prompt asked for a bare array.
- Memory block: WHAT YOU KNOW ABOUT THIS PERSON, built per turn behind the
wire egress check (nothing is read on a refusal), fenced as data, at most
1,500 characters, never trimmed before the inventory. Confirmed facts
first, then the newest; active plans with a server-computed progress line;
reminders the person asked for. Medication facts stay out while the
medications module is off. Building it stamps last_used_at without moving
updated_at. buildCoachToolRequest and buildCoachProviderPrompts take it as
an optional memoryBlock; the snapshot memory keeps only the narrative and
trend recall.
- remember_fact: preferences, goals and context are saved at once
(source coach, source_message_id set); anything that names a condition,
limitation or medication, by category or by wording, is only proposed.
A fact must come from the person's current message; one note per answer.
- The person's answer reads the proposed fact from the encrypted trail on
the assistant message, never from the request.
- propose_plan writes a proposed plan; the tap activates it with its review
window counted from the tap. One per answer, three open at most.
- Progress: one sentence per active plan from the person's own day series
and units, also offered to the briefing (two lines at most, behind the
briefing capability). The plan review and the progress line read a step
plan as daily totals instead of per-sample means.
- Background extraction asks for {"facts":[]} / {"plans":[]} and reads
either shape, runs once a conversation has been quiet for 30 minutes, and
stores health facts as proposals offered once at the next turn.
- Unanswered plan and fact proposals lapse after 14 days in the reminder
sweep.
- Memory list API: POST saves one of the caller's own messages (the
remember button), PATCH edits a fact's wording, GET reports source, origin
and last use and never lists a waiting proposal.
- A structural guard keeps the egress check ahead of every read and the
callers of both builders to the turn and the briefing.
Replace the fixed three tool rounds with a budget fixed by who pays:
120k tokens, 150 s and 12 rounds on the person's own plan, key or local
model, 40k, 90 s and 6 rounds when the operator pays. Cached input counts
at a tenth. The final answer is reserved from the start, so a turn whose
budget, clock or round cap runs out answers from what it read. A
no-progress brake stops a turn that repeats calls or reads nothing new,
and a repeated call is never run again.
The day's ledger is booked round by round: round one and the final
reserve when the turn starts, each further round right before it runs,
every round settled as it returns, the unused reserve given back at the
end. A round the daily ceiling refuses ends the turn with an answer.
Every phase goes out live as an activity frame (thinking, fetching,
summarising, asking, remembering, planning, the stop, the answer), with
screened reasoning titles and summaries; the structure rides the
provenance and the text is sealed in trail_encrypted, served by
GET /api/insights/chat/{id}/messages/{messageId}/trail. Tables go out as
interim results as soon as they are read.
Add ask_clarification with a server brake, remember_fact and
propose_plan over the memory contract, model-free decision turns,
compare_series with the compare chart, eight tables per message,
assumptions with change-assumption chips, a reasoning level per turn and
at most two concurrent turns per person.
Add the shared contract the reasoning release builds on, with no change
to how a turn runs today.
- Schema: operator reasoning controls on app_settings (0376), the
encrypted trail text on coach_messages (0377), and source,
last_used_at and source_message_id on coach_facts (0378).
- Provider contract: CompletionParams.reasoning and onReasoning,
CompletionResult.reasoning and providerState, AiMessage.providerState.
- Reasoning levels, thinking budgets and the background-job table in
src/lib/ai/reasoning/levels.ts; coachPrefsJson.reasoning, absent
meaning medium, salvaged field by field like every other pref.
- Stream: activity, memoryNote and planProposal frames, done.stop and
done.withheldResults, result.interim, a compare chart kind, eight
table refs, three more clarification kinds with an assumed choice,
the change_assumption chip, and assumptions, activity, stop,
memoryNote and planProposal on the provenance. Zod mirror, type
parity and the native-client key rule cover every new frame.
- Chat request: memoryDecision and planDecision, parsed and not yet
acted on.
- Stub contracts for the memory functions and the activity recorder,
inert until their implementations land.
- Backup, restore and rotation carry the new columns; a fact's message
reference is resolved against the restored messages like its
conversation reference.
- Message keys for the trail, memory, plans, assumptions, replies,
interim tables, comparison charts and reasoning settings in all seven
locales, each reachable from a catalog and checked both ways.
The summary of a condensed block was taken over the block's mix of daily
values and weekly means: over a year the newest fourteen days weighed as
much as fourteen weeks, min and max were extremes of means, and change was
one day minus a weekly mean.
Each series block now hands the budget pass its daily values over the
whole read, computed only when the block is condensed. The summary states
the first and last day, the lowest and highest day, the mean of the days,
and the change from the first calendar week to the latest one. A block
without daily values gets no summary rather than one of what is left. The
coverage note says how the summary is taken.
The condensing step read every block as a value series: daily rows under
value, weekly rows under mean and count. Sleep nights carry minutes, so a
condensed sleep summary ended on the last weekly mean weeks before the
newest night; adherence rates fold by doses, not readings; glucose keeps a
series per measurement context; workouts are a sessions list.
Condensing now knows each block's shape: sleep reads minutes, adherence
folds rate weighted by total doses, glucose condenses every context, and a
workouts block keeps its newest sessions beside the whole-window rollup.
The Coach's figure check ran before the MCP branch, so an MCP client was
told retrieval_failed for a block it used to receive, and the plain-text
renderer then said no data was on file. Any single-source build also
condensed its own block, which reached MCP through the source snapshot's
fallback to the full builder.
The figure check now runs after the MCP branch, and condensing is an
explicit option of the snapshot build that only the Coach tools set (it is
part of the cache key). MCP output is what it was before.
A block trimmed to its unit and freshness stamp passed the figure check,
because asOf.daysAgo is a number. The check now skips the asOf and scope
fields, so that shape answers retrieval_failed like any other block
without a reading.
Inside a chat turn every tool read lands on the turn's shared full-source
snapshot when its window matches. On an account with many synced series
that build passes the prompt budget, and the budget pass collapses blocks
to a bare unit and then to an `{ omitted }` marker. get_metric_series
sliced the marker out and answered `present: true` with it, so the model
was told the readings exist and was given no figure. The same held for
sleep, glucose, compliance and workouts.
- The snapshot result names the blocks the budget pass cut. A tool whose
block was cut in the shared build reads it again from its own source at
the same window.
- A read scoped to one source condenses its own blocks last and never
swaps them for the marker: it keeps a summary (first, last, min, max,
mean, change) taken before anything is cut, the newest daily values,
weekly means folded into monthly means, and names each cut.
- get_metric_series never answers present with a block holding no figure;
that read reports retrieval_failed instead.
- A chat series read carries the readings it counted, so the step shows
them, and on a long or condensed window a note on what it covers. The
tool description and a new tool-mode rule tell the model to call
get_metric_table itself for longer history instead of asking.
- The screen-reader comma between step parts is unselectable, so a copied
row no longer reads "Weight · ,all time".
MCP output is unchanged: the source-snapshot equality suite stays green.
Reading a stored reply back drops a view chip whose table the answer
already shows with its toggle, but the resolver reads the raw stored
chips. A tap from a client that still shows such a chip is answered, by
design: the chip was a valid offer when it was stored. The test pins both
halves for "as a table" and "as a chart".
A table the answer displays with a chart renders its own chart/table
toggle, and the "Show as a chart" / "Show as a table" chips under it
repeated that control. The server no longer derives a view chip for such
a table; a table the answer only used, folded under "Data used", still
gets "Show as a chart", since that brings it into the answer. A reuse
turn shows its table with the toggle and so offers no view chip either.
The rule is one predicate in a leaf module, applied where chips are
derived and where a stored reply is read back, so a reply stored before
the change offers the same chips as a new one on every client.
The title encryption backfill moved each Coach conversation's title into
ciphertext with a plain Prisma update, which stamps updatedAt. The Coach
panel orders and groups conversations by that column, so threads untouched
for months showed up under Today. The memory summary refresh did the same
in the background.
Storage-only rewrites now carry the row's own updatedAt and match on it, so
a turn that lands meanwhile wins and the row is retried on the next pass:
conversation titles, practitioner contacts, measurement and mood notes
(where updatedAt is the sync cursor), inventory notes, and the mood-note
script. The summary refresh writes through raw SQL. A guard freezes both
rules: every backfill write on a model with @updatedAt names the column or
is listed with the reason the stamp is wanted, and every conversation
write outside the person's own actions keeps it.
The derived block shortened each engine's window to the lookback limit,
so a 7-day limit produced a readiness computed over 7 days instead of the
30 the ring on the dashboard uses, under the same name. A block whose own
window does not fit is now left out, as the lookback rule says, and the
coincident-deviation flag follows its 30-day baseline the same way.
The GLP-1 block sent every dose change with its decrypted note whatever
the lookback limit said. Dose changes are entries, so only those inside
the limit reach the prompt; the dose in force stays as the current dose.
The earlier-tables block names each stored table with its window and row
count. The list was filtered by module and exclusion only, so a table
stored before the limit was set or narrowed still told the model how much
history it covered. It now answers to the same range test show_result
uses, and the history-reach guard checks both the filter and that the
call site passes the limit.
The saved Coach window only seeded the snapshot and each tool's default.
A tool could still read back to all time, the availability probe reported
counts and means for history outside the window, and several blocks read
fixed windows of their own (labs 12 months, correlations 180 days, the
coarse tail, the all-time extremes) whatever the setting said.
coachPrefs.defaultWindow is now enforced as a limit through one module,
history-reach.ts: the turn context clamps the client window, the snapshot
clamps its window and gives every fixed-window block either the shorter
window or, when the block's own window cannot fit, leaves it out. Every
retrieval tool clamps its window, an earlier period or a stored table that
reaches past the limit answers outside_reach, and the probe reads only
inside the limit and names older history as outside_reach without any
figure. Follow-up chips that would read past the limit are not offered.
The default (allTime) is no limit and reads as before. MCP clients pass no
limit and keep their own window arguments.
Adds the twelve-month option, rewrites the picker text to state what is
read with the numbers the readers use, shows the same options in the
managed-profile settings, counts cluster members with a singular, and
drops the opacity on switched-off rows that put their text below WCAG AA.
get_metric_series (and get_metrics, the metric resources and the review
prompts, which fan out over it) answered with one section of a Coach
snapshot, but built the whole snapshot to get it: the all-type feature
extraction, the derived wellness layer, trajectory, correlations, memory,
GLP-1, cycle, illness, labs and visits. On a seeded account with about
215k measurements a resting heart rate read ran 208 SQL statements and
blocked the event loop for 175 ms to return under 2 KB.
The MCP surface now asks the executor for a source snapshot. It runs the
same prelude (prefs, module gates, admitted sources, cutoffs, now shared in
snapshot-prelude.ts) and the same block builders over the requested
source's reads only. The core clinical aggregates come from extractFeatures
scoped to the one block, whose bulk read keeps its all-type cap so the rows
a block sees do not change. A section large enough for the full builder's
prompt-budget pass to trim, and the environment cluster, go to the full
builder, so their results are the old ones by construction.
The integration test compares the tool's serialised result with the full
builder's section and grounding for nine sources over four windows, the
get_metrics batch and the metric resource, on a seeded account dense enough
to put the feature row cap into play.
The insights comparison snapshot took its current and baseline pulse figures
from a reading AVG, and the Coach's data inventory told the model a reading
mean of the whole pulse history. Both now read the mean of the user's local
days, each the mean of its hours' means, through the day-mean helper. Every
other type keeps its aggregate.
The feature extraction, the signals of the day, the status cards, the graded
series, the Coach snapshot, the mood correlations and the report's trend arrow
each still averaged pulse over its readings, so a dense workout hour decided
the day and a workout day the window. They now read a pulse day as the mean of
its hours' means and a window as the mean of its days, through the day-mean
helper. Counts, extremes and spreads stay over every reading; every other type
is unchanged.
A medication taken now and then (#1024) keeps one row and gains course
rows: calendar spans, both ends inclusive, an optional encrypted note, no
schedule or dose of their own. Courses never overlap and every course but
the latest has ended, so the latest course is always the one the
medication's startsOn/endsOn project. src/lib/medications/courses.ts is
the only writer of those two columns for a medication with courses; it
writes under a per-medication advisory lock and re-projects in the same
transaction, so reminders, the next-due line and intakeActionable keep
reading the columns and stay right.
- Migration 0368 adds medication_courses and backfills one course per
medication window (an end-only window starts on the creation day, or
on its end day when that lies earlier); a medication without a window
stays course-less. Idempotent, data-preserving.
- GET/POST /api/medications/{id}/courses and PATCH/DELETE
.../courses/{courseId} (owner-only); create seeds the first course
from a window; the medication PUT edits the latest course and refuses
a window that would break the rules (medications.course.*).
- List, detail, create and update carry courses, courseCount,
previousCourseEndedOn and canStartCourse, resolved on the server.
- Compliance expands over the span of all courses and drops the slots
between them; every compliance read selects the courses. A rolling
cadence is not anchored by an intake from a previous course.
- The card names the last course's end and the running course's number;
the detail Lifecycle group lists the courses with start, edit and
delete. The Coach compliance block, the MCP resource, the doctor
report window and table, the clinician view and FHIR (one
MedicationStatement per course with effectivePeriod and status) read
the courses.
- Backup carries courses per medication and the restore writes them and
re-projects; an older file derives the one course its window
describes. Wipe plan, encrypted columns, rotation and OpenAPI follow.
The Coach illness block now carries the last 14 days of the person's own
symptoms (name sanitised for the prompt, count, highest intensity, last
occurrence; at most ten, most frequent first), and exists when there are
symptoms but no episode. The prompt describes the field in English and
German. The MCP get_illness_recovery result leaves it out in this
release.
Every active symptom definition enters pattern discovery as an outcome
channel SYMPTOM:<id>, labelled with its decrypted name (the data label
the interpretation already prefers over the key). The daily value is the
day's highest intensity in the person's zone; days without an occurrence
count as 0 only between the first and the last logged day. At most eight,
the active-definition cap. The SYMPTOM: prefix resolves to the illness
module in the surface map, so the family leaves the matrix with the
module, and the discovery guard lists the new fetcher.
The cycle block's headline contrast labelled weight in kg and the
temperatures in °C, and left a glucose contrast without any unit. The
metric table, which the person sees and whose summary the model reads,
listed weight in kilograms, temperatures in degrees Celsius, walking
speed in metres per second and distance in metres.
The cycle contrast now converts both means and the gap into the
reader's units (the gap by the factor alone) and names the unit in its
sentence. The table resolves each column's unit and decimals from the
reader's preferences and converts every folded period value once; the
snapshot carries the preferences it was written in, so the table and
the snapshot of a turn always agree.
Refs #1067
The Coach snapshot carried weight, body composition, temperatures and
walking speed and distance as stored: kilograms, degrees Celsius, metres
per second and metres, with no unit on the block. The evidence-block
rules then listed kg and both glucose units as the units to cite. A
reader on pounds, Fahrenheit or miles was answered in the units they do
not use.
Every value block now converts its timeline (and the weight block its
aggregate and coarse tail) into the reader's units after aggregating on
the canonical value, and names the unit on the block; the grounding
scalars stay canonical. Workout distances arrive in kilometres or miles
with the unit named once. The prompt in every locale tells the model
that a block's figures are already in the reader's unit and takes the
evidence-block unit from the block instead of a fixed list.
Refs #1067
The reference grounding printed the fasting glucose bands in mg/dL and
the body-temperature band in °C, and the system prompt described time
in range as 70-180 mg/dL. For a reader on mmol/L or Fahrenheit the
model was handed one quantity in two units: the glucose block in
mmol/L, the band beside it in mg/dL.
The grounding block now takes both unit preferences and prints the
general, the diabetes goal and the temperature band in them, while the
placement is still classified on the canonical value. The glucose
block carries the time-in-range band in its own unit, and the prompt
points the model at it instead of naming a unit.
Refs #1067
A guard that compares a sweep against an allowlist, or against "none of
these may exist", passes when the sweep comes back empty. walkSourceFiles
already refuses a short walk; this adds the same refusal for what a guard
does with the walk.
- requireFloor(label, found, floor) throws below the floor and refuses a
floor of zero.
- scanSourceMatches(root, pattern, { fileFloor, matchFloor }) walks, matches
and floors both the files read and the hits found.
Floors added, pinned at the current counts:
- coach_usage owner columns: the clauses it actually examined (6).
- measurement reference index: the foreign keys it extracted (2).
- job outcome fact inventory: job files (130, floor 120) and fact keys
(399, floor 350).
- tone harness: assessment prompt modules (11).
- coach locale default: turn pipeline modules (15, floor 12).
- locale integrity: message bundles (7).
- settings card header coverage and settings anatomy: the walked cards.
- coach metric scope inventory: insight route sources.
- side-effect drift guard: both enums populated, since two empty keysets
are equal.
The pill-mapper guard walked with globSync, which skips dot-prefixed
directories; it now walks through walkSourceFiles like the rest.
`instant.toISOString().slice(0, 10)` and parsing a key back as UTC
midnight are the idioms behind every day-boundary fix this year: the
evening west of UTC and the small hours east of it land on the
neighbouring day. A new rule, healthlog/no-utc-day-key, refuses them in
src/ outside src/lib/tz and tests, and refuses a disable directive for
itself that is not next-line or gives no reason.
The time zone module now names every legitimate form: userDayKey for a
person's day, dateOnlyKey / dateOnlyAtNoonUtc / dayKeyAsUtcMidnight for
date-only values and day labels, daysBetweenDateKeys / weekdayOfDateKey /
isCalendarDateKey / statedDateKey for key arithmetic and parsing.
Of the 133 sites the rule found, 84 are gone. Most moved onto a named
helper with no change in behaviour; these changed behaviour:
- cycle calendar, insights, Coach snapshot and prediction refresh start
their lookbacks at the first instant of the user's day
- a model-transcribed date like "June 10, 2026" no longer lands a day
early in a process east of UTC (documents and lab scans)
- a custom report range covers exactly the picked days in the profile
zone instead of starting the evening before west of UTC
- date-of-birth pickers cap at the user's today, not the UTC one
- an allergy onset date is stored at noon UTC like other date-only values
- the report's cycle summary, the Coach's usual BP range and the intake
projector's rollup recompute use the user's day
- vaccination dates render as dates instead of the previous evening
49 remain, each with its reason on the line above: 39 UTC by design (15
file name stamps, rollup and object keys, the UTC-day score engines) and
10 marked baseline. A guard test requires a classified reason on every
directive and holds the baseline to a ceiling that only goes down.
The snapshot cut the last injection, the dose history and the projected
next injection on UTC days and compared getUTCDay() with the schedule's
local weekday. An evening injection west of UTC landed on the next day,
and east of UTC 'tomorrow' was today. Day keys now come from the user's
zone, which the Coach snapshot already resolves and now passes in.
"As a table" used to drop the table's chart, so the reply offered no way
back to it. The copied table now keeps its chart and carries view: table,
which the panel opens on the table (the chart one tap away on the toggle)
and the chips read to offer "as a chart". show_result with view table
does the same. The field is optional on the result metadata; a client
that ignores it shows the chart first as before.
The Coach's recent nights carried the wake instant in UTC as their date,
while the sleep table under the answer keys a night by the local day it
ends on. East of UTC the two named different days for the same night.
The snapshot now gives the night's own key.
The recheck and follow-up rules now ride the tool-mode prompt only once
the conversation holds a table, or from the round after the turn produced
its first one. The clarifying-question rules are shorter and English for
every locale, like the other instructions to the model; the German copy
of them and of the recheck rule is gone. A turn without tables sends
about 1 300 characters less on every round.
The golden transcripts change only in the system prompt's hash and
length (4 509 to 3 187 characters; 4 728 to 3 280 for the German turn).
Table refs, step ids, chip ids, a chip's anchor ref, clarification choice
ids and a reused table's source are now validated against their exact
server-minted patterns when a stored reply is read back. Some of them are
written into the model's context on a later turn (the earlier-tables list,
the "keep looking" fetched list), so a stored row edited by hand can no
longer carry text into a prompt through them.
show_result now passes the same scope gate a fresh read does before it
decrypts a stored table: a metric the person has since excluded from the
Coach, or one outside the conversation's scope, answers
unavailable_in_scope and nothing is sent. The list of earlier tables in
the context drops those metrics too, so the model is not told they exist.
The exclusion (the person's own list plus switched-off modules, and no
adherence when medications are excluded) is the snapshot builder's, now
one exported helper both paths use.
"Keep looking" no longer copies the unfinished question into the
server-written context lines. The question stays in the transcript as the
person's own turn; the line only points at it. Lifted into the system
role, whatever the question said would have carried the system's
authority.
A turn now reads its conversation's newest messages once and shares the
rows between the chip, clarification, "keep looking" and repeat-question
checks, instead of three identical reads.
Earlier tables keep their m<k> names once a conversation outgrows the
window a turn loads: the index counts the assistant messages before the
window too, so a name the model read on an earlier turn still points at
the same table.
The backup validation now reads the per-message table ceiling from
results/refs, which imports nothing but types. Reading it from
results/project pulled the tool executor and the message store into the
snapshot builder's import graph, and the free-text sanitiser guard then
walked files that never build a prompt.
The grounding ledger derives differences only between figures of one
payload, so a correct "up 4 mmHg on last week" drawn from two separate
table reads was withheld as unverified. A table of the period before or of
a year ago now reads the current window too, and its summary carries the
current figures and the change (delta and percent change) beside the
earlier ones. A wrong delta is still flagged.
The live evaluation now captures the tool calls a generation makes, with
their arguments, and hands them to the judge. runScenarioLive puts a
scenario to the model in the chat turn's own context (system prompt,
tool-mode and dialog addenda, DATA INVENTORY, EARLIER TABLES, transcript)
over the real tool catalogue, answers each call with a stand-in result,
and reads the reply through the turn's clarification and chip parsers.
runScenarioJudge grades each scenario a few times with the same
evaluateScenario the deterministic suite uses and reports a pass rate per
scenario; the server-only scenarios and checks are left out.
Like the existing judge it runs only with COACH_EVAL_API_KEY, from the
nightly workflow's non-blocking step, and never throws.
Marc-André Bombeck