A rolling cadence yields only lastIntakeAt + N, so a slot the auto-take
pass never records (before the switch-on floor, past the catch-up
window, inside a pause, skipped or removed by the person) held it still
for good. Step the cadence forward by N past every such slot, keeping
its rhythm, and feed the stepped anchor to the card, dashboard and
schedule-tool next due so they show the slot the pass will record.
The token ledger keyed every day on UTC while the daily briefing and the
other per-day generations run on the account's local day, so east of UTC
the first generations after local midnight were refused against the
previous day's spent ceiling. The budget day is now the calendar day in
the account's time zone; each request still resolves it once and books
its reservation and reconcile on the same row. The medication extraction
reads relative phrases against the same local day, the operator's spend
card shows the operator's own day, and the limit copy says the budget
resets at midnight in the person's time zone.
Since the backfill resumes per collection, a slice that ends before the
whole history is in threw "backfill incomplete" so pg-boss retried it,
which logged job.run.failed at error and grew the queue's failure count
although nothing had failed. A slice that finished at least one more
collection now queues the next slice on the admission queue under its
own singleton key and completes. Only a slice that finished nothing new
still throws, so a backfill that is actually stuck keeps its error. The
import lock keys on the import identity without the slice number, and a
slice that finds the import already stamped walks nothing.
Every raw-SQL clock read against a timestamp column without a zone now
writes NOW() AT TIME ZONE 'UTC', so passkey and login challenges, rate
limits, provider pauses, coach usage stamps, tombstone and measurement
windows no longer pass through the session time zone even where the
per-connection pin cannot apply. The regression-window bounds stay naive
UTC timestamps instead of being turned into timestamptz. A structural
guard keeps bare NOW() / CURRENT_TIMESTAMP out of raw SQL outside a short
list of timestamptz targets. The integration test sets the database
default to Europe/Berlin and drives challenge claims, session expiry, the
native sign-in handoff freshness check and the rate-limit window through
the app's client, plus the challenge claim through an unpinned client.
Self-hosting docs describe the pin for an existing or shared Postgres.
Stored times are UTC in timestamp columns without a zone, and Postgres
reads now(), CURRENT_TIMESTAMP and offset-free timestamptz parameters
through the session time zone. On a database whose default is not UTC
every expiry check ran off by the offset. The Prisma pool, pg-boss, the
lock and the maintenance connections now send TimeZone=UTC as a startup
option, or SET it on connect when the startup options are disabled for
a pooler. A boot check reads SHOW TimeZone and warns on anything else.
A history backfill carried one access token through walks that outlive it.
After an hour every request answered 401, which classifies as a revoked
grant, so the account was parked at error_reauth with a refresh token that
was still valid, and every later run skipped it until a manual reconnect.
Running beside the hourly sync, the backfill also pushed the account over
Google's per-minute quota, and each 429 failed its whole collection.
- Walks take a token source that renews the access token before it expires
and once more when Google refuses it mid-walk.
- A 429 is waited out (Retry-After, RetryInfo, else a doubling backoff) and
the request repeated; it is never a reauth.
- error_reauth only short-circuits a run when the token endpoint refused the
grant (needsReauth). A connection parked by a data request's 401 refreshes
on its next run and clears itself when that run completes.
- One run per account at a time: syncs, the backfill and the sleep repair
share an advisory lock. The hourly poll skips a held account, a manual
sync answers 409 sync_in_progress, the backfill waits a bounded time.
- A backfill attempt records each collection it finishes cleanly
(backfill_progress) and the next attempt resumes with the rest.
- The incremental watermark is stamped with the cycle's start, not its end.
Overlapping sleep segments that share a stage and an end share the natural
key (type, measuredAt, source, sleepStage). A re-score that moved one row
onto a sibling's slot threw P2002 and failed the sync on every run.
- A batch keeps one reading per natural key, chosen the same way each time
(longer segment, then smaller externalId).
- A move that still collides is settled by reconcileExternalMeasurement,
which retires the stale row, instead of failing the sync.
- Updates run before inserts, and the key-format rescue no longer re-keys a
row this batch is moving.
- The sleep replace-by-window protects every session's fresh rows, so two
overlapping sessions no longer tombstone and resurrect each other hourly.
Refs #1194, #1195
Next 16.4.0, pg-boss 12.37.1, the AWS SDK 3.1146, @napi-rs/canvas 1.0.10,
@node-rs/argon2 2.2.2, @simplewebauthn/server 14.0.3, TanStack Query and
Virtual, lucide-react, mmdb-lib, nodemailer, otpauth, pg, radix-ui,
react-day-picker, react-hook-form, tailwind-merge and yaml.
pg-boss moves its schema from 42 to 45 (three additive migrations, nullable
columns and one partial index, no table rewrite); the worker applies them on
start as before. The retention test pins the installed version, so it moves
to 12.37.1 after re-checking every source line it asserts.
zod stays on 4.4.3. 4.5 requires seconds in `z.iso.datetime()` and 4.6
rewrites nullable unions in the generated JSON Schema into `type` arrays,
which reshapes the OpenAPI document and the MCP tool schemas. Dependabot
skips zod 4.5 and later until that move is made on purpose.
The fold leftovers were told from a person's deletions by syncVersion = 1,
but raw samples are raised in place before a fold takes them (the collision
merge, a changed Health Connect value). The repair then recomputed windows
without those samples and could overwrite a correct mean, also in windows
one run had folded whole.
A tombstone of fold age now counts as a leftover when it carries
syncVersion = 1 or shares its deletion instant with such a tombstone of the
same day; anything else of that age is ambiguous and leaves its window as
stored. Only a window taken in two parts (fold runs at least an hour apart,
or a run and later live samples) is recomputed.
Once the repair has been through an account, the purge may remove class-A
leftovers inside the horizon, so the folds' re-fold of a stored day or hour
leaves it as stored, and the purge takes the fold lock per account. The
fold-lock transactions get an explicit timeout and maxWait, since the lock
wait counts against Prisma's five-second default.
- Place a day at the home only from homeSince on, through the same
resolver the nightly fetch uses; a home without homeSince places no day.
- End a run at the first request the feed refuses (429, 5xx, network)
with status error, no follow-up and no pg-boss retry.
- Fetch the weather of days without a row before their air quality, and
keep the weather rows when the air-quality request fails, so no paid
request is asked for twice.
- Create no row for a day the archive answer leaves out.
- Re-read the module, the account switch and the operator switch before
every range.
- Check the history's own ceilings in the budget's locked transaction
together with the charge, instead of reading usage first.
- Read the days with entries once per chain and hand them to the
follow-up in its payload.
A background job fetches air quality, pollen and UV for every earlier
local day with a reading, mood entry, dose, symptom or workout: after the
update (boot and nightly discovery), when the home is set and when air
quality is switched on. A day with a weather row gets its air quality; a
day without one gets both, at the covering location period or else at the
home, also before the home was set. Days the source does not reach (before
2013 in Europe, before August 2022 elsewhere) stay empty and are not
counted.
Pending days are grouped by coarse location and cut into ranges of up to
90 days, joining days at most 14 apart, newest first. A run holds itself to
300 of the account's 400 daily calls and half of each instance window,
stores its progress on the user (migration 0391) and sends a follow-up while
it makes progress; a budget stop waits for the next discovery. The
Location & weather card shows the progress.
The air-quality client now asks pollen and dust from 2021 and skips the
request outside Europe before CAMS global starts.
The fold repair and the regular folds rewrote an existing daily or hourly
stats: mean from the samples still in the table. Fold tombstones are purged
75 days after they were written, so for an older window those samples can
be a fragment: a re-upload after an anchor reset, the rows of one later
fold run. A mean over them replaced a stored mean that had been taken from
the whole day, and on old days it moved a lot.
A stored mean is now recomputed only for a day that starts at or after the
tombstone horizon (foldConstituentHorizon): every fold tombstone of such a
day was written after dayStart + the fold threshold and is still inside the
retention, so the live samples plus the fold leftovers are the whole day.
That is the last 75 days for daily means and, since dense days fold 90 days
late, the 75 days before the raw window for hourly means. A window also
needs at least one fold leftover; without one the rows in hand are not
known to be all of it.
Beyond the horizon:
- The repair leaves the day as it is and counts it per type as
windowsSkippedBeyondHorizon.
- The daily-mean consolidation and the dense hourly fold leave a day or
hour that already has a live stats: row exactly as it is, the stored
mean and its live samples alike (daysLeftAsStored, hoursLeftAsStored).
Up to v1.41 the fold overwrote such a row with the mean of the live rows
alone, and the first v1.42 cut with the mean of those plus whatever
tombstones were left; both are a mean of part of a day. Dropping the
live rows instead would lose samples that may not be in the stored
mean, so they stay as synced. They are few, the folded_window guard
refuses new ones, and the nightly pass walks over them without writing.
A day or hour without a stored mean still folds as before.
Inside the horizon, a pass that recomputes a stored mean soft-deletes the
live samples it took in as fold leftovers instead of deleting them, so a
second pass computes the same mean from the same samples. Before, a fold
that deleted them left the repair a fragment, and the repair then rewrote
the mean from the tombstones alone.
The repair, the mean consolidation and the dense fold now take a per-account
advisory lock after the restore lock in every day transaction, and the
repair reads the day's samples inside it, so the two never compute the
same window from two views of its samples or write the same row at once.
The repair also stops at the fold boundary: a day the fold has not
finished is the fold's to complete.
The backfill route shared the analytics-read bucket and sent every
explicit range without a key, so one account could queue two-year
backfills until the instance-wide daily budget was gone for everyone.
The route now allows three backfills a rolling hour per account (429
environment.backfill_rate_limited) and one queued backfill per
twenty-minute slot, keyed on the account (409
environment.backfill_pending). Every fetch made for an account is also
charged to that account's own share of the daily budget, one twentieth
of it, so a refused request ends the run as budget-blocked and the
nightly run continues.
The daily-mean consolidation and the dense hourly fold cut at
`now - threshold` instead of a local day boundary. A day was folded in two
runs and the second run overwrote the stored mean with the mean of the
later part alone: every mean-type day, and the boundary hour of every
heart-rate, HRV and SpO2 day, plus the derived resting figure and the DAY
rollup of such a day. The Health Connect importer split its raw window at
the same unaligned instant.
- Fold up to the start of the account's local day the threshold ends in
(`foldBoundary`), for every consolidation pass and the Health Connect
importer's raw/hourly split.
- A window that already has a live `stats:` row is refolded from every
sample of it, live rows plus the fold's own tombstones; with none left
the stored mean stays.
- `folded_window` refuses a raw sample only when its local day lies before
the aligned boundary and a live `stats:` row of its source covers it,
also within reach of the sample after a timezone change.
- A deletion inside a day the fold had not finished is class B.
- Add a one-time, resumable fold repair per account that recomputes each
hourly and daily mean from its remaining samples and writes only values
that differ, recorded in `measurement_fold_repairs` (migration 0389).
- The compaction-tombstone purge waits for the repair per account, and the
boot sends no purge run before one account is repaired.
- The restore keys live `stats:` ids by source, and writes compaction
tombstones back from a file older than the account's repair, resetting
the repair.
The nightly dense-intraday fold scanned Apple Health rows only, so the
per-minute heart rate a Health Connect import writes for the last 90 days
never aged into hourly means and every new export added another quarter
of raw samples.
The fold now walks each fold source in turn, Apple Health first and
unchanged, and folds Health Connect samples into hourly stats: rows under
their own source, deleting the raw rows outright. The folded_window rule
covers both sources for the hour window and looks a covering row up under
the sample's own source; the day window stays Apple Health only. The
Health Connect importer applies the same rule to raw samples, so an
export that still holds folded minutes does not put them back.
- Load each settings section on its own: the section route imported all
twenty-one statically. A preload starts the chunk beside the record
gate's read, and the loading state paints the settings card shell.
/settings/[section] goes from 611 to 403 KB gz.
- Take Zod out of the shell every route shares: the account-access parse
run on each boot is plain TypeScript now, held to the published schema
by an equivalence test over good and broken blocks. The canonical unit
table moves to its own module, so the display transform no longer
pulls Zod in either. Layout entry 199 to 136 KB gz; /insights 369 to
304, /medications 428 to 365, /labs 400 to 337.
- Open a saved Coach conversation at its last answer: an instant jump
instead of the smooth scroll that lost the end, then follow late
layout (charts, tables) for a few seconds until the reader scrolls.
- Show the document archive as a month grid of preview tiles, two
across on a phone and three or four on a desktop, with the month
headings kept.
- Add the admin card for the measurement table maintenance: a status
read on the maintenance route, one calm confirmation, and the run's
state while it is queued, running, finished or stopped.
- Re-base the bundle budget: /insights at 304, /settings/[section]
watched at 403, aggregate cap 4335 with the reason written down.
- Publish timezone, glucoseUnit and a features block on the account
payload; glucoseUnit reads as the closed pair the contract names, and
features.trackIntake tells a separately released client that
medications take the record-only switch.
- Refuse a new dose for a record-only medication with
medication.intake.notTracked on every dose-writing route; the bulk
route reports it per entry as intake_not_tracked, and correcting an
already resolved event stays open.
- Answer passkey sign-in failures by name instead of a 500:
passkey.challenge.expired, passkey.unknown, passkey.response.invalid
and passkey.verification.failed, each with an audit row.
- Carry one account-wide adherence figure in meta.aggregate on the
batched compliance read, weighted by expected doses, and accept
?days=30|60|90|180|365 for a report window.
- Accept ?days=30|90|180|365 on the mood insights read so stability
covers the period shown, and name the window it covers.
- Keep only the reminder that announces a dose off APNs when the phone
owns medication reminders; follow-ups for a dose still open are
pushed again, marked by a followUp flag the reminder cron sets.
- Document which batch validation issues belong to an entry and pin it
with a test that splits and resends a batch; publish rejectedFields
items as ValidationIssue.
Fetch the Open-Meteo air-quality feed (CAMS, domains=auto) with the weather:
17 hourly variables folded into local days by fixed rules (means and maxima
only with 18 of 24 hours, ozone as the highest 8-hour running mean, UV only
with the midday hours), a value the feed did not cover stored as null, never
as zero. Requests are cut into 90-day chunks, newest first, and chunks before
mid 2022 ask for the eight pollutant and index variables only. An air-quality
failure never costs the weather row; the day stays unfetched and the nightly
gap fill (up to 90 days per account, newest first) asks for it again.
Every Open-Meteo request asks an instance-wide budget kept in Postgres first
(500 a minute, 4000 an hour, 8000 a day, weighted by the hosted pricing
rule). A refused request is not sent and the run ends as done with
budget_blocked. OPENMETEO_AIR_QUALITY_URL and ENVIRONMENT_AIR_QUALITY_DISABLED
are honoured; the account switch lives on PATCH /api/environment/preferences.
The home, each dated location period and each stored day keep their coarse
location sealed as one value under healthlog/environment-location/v1. The
writers write the sealed value only, the readers open it and fall back to the
readable columns for rows the free-text encryption backfill has not reached,
and the backfill now seals all three and clears the readable columns. The
backup carries the location readable in a portable file and sealed in a
disaster-recovery file; a restore seals either under the receiving key.
Three correlation channels join the environment family (PM2.5, the ozone
8-hour high, the day's highest pollen count), read through the averaged
same-day window and left empty while air quality is off. The Coach and MCP
gain get_environment: module-gated, clamped to the lookback limit, no
coordinate or place name anywhere, with provenance and the attribution
lines. The settings section gets the switch, the coverage and the newest
day's values; the dashboard shows a quiet chip on a day with high pollen,
a warm night or very poor air. Copernicus and Open-Meteo are credited next
to the values.
Upload the ZIP the Android Health Connect app writes and import its
`health_connect_export.db` in the background, the same way an Apple
Health export is imported: streamed to disk with a SHA-256, capped at
1 GiB, one import per account, a job row of kind `health_connect` that
the shared reconcile cron and staging sweep now look up on its own
queue.
The worker extracts the database (4 GiB ceiling, zip-bomb checks) and
opens it read-only with `immutable=1`, `trusted_schema=OFF` and
`query_only`. A view in place of a record table refuses the file,
columns are found by name, and schema versions below 9 are refused.
Records are read in batches of 500 and aggregated in SQLite: steps,
energy and distance as one total per day from the app the person's
Health Connect priority list ranks first, heart rate per minute for 90
days and as hourly means before that, sleep stages with the unknown
stage counted only in sessions without real stages, workouts, cycle
days, water and micronutrients. Apps whose integration the account has
connected are left out, and a reading already stored under another
source is skipped. Rows carry source HEALTH_CONNECT, `hc:<uuid>` or the
`stats:` ids, so a second import of the same export writes nothing.
Both staged files are removed on every way out; the job result holds
counts per type and per app only. Settings gain a Health Connect card
with the export steps, progress and the outcome, in all seven
languages, and the import is documented in
docs/integrations/health-connect.md.
A catch that drops what it caught is where the silent nightly failures
started. Every `catch` and `.catch()` under src/lib/jobs now rethrows,
fails the job, writes to a ledger an operator reads, counts into the run's
failure facts, hands the error back to its caller, or logs at warn under a
stable action. The enqueue helpers that swallowed a failed send (briefing,
status, narrative, morning digest, document index / summary / thumbnail,
workout insight, boot-time sweeps), the Coach nudge's provider fallback,
the workout and reaction-line cleanups, the advisory-lock release, the
backup attempt bookkeeping and the cycle reminder's dedupe reads now leave
a warn line instead of nothing or an info annotation.
`error-level-discipline-guard` holds the rule. The catches that stay
silent are listed with their reason (the pgboss readers' documented null on
a web-only deployment, the key-ring trial, the restore drill's per-account
result, an already-deleted Telegram message); a new silent catch fails, an
entry that no longer matches fails, and each entry is shown to be
load-bearing by removing it.
A Coach turn cut short by its token, time or round budget writes
`coach.turn.budget_stop` at warn.
The nightly briefing warm completed every night and read "all-failed",
because one run event held every account's annotations and the last
account won. Each candidate now runs in its own wide event and the run
reports every outcome through `reportJobRun`, so ok, dead sign-in,
withheld and failed accounts show side by side.
A dead Codex refresh token threw a bare error with no status, so the chain
read it as a network failure: the health ledger recorded hard_failed, no
"reconnect" hint appeared, and every night tried the same token. The
refresh now throws `CodexRefreshError` with a 401 for a dead grant
(invalid_grant / invalid_refresh_token on a 400 included), as do a missing
or undecryptable stored credential. The ledger records auth_failed, the
chain marks the primary credential expired, and the nightly pass reports
the account as auth_failed without queuing a retry.
A briefing the outbound screen withheld (a causal claim) failed outright,
and the 45-minute provider retry then asked the same question eight times
a night. The generator now makes one bounded repair that names the broken
rule and keeps the repaired payload only if it passes the same parse,
grounding and screen; a withheld outcome queues no retry.
A provider that keeps failing is paused: from five consecutive hard
failures the skip window starts at 30 minutes and doubles per further
failure, capped at a day. A paused provider is walked last and tried again
when the window lifts; one success resumes it. The operator card shows how
many accounts it is paused for and until when. After a 5xx from a
non-canonical OpenAI-compatible endpoint the client asks the free model
listing, and a configured model the endpoint no longer offers is named in
the error, on the hop line and on the card.
Also:
- The tiered context's live day band read `measured_at AT TIME ZONE tz`,
which treats the stored UTC as local time; a far-from-UTC reader's
evening readings landed on the next day. It converts from UTC first now.
- Google Health intraday heart rate re-reads two hours per sync instead of
a day; daily summaries and sleep keep the day (#1023).
- The first filtered daily read that answers 400 and triggers the
snake_case retry is recorded as a note, not an error (#1023).
- The single derived-score read goes through the derived cache with
stale-while-revalidate, like the batch beside it.
A wide event carries one action and the last annotation wins, so a hop
that failed before the next provider answered, one failed account among
six, or a caught error the code carried on past was either a meta key on
someone else's info line or nothing at all. `emitSignal` writes one extra
line under a stable action and level, joined to the surrounding event by
request and trace id; `logCaught` / `caughtAs` are its warn-level shape
for a catch that degrades on purpose.
Every failed pg-boss run now writes `job.run.failed` at error with the
queue and how many runs of it in a row have failed, counted from pg-boss's
own rows (`readConsecutiveRunFailures`). `reportJobRun` takes a fan-out
pass's whole candidate list: an all-failed run is `job.run.failed`, a
partial one `job.run.partial` at warn, and the run reports `all_failed` in
its facts so a completed night with nothing done still counts toward the
streak. The repeated-failure alert keys on that one action and number.
Retention is per queue now. `failureReaderRetention` refuses a queue
retention under 96 hours (the 72-hour reader window plus a day), and a
guard refuses any `deleteAfterSeconds` that is neither that helper nor a
literal at or above the floor.
A refusal the request was meant to get (any error carrying a 4xx status)
is recorded at warn instead of error. Expired-token 401s were most of the
error lines and hid the real failures between them.
The nightly folds soft-deleted every raw Apple Health sample they folded
into an hourly or daily stats row and kept it for 75 days. Nothing ever
read those rows as deletions; they only held a re-upload of the sample
off, and they were most of the measurements table and its indexes.
Ingest now refuses such a re-upload directly: a raw sample of a dense or
mean type, old enough to have been folded, whose hour or day a live
stats row covers, is a duplicate with reason folded_window. The batch
route and the Apple Health export import apply the same rule. With that
in place the dense retention and the daily-mean consolidation delete the
raw rows outright, and a new compaction-tombstone-purge job removes the
backlog: 5,000 rows a batch, one account at a time under the restore
lock, a short pause between batches, resumable because the predicate is
idempotent, queued at every boot and continuing itself while it makes
progress. A person's own deletions are never in its predicate. The sync
feed leaves compaction tombstones out and a restore does not write them
back; all three classify through the same function.
The batch route answers an exact re-sent sample without a transaction:
one read over the batch's sample ids, live or deleted, decides what the
reconciler would have decided, at the same point in the loop, so every
status, reason, counter and the cross-source merge stay as they were. A
batch of duplicates only no longer queues PR detection, a reminder sweep
or the detection audit row. Caches are evicted only for a foreground or
manual sync or a batch that wrote a hand-entered reading, and marked
stale otherwise; the targets cache follows the same split.
Also:
- POST /api/admin/maintenance/measurements queues VACUUM (ANALYZE) and a
REINDEX INDEX CONCURRENTLY per index, largest first, on a connection of
its own; it refuses while the purge runs, drops invalid leftover copies
and stops at the first failed rebuild. Runbook in docs/ops.
- The pure volume queues keep their finished jobs for 96 hours instead
of seven days, never less than the failure window plus a day.
- work_mem is set per app connection from DATABASE_WORK_MEM (16MB), also
when the statement timeout is disabled.
- The PR-detection fallback tick re-scans only accounts whose readings
or workouts changed since the previous tick.
- Type discovery for the status refill walks the live-rows index as a
loose index scan instead of grouping every live row.
- The host metrics chart polls every five minutes, like its sampler.
- The batch route records per-type HealthKit arrivals.
Land every file the v1.42 work shares before any of it is built, so the
parts built side by side never edit the same file.
Schema and migrations 0379-0386: the HEALTH_CONNECT measurement source,
ImportJob.kind, the air-quality columns and the per-user switch on the
environment tables, sealed environment locations with nullable plaintext,
managed-profile handover links, the per-type HealthKit arrival ledger, lab
unit-conversion provenance, and user-defined vaccines. Every new column is
annotated for the column-reader sweep.
HEALTH_CONNECT reaches every exhaustive list: the source labels, the
read enum (not the write allowlist), the source-priority ladder cap, now
the number of sources rather than a fixed eight, and the write-path smoke
test. RMSSD gets its own priority class and a 300 ms ceiling. The batch
syncTrigger gains `manual`, with its label on the Apple Health card.
Queues for the Health Connect import, the compaction-tombstone purge and
the measurement-table maintenance are registered with stub handlers;
host-metric sampling moves to every five minutes. Environment and work_mem
variables reach the compose whitelist, the production example and the
manifest. The proxy admits the handover link and redirects it at the edge,
and the Health Connect upload bypasses the proxy body limit.
The new routes answer 501 behind their final authentication and are
published in the contract; backup, wipe and key rotation classify the new
models and columns; the seven message bundles carry the base copy. Guards
that list release-pending keys and columns fail once an entry is wired, so
those lists empty themselves before the tag.
Also re-anchor a login-throttle fixture to the wall clock, where its
expiry is compared, and re-base the aggregate bundle cap on a measured
12 KB of new locale strings.
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.
With signals of the day resting on today's readings only, the briefing
had to be written after those readings, and it was only written at the
04:30 Berlin warm, the post-sleep morning refresh, or a page open on a
day-old cache. A morning blood pressure or weigh-in never became a
signal, and since an earlier day's briefing is no longer served, a
reader in Los Angeles or Seoul had none for most of the day: the Berlin
warm lands the previous evening there, or at noon.
The snapshot read every surface goes through (dashboard, iOS, digest,
morning push) now asks for a briefing for today at two moments: the
first read of the reader's local day with no briefing written today,
and the first read after a signal metric's first reading of the day
that the briefing has not read. Each reason is claimed once per user
per local day in a Postgres bucket and remembered in-process, so later
reads cost no query; no write path needs a hook.
A claim enqueues one today warm: immediately for a new day, after a
five-minute window for readings, so a burst (weigh-in, then blood
pressure) folds into one job. The warm writes the comprehensive
briefing only, past the one-hour freshness gate but under the failure
backoff, the daily forced-warm cap (two per 24 h) and the content-hash
gate.
Worst case per user per local day: one enqueue per reason, at most
eight (the day plus seven signal reading types, measured over a day of
reads every two minutes); at most two forced comprehensive attempts
under the existing cap, each at most three completions (the answer, a
JSON retry, a grounding retry), so six provider calls at the ceiling.
A warm whose data has not changed costs at most the existing
once-a-day paragraph re-roll, inside the same cap.
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.
Taking the weekly preview from the writer dropped the schema check the
old scanner ran at the end, so a copy that violates the schema was
described as restorable. The collector now checks each section in its
JSON form as the writer shows it, an array section and a bulk table one
element at a time, plus the schema's one whole-document rule (a
canonical copy names every measurement by id). A copy that fails gets
no stored preview, and the preview route reads the copy itself, as it
did before. Nothing is kept beyond one element.
Measured on the preview memory test under --max-old-space-size=524:
held after the producer 3 MB with and without the check, raw peak
heapUsed 293 MB with it (335 MB without). The test now loads the
backup schema before its baseline: a module loaded once per process,
not memory that grows with the record.
An account whose last backup attempt never came back was kept behind
every other account for as long as its attempt stayed open. A pass that
regularly stops on its time budget never reached it, so it never got a
copy again. It is now held back only until a later run has started on
another account; if its record kills the process again, its attempt is
the newest once more and it goes last again. A crash loop still leaves
the other accounts their copies, and the account itself is attempted at
least every second run.
The drill now reads every account's newest off-host object but still ran
on the default fifteen-minute expiry, without a lock and with two
retries, so a large instance could see it killed, run beside itself, and
page every failed account three times. It now carries a four-hour expiry,
stops between accounts on its budget and reports the objects it did not
reach, holds a pass lock at the binding, and is not retried.
backupPayloadSchema is built when validations/backup.ts is evaluated and
holds about 12 MB of heap. The restore job, the stored-backup readers and
the export writer import that module, and they sit on the boot path, so
every server process built the schema at start whether or not it ever
touched a backup.
The schema version, the compatibility check and the per-section summary
move to validations/backup-summary.ts, which never builds the schema. The
call sites that validate a file (upload, restore, the summary and download
routes, and the streamed reader's element check) load backup.ts with a
dynamic import at that moment; everything else takes only types. The
OpenAPI registry keeps its static import, since only the spec generator
loads it.
After warm-up on a production build under --max-old-space-size=524, the
used heap drops by about 29 MB, and the schema no longer appears in a heap
snapshot. A structural test refuses a new static value import of the
module outside the registry.