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.
A restore now leaves out measurement tombstones deleted longer ago than
the tombstone retention. The nightly purge removes exactly those rows,
the delta feed answers cursorExpired for every cursor that could still
be served one, and the consolidation repairs look only inside the same
horizon. Younger tombstones keep their readers (a lagging paired client,
the undo of a recent delete, the repairs, the duplicate guard on
re-uploaded samples) and are written as before. The backup file is
unchanged. On a seeded account of 1.89 million readings whose 1.68
million tombstones were past retention, the restore on a slowed database
took 5.4 min instead of 15.
The rollup fold after the commit rebuilt five years of buckets under the
60 s request statement limit and could be dropped on a slow disk, leaving
the charts empty until the next boot. `recomputeUserRollups` takes a
`statementTimeoutMs` that scopes each aggregate and persist transaction
with SET LOCAL, and the restore passes its after-commit allowance.
Fixture tombstones in two round-trip tests were dated in July 2026 and
had aged past the retention; they are now relative to the run.
The test diffed the shared system tmpdir before and after the extraction,
and other test files extracting in parallel stage their files there too,
so it failed on their files rather than its own. TMPDIR now points at a
directory of the test's own for the call.
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.
A restore is refused before it deletes anything when its transaction limit
would run past three quarters of the job's expiry. At two hours that refused
a same-size replacement of about four million readings. The expiry is now six
hours, which admits ten million over ten million; a dead worker still frees
the account within minutes through the heartbeat, so the longer expiry does
not hold an account longer.
The refusal now names both counts, the readings in the backup and the ones the
account holds now (removed readings included until the nightly clean-up),
since clearing the latter costs time too.
run-case.ts now calls a provider for the live scenario run, and judge.ts
reads COACH_EVAL_SCENARIO_REPEATS. Both belong to the offline evaluation
harness, like the judge and its model knobs.
A turn never stores more than 128 KB of table JSON, but a restore accepted
any size in resultsJson or resultsEncrypted and sealed it into the database.
Both are now bounded: the readable form by its UTF-8 size, the sealed form at
twice that, above the ciphertext's measured expansion. A file this release
writes always fits; a larger value is refused like any other malformed field.
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.
A tapped follow-up chip goes away and a clarification choice disables as
the turn starts, which dropped focus to the page. Both now hand focus to
the composer, which stays focusable while a reply streams (read-only with
aria-disabled rather than disabled). The clarification question is
announced through a live region that is mounted before the question
arrives; the card itself no longer carries one.
The encryption backfill now also removes the two copies of a practitioner's
contact details that outlived the sealing. An appointment reminder's copy of
the address is cleared, since the reminder resolves it from the practitioner
when it fires. An address-book edit's audit row written before the sealing
kept the old phone number and address in its previous values; those values
leave the row and the changed fields stay named. Rows of a deleted account,
which no per-account job reaches, are scrubbed by the discovery pass itself.
A route whose readable geometry is a JSON null rather than SQL NULL had
nothing to seal but matched discovery every night. The pass now sets it to
SQL NULL so it drops out.
Sealing the practitioner's address left the appointment arm reading the old
readable column, which is now empty, so a booked visit's push lost the address
on booking, rescheduling and restoring a deleted visit.
The reminder no longer keeps its own copy of the address. The reminder tick
resolves it from the owning visit's practitioner when it builds the push and
decrypts it there, so an edit of the address also reaches a nudge that is
already booked. MeasurementReminder.location stays NULL on appointment rows:
booking writes none, a re-anchor clears one an earlier release wrote, and a
backup neither carries nor restores one. A checkup's own free-text location is
unchanged.
A withheld table now leaves a line saying why (module off, or the stored
tables could not be read) and a failed read shows an error row with a
retry, so the Data used count matches what is listed. A table copied from
an earlier answer names that answer's date under its title. The preview
shows the latest twelve periods of a time series, and the truncation note
says the latest rows are kept, which is what the server keeps. The title,
copy and view controls sit above the scrolling table with the title kept
as a hidden caption, and a reply with tables takes the full column width.
The step headers and the related-metric chip read as label: value in all
seven locales, so a metric name no longer sits capitalised mid-sentence.
The "compare with the period before" chip now needs a first reading
before the window's start, the same history check the other history chips
use, and is never offered on an all-time table, which has no period
before it: the table tool reads all time as the current period whatever
it is asked.
Tables are now projected outside the provider's failure path: a
projection defect costs the turn its tables and is annotated, instead of
refunding tokens already billed and dropping the reply.
The at-rest ceiling for a message's tables is applied once, when the
dialog is assembled, before anything streams. A table trimmed only at
write time used to show live and read back as withheld after a reload.
Ten scenarios, each in German and English, describe what a turn must do
with a question: which tool it calls and with which arguments, whether it
may read anything new, whether it asks back, declines a dose, offers
chips only for what the record holds, answers a chip without a provider,
offers to keep looking after a forced answer, and keeps an earlier
table's title out of the model's context. evaluateScenario grades a run
against them.
The deterministic suite drives POST /api/insights/chat with a scripted
provider; the tool loop, executor, table tool, chips, clarification
filter, method line, reply guards and stream run for real. Every check
also runs against a misbehaving model or a broken record and must fail
there.
route-followups.test.ts pins the front door for chips and answered
questions: validation, the rate limit and capability gate on a reuse
turn, the stored chip winning over anything the request adds, stale
chips and answers degrading to a plain message, and the continue line.
A total's weeks and months are now the sum of its days, labelled as a
total, and a level's stay a mean. All time is read as UTC calendar months
from the rollup tier's MONTH fold, or the same fold of the live table when
the tier has no rows, so a month no longer reads as a mean of daily totals
for a short history and a monthly total for a long one. Months are keyed
by their UTC start instead of being re-read in the user's zone, which
named every month one early west of UTC.
The chart's long ranges had the same shift: tier rows are now stamped at
noon UTC of the bucket's first day, which falls inside the bucket's own
week and month in every zone.
Step length and walking speed keep two decimals, loud-sound events count
as a total in whole events, and HRV counts a night measured by both
estimators once and charts the estimator the person mostly has. The model
summary names what a period's value is and gives a total's range total.
The outbound screen matched "Erhöhe auf 2,4 mg" but not "Erhöhe deine
Dosis auf 2,4 mg": the German change verbs had no slot for an object
between the verb and the target, unlike the English "increase your dose
to". The raise and lower patterns now accept one.
Asking for an earlier table "as a chart" went through show_result, which
builds the requested view, but collecting the turn's tables then reset
every chart to null and only derived a new one for fresh tables. The table
came back without its chart. A table shown again now keeps the view it was
given.
A damaged or foreign-key route ciphertext threw out of readRouteGeometry, so
one bad track answered the workout detail with a 500 and failed the activity
insight job. It now reads as no track with a warning on the request event, and
still never falls back to the readable column.
The line and bar charts under a Coach reply let recharts place the y-axis
ticks over an auto domain, and after thinning the ones that would collide
a blood pressure line could read 75, 90, 105, 135. The chart now computes
a nice domain and hands explicit ticks one step apart (1, 2, 2.5 or 5
times a power of ten); bars keep their zero baseline.
A table the answer referenced opens under the prose as its chart; the
others sit under Data used in the evidence disclosure and open as
tables. The plot is one image whose name points at the table view, two
series carry legend buttons, and the chart loads through the shared
chart runtime. A metric chart token for a metric a shown table already
covers is dropped.
When the tool loop forces an answer at its round cap, the reply leads
its chips with "Keep looking". Tapping it runs a normal turn, under the
budget gate as ever, with two server-written context lines: the question
that was left unfinished (at most 1000 characters) and what the forced
turn already read, each table by the name show_result takes, so the
continued turn shows those again instead of fetching them twice.
An answer is continued at most once. The continued reply records the
message it continues in metricSource.continuationOf and offers no
further "Keep looking", even when it is forced again; a chip on a reply
that is itself a continuation is refused.
Up to three chips, derived on the server from what the turn read: the
other view of a table with a chart, the period before, a year earlier,
a wider window, and a metric a discovered pattern links to. A chip is
offered only when its read was done this turn and the record holds the
metric; the year-ago and wider-window chips also need the record's
first reading to reach back that far. The model may name chips in a
FOLLOWUPS block, but only to reorder ones the rules already allow; every
label is rendered from the catalog.
A tapped chip resolves against the chips stored on the conversation's
latest reply. "As a chart" and "as a table" are answered from the
stored table without a model call or a budget reservation: the rows are
copied unchanged and the message is tagged providerType "reuse". Other
chips run a normal turn with a server-written context line. A chip that
is no longer current falls back to a plain message.
A setting turns the chips off (absent means on).
A table over time is drawn as a line for a level metric and as bars
for a total, with at most two series in one unit. Category counts get
bars, horizontal above six categories, the top eight and one other.
Fewer than two or more than 400 rows, or no numeric column, stay table
only. show_result's view now counts: table drops the chart, chart turns
an earlier day table into counts per range with a histogram (fixed
widths for the common metrics, Freedman-Diaconis otherwise, none below
ten values).
The line that resolves an answered clarifying question was already
appended to the system prompt on both paths; nothing proved it for the
no-tools (local model) path. Three route-level cases now do: the local
prompt and the tool-mode prompt both carry the server-resolved line, and
a turn without an answer carries neither. Removing the line from the
local prompt fails the first case.
The golden harness gains a coachMessage.findMany mock for this; it
defaults to no rows, so no existing transcript changes and the golden
snapshots are untouched.
The result-table caption test used the regex dotAll flag, which the
TypeScript target rejects (TS1501); it matches across lines with
[\s\S] instead.
The method line merged a repeated read by keeping the larger count. With
show_result that could let an earlier answer's count stand over the read
this turn just made. A fresh read of the same source, window and period
now always wins, whichever order the steps ran in, and the two are never
added. An entry that rests only on an earlier table says so ("from an
earlier answer"), including a reused table no step accounts for.
The new catalog string is in all seven locales.
get_metric_table now gets a step like every other read: the domain is
its validated metric, and once the table is in, the window, period and
granularity are the ones the table was built with (after the executor's
defaults). The count is the readings the table folds, taken from the
summary the server computed, so a read past the sixth table still counts.
A step whose call produced a table now names it (resultRef), for the
projected tables of the older tools as well.
show_result gets its own step. It reads nothing new: while it runs it
names no domain, and once the stored table is found it takes the source
of the read that built it and the name the copy carries in this turn. A
table that could not be shown is a failed step with the reason, never an
empty read.
The property test now drives show_result too and checks resultRef; the
coverage test gives each tool the smallest arguments it accepts instead
of passing empty arguments to a tool that requires a metric.
The table shows its caption, scoped headers, right-aligned tabular
numbers and a dash for a missing reading, twelve rows before Show all.
Copy writes TSV and an HTML table together, falling back to TSV, with
formula starts neutralised and separators flattened inside cells.
get_metric_table reads one metric as days, weeks or months through the
chart reader, cut in the user's zone and collapsed to one source per day,
with null for a period without a reading. The model reads a bounded
summary; the full table rides the turn, encrypted at rest. show_result
shows a table an earlier reply of the same conversation fetched, resolved
only against that conversation's own messages. Workouts, labs, sleep and
adherence results project into tables, earlier tables are named in the
context by metadata alone, and result:rN marks are stripped from the reply.
Marc-André Bombeck