Every interactive path, every import, MCP and Telegram checked an incoming
measurement against VALUE_RANGES before storing it. The provider sync writers
did not, so a decode slip or a glitched sensor landed a row the application
itself calls impossible, and every mean, median and personal band derived from
it inherited the impossibility. Filtering the read side hid the row. It kept
landing.
The gate is one module reading the domain the app already declares, applied at
two kinds of seam. reconcileExternalMeasurement is the shared identity writer
for Polar, Oura, WHOOP, the Withings sleep leg, the Apple export importer and
the batch endpoint, so gating there means a provider added later inherits the
refusal rather than having to remember a check it never knew about; the new
rejected_range verdict carries no row, which is what makes the compiler force
every caller to handle it. The five writers that hold Prisma themselves gate at
their entry, before the existence probe decides insert versus overwrite, so an
impossible reading can no longer replace a good stored value.
An out-of-band value is dropped. Not clamped, because a clamp invents a reading
nobody took. Not stored behind a flag, because a quarantined row is still a row
and every consumer would have to learn the flag. Not fatal to the batch, because
one bad sample must not strand the run's other readings behind a held watermark.
Absence reads as absence.
A silent drop, though, is indistinguishable from a provider that sent nothing,
so every drop is tallied on the wide event under measurement.range_rejected with
the metric, the provider and the edge it broke. Counts only: the reading itself
is health data and never enters the log line, and the direction plus the
metric's own declared bound is what a diagnosis actually needs. The tally
accumulates rather than replacing, because a provider gates once per resource
leg and the reconciler once per row. One warning per provider lifts the event to
warn so a drop does not hide inside a green line.
A structural guard enumerates all twenty measurement writers with the reason
each is allowed to write, so a new one has to answer the question before it
ships. scripts/diagnose-implausible-measurements.ts reports the rows already
stored, aggregated per metric and source with no value beyond the bucket's own
extremes. It repairs nothing: whether a stored impossible row is a decode slip
or a band that is too narrow is not a call a script gets to make.
A sync failure the classifier bucketed as `transient` had no exit. The park
test read `kind === "persistent"` and the persistent streak anchor was only
ever stamped by a persistent failure, so a streak could run for a thousand
attempts with the persistent counter at zero, the state frozen at
`error_transient`, and no ladder it could climb. `transient` is a hypothesis
about retryability, and the classifier defaults every timeout, 429, 5xx and
unrecognised response to it precisely because the wire cannot tell a reboot
from a permanent break. That default is right for one attempt and wrong for a
thousand: an unbroken streak retried hourly for a full day has refuted it by
experiment. Record it as `persistent` from there, which feeds the park ladder
that already exists — 48 h of unbroken failure end to end before an
integration stops attempting.
The counters could not accumulate anyway. `IntegrationStatus` is keyed
(user, integration) while Withings runs four legs on four crons and WHOOP four
resources on four more, and a success cleared the error, every bucket, the
streak anchor and the state whatever leg had recorded them. The ECG tick at
:41 wiped what the sleep tick recorded at :15 — including a reauth park — so
the strike ladder was reset several times an hour and the card painted a green
"connected · 12 minutes ago" over a data class that had stopped arriving. A
success now clears only what it is entitled to clear.
Two columns carry it: `failing_since_at` dates the streak, `failing_leg` names
its owner. The migration backfills the anchor for rows already in an error
state so a connection broken for weeks escalates on its next attempt rather
than restarting its clock.
Port the hard-fail ledger the Fitbit transport already carries. A revoked WHOOP
grant makes every resource short-circuit and return zero rows without throwing,
so the cycle read as clean: it stamped success, un-parked the reconnect state,
reset the failure streak, and the hourly cohort hammered the dead refresh token
on every tick. Record the three dead-token null-return paths on the ledger and
fold that verdict into both drivers — the full orchestrator and the per-resource
cron path that runs most often — so a dead-token cycle refuses the success
stamp and the park holds.
Return an imported-plus-failed verdict from the orchestrator and gate the
history backfill on it: an interrupted full-history walk now throws for a
pg-boss retry instead of being stamped complete, and a connection parked at
reauth returns without burning the retry budget. Thread the same verdict into
the one-shot sleep-timeline backfill.
Also catch a write-path throw on both drivers so a persistent measurement-write
failure surfaces on the status ledger instead of leaving a stale-green pill, and
widen the cycle overlap to the seven-day re-score window the other collections
use so a late-revised day's strain and energy are re-fetched rather than
stepped past.
Marc-André Bombeck