The past-days backfill placed every day without a row at the current home,
also before the home was set. Weather and air of a place the person was not
at would flow into correlations, the Coach and the day view as fact. A day
before homeSince is now filled only when a location period covers it; with
no homeSince stored the home still covers every day. Days without a place
are not counted in the progress.
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.
Since correlations are tested on seasonally adjusted residuals with an
effective sample size, an account with history sees far fewer
relationships, which reads as a defect without a word about it.
The correlations response now carries findingsBeforeSeasonalAdjustment:
true when the record kept a discovery pattern first stored before the
change, one indexed lookup on the pattern store. An account created since
can never qualify. The mood "What stands out" card shows a short note
for those records, closable, remembered per viewer like the AI setup
hint. Copy in all seven locales; OpenAPI regenerated.
Deleting a conversation hides the row and sends the DELETE only after
the six-second undo window. The pending delete lived in a timer alone,
so a reload, a closed tab or an installed app sent to the background
inside the window dropped it without a request, and the conversation
came back. When the window did run out, the hide filter was released
while the paginated history still held the row, so it showed again
until the refetch landed.
Every surface that deletes a conversation (the Coach panel, the
conversations page, the settings memory card) shares one hook, and the
fix lives there:
- a scheduled delete commits early when the page is hidden, on
pagehide and on unmount, and the request is sent with keepalive
- a committed row leaves every list cache, head list and every history
search variant, before anything else changes, and stays hidden
- deletes not yet confirmed are kept in sessionStorage; the next page
hides them and sends them again, so its list read cannot bring the
row back before the old page's request lands; a 404 counts as gone
- a failed delete restores the lists and says so instead of the row
quietly returning
- Undo after the commit reports false, so the rail does not reopen a
thread that is gone
A find-then-delete pair let two deletes of the same conversation both
find the row, and the slower one failed with a 500. A client that
re-sends a delete it could not confirm now gets 404 for the second
request, which it reads as already gone.
The integration test covers the thread, its messages and its document
links going with it, the document itself staying, the list no longer
carrying it, a foreign account getting 404, and the concurrent pair.
Every bucket with a reading gets its point; one with fewer than three
readings is drawn hollow. A short gap (one quarter, two months, four
weeks) is bridged with a dashed stroke between the two real points; a
longer one leaves the line open. The selection bar and the value table
for screen readers name each bucket's readings and say "no value" where
there is none. The phone chronicle names each bucket once, at its newest
month. The value menu takes up to six lines. The legend's causality line
is gone with its keys in all seven bundles.
The server picks the bucket per zoom and sends it as `bucket`: quarters
for an `all` window over two years, months for `year` and a shorter
`all`, weeks for `quarter`. Each point carries the readings behind its
mean as `count`; a bucket without a reading stays absent. Means remain
the mean of day values, so pulse-like types keep the hourly-mean day.
`values` accepts up to six series.
From 1280 px the day sits beside the page, and its control at the top
left took it away entirely: after a page change there was no way back to
the day short of finding it again. The control now hides the day the way
a sidebar collapses. A narrow edge stays on the right, holding the date,
with a button that opens the last day again on this page or any other.
Focus lands on that button when the day collapses and on the day's
heading when it comes back.
The last open day is remembered per browser, bound to the account and
the record it was read in, and removed at every session end. The Coach
button keeps clear of the edge as it does of the open day. Sheets below
1280 px and on phones close as before, and Back still closes them.
An account that never picked an arrangement now sees the months flowing
across the width. The layout PUT stores only the fields the person chose,
so an untouched field keeps following the default while a saved choice
stays.
The "Running that day" list carried the profile facts (smoking status,
alcohol pattern, shift schedule) as running items, titled by their raw
codes and counted "day n" from the date the fact was filed. A person who
quit smoking years ago and entered it last week read "day 13". Those facts
describe the person, hold no start the person gave, and are not part of a
day; the lifestyle section and its running kind are gone from the day, its
index and the model-facing get_day.
The same false count reached a medication filed without a start date: its
creation date stood in as the start. `since` is now null there, with no
day count, and the entry's date only bounds which days show it. A life
event's category is worded in the reader's language instead of its code.
The repair covers the last 75 days of daily means and the 75 days before
the raw window of hourly means; older days stay as they are. The ops page
names the two new run counters.
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 boot script interpolated its two colours; it is one string literal now,
checked against THEME_COLOR by the script's own tests. The retired-route
guard escapes every regex metacharacter including the backslash, and the
two markup tests read text by walking the string instead of stripping tags
with a regex.
A batch without a trigger only marks the dashboard snapshot stale since
v1.42, so a snapshot an earlier spec left for the account painted the day
final before this journey's own night counted. Foreground evicts.
Since an empty mood list carries the one add control and the header drops
its own, the journey clicked a header button that unmounted as the list
resolved. Start the profile without entries, assert the empty-state add,
then the header add once an entry exists. Both carry a data-slot.
The band writes one strain row per physiological cycle, stamped at its
start, and a bedtime either side of midnight puts two starts on one
calendar day. The strain page counted those rows as days and drew an
extra sparkline point the recovery chart does not have. Collapse the
rows to one per local day at the day's mean, the chart's own value.
The tooltip shrinks to the room right of its clamped left edge and
wraps, so it never ran past the screen, but near the right border it
filled that room to the last pixel. Cap its width at the room less the
8 px gutter the clamp was meant to leave.
The mood and dashboard prefetches and the sign-in location lookup raced
their work against a bare setTimeout that was never cleared, leaving
one live timer behind per render or sign-in after the work had already
answered. A shared helper races the work and clears the timer.
A first read of the vault layout that was still running when the switch
was flipped landed afterwards and painted the old layout until the save
answered, so the switch jumped back for a moment.
The allergy card had two action rows, the list's add button and the
free-text note's save, and the family-history card on a shared record
put its sentence about relatives after the list's actions. The add
button now rides the note's row before save, and the sentence moves
above the list.
The empty adherence tile linked to the retired /medications/new route,
which only redirects. The sidebar's short-touch rules keyed on a touch
screen at most 520 px tall, where the phone shell hides the rail, so
they could never apply.
A span marked both shrink-0 and truncate never gives up width, so a
long source name on a collapsed priority group squeezed the metric
label at phone width, and a long month marker in the document flow
could overrun its cell. Both shrink now and cut with an ellipsis.
The encryption column table and the source-priority groups, overrides
and device ladder rendered their panel only while open, so a collapsed
toggle's aria-controls pointed at nothing. The panels now stay mounted,
hidden and empty, while closed.
Unlayered, the safe-area, keyboard and sheet-clearance rules outranked
every utility on the same element, so a later bg-* on the shell or a
bottom-* on a sheet would have been ignored without a trace. They now
sit under the utilities. The bottom sheet drops its bottom-0, which the
layered keyboard rule replaces with the same zero until the keyboard
opens; the keyboard rules that must win keep their !important. The
toast lift stays unlayered to outrank the toaster's own rules.
A bottom sheet standing on the on-screen keyboard kept its safe-area
inset as bottom padding, leaving about 34 px of empty space between
its last row and the keys on an iPhone.
The theme-color tags were rewritten to the app's theme only after
hydration, so with a light system and the app in dark mode the status
bar stayed light on a cold start. The existing boot script now sets
them along with the theme class, and once more after parsing in case
the tags stream in after it.
The reload cleared every HealthLog cache after asking the worker to
update, but the update resolves as soon as the new worker starts
installing. Its fresh precache of logos and favicon was deleted with
the old ones, so the installed app lacked them offline. Leave the
target release's caches in place.
A time-axis chart drawn from real instants tested each day's noon
anchor against the readings themselves, so the first day dropped out
when its reading came after noon UTC, the last when it came before, a
single reading never got a dot, and the span was cut in UTC while the
days were local. The row now spans the points' own calendar days, as
the chart hands them to its day links.
The day view translated any title reading "cycle" or "travel", so a
medication or a document with that name showed up as the localised
cycle or travel label.
The in-use probe matched tombstoned readings, so a source stayed marked
on the priority ladder after every reading from it was deleted. Both
the read and the save response of the source-priority route go through
this one probe.
A body held back until its reads settle was hidden when the page
arrived, so a link such as the pill-order anchor in the insights
settings had nothing to scroll to and the page stayed at the top.
Marc-André Bombeck