The averages on a dashboard tile open their colour hint on a tap now,
which made them touch targets of 16 px height. On a touch screen the
trigger is 44 px tall, and a negative margin keeps the tile's line
where it was.
Opening a sheet pushed a history entry, and closing it took that entry
off again with a history back. The browser answers that back later, so
when the close started a navigation it landed after the router's push
and undid it: the medication wizard stayed on the list instead of
opening the new medication, a section picked from the settings list on
a phone stayed on the account page, and a link followed from a sheet
could put the page under it back.
The sheet history goes, with its wiring in ResponsiveSheet, the
document sheet's opt-out, the account form's deference to it, Back
answering the discard question, and their tests. Back behaves as
before: it leaves the page and the sheet goes with it, and the account
form's own Back guard answers every Back again. A tap outside, Escape,
the X and a swipe still ask before dropping typed input. The two size
baselines return to their earlier numbers.
A badge call from the service worker stalled Chromium's worker thread
while it activated, awaited or not: the worker reached "activated" but
never claimed its pages, so no page was ever controlled and every
offline path was gone. The worker no longer touches the badge. The
installed app clears a leftover count from the page once the worker is
registered; a browser tab leaves it alone.
The filter-sync test no longer reads `history.length`, which the
browser caps; the recorded push without a replace already proves the
new entry.
The two ketone cases repeated the glucose cases with one type changed; they are now one parametrized case each, which keeps the mocked client casts at their count.
The Fitbit guide still called the integration experimental and spoke of
the 30 October 2026 turndown as ahead. Three new files carried em
dashes in a doc comment and two test titles.
- The About button opens the release notes and now says so in all ten
languages; six of them still showed the English word "Changelog".
- The discard question now also covers edit sheets, where "Discard this
entry?" read like deleting the target or episode being edited. It
asks about the changes in every language; Russian also loses the
ambiguous "Отменить" on both the title and the button.
- The 7-day trend figure on a dashboard tile had an English screen
reader label in every language.
- Outlier terms in the new checkup and ketone strings follow each
catalog (ru "Кетоны в крови", "Коуч"; es "Desactivado"; nl
"rusthartslag"), the French steps question reads naturally, and the
shared Polish and Russian Coach question no longer assumes a
masculine speaker.
The case with a year of earlier takes proved the ledger stays the same,
which the old read of every intake satisfied as well, so a revert of
the bounded read passed every test. It now also counts the rows the
route read and expects only the window's.
- README: the integrations line now names Strava, Garmin and the Health
Connect import, the reminder channels include email and the webhook,
and the last two feature labels say what they cover.
- UI guidelines: the password form was said to carry a hidden username
field it does not have; the line now states when one helps.
- CONTRIBUTING: strings go into every locale file, not only English and
German.
- The translation issue template called eight locales "AI-initial",
left Russian out of its list and its dropdown; it now says they
started as machine-translated drafts reviewed by the community.
- The pull request template asked for two locales where the checks
require all ten, and pointed at an ignored docs folder.
- Code comments named the maintainer notes file, a release by its
working name and an internal host; old changelog entries explained
the notes files by their tool-specific names. A fixture token and
the deploy webhook test's application id (a real one) are replaced.
The git and docker ignore files and the lint config listed the folders
and notes of particular editors and tools. Local workspace folders
belong in a contributor's own global excludes. The image copies only
named paths, so nothing local reaches it, and the lint config now skips
every hidden top-level folder, which covers local worktrees under any
name; no tracked hidden folder holds code.
The commit-msg hook listed the tools and the session link format it
refused, so the file read as a description of how commits are written.
It now states the rule: commits are authored by the maintainer's
identity, and a co-author trailer is only for a contributor. It refuses
a generator line, any "<Word>-Session:" trailer and a co-author trailer
whose address is not a GitHub noreply address, so a contributor's
credit still passes.
Several test titles and comments said whose readings a fixture came
from, with a host name beside them: a blood pressure average, a month
of paired readings with dates, a cached insight about weight, cycle
history and dose timing. They now describe the fixture itself (a
profile averaging 134/87, a ten-pair month, a long history); the values
and what each test proves are unchanged.
Four comments still named `queue-runtime.ts` without its folder after it
moved under src/__tests__/helpers/, and the v1.45.0 boot page described
two dumps that are no longer kept.
Water, caffeine and the nutrient channels feed the correlation scan
behind /api/insights/correlations and the Coach read line on the metric
pages, but the water and nutrient batch writes evicted only the
dashboard snapshot. After logging water both could serve the read taken
before it until the cell went stale. Both writes now sweep the record's
analytics cells, as measurement and mood writes do.
The tab strip shows the Workouts pill from the presence read, which
stays fresh for five minutes and was not part of the workout write's
invalidation. After the first manual workout, or after deleting the
last one, the pill stayed missing or stayed on for up to five minutes.
Inside a managed child's record, or as a delegate with manage access,
the tab strip showed Weight, Pulse, Blood pressure and the other metric
pills, and every one of them opened an error. The strip now reads the
record's own presence answer, but the metric pages read /api/analytics,
which serves only the signed-in account's own record and refuses a
switched session. The strip again shows those pills only in the
account's own record, as before; mood, medication, workout and ECG
pills keep following the presence answer.
A "Measure blood ketones" checkup kept sending its push, Telegram or
email with the Glucose module switched off, while the glucose checkup
beside it went quiet. The checkup job kept its own copy of the module
map, which knew glucose but not ketones.
The job now asks the one ownership table the rest of the app gates on.
That table did not list the four mental-wellbeing screenings, which only
the job's copy knew, so they join it under the mental-health module; a
screening score row now also follows that module in the day view and
search.
- The Coach thread kept 80 px free under the last message for the New
conversation button while the keyboard hid that button.
- The remove buttons on lab alias chips and attached-document chips
reached sideways into their own label and the next chip, and an
alias is removed at once. Their tap area now stays inside the chip's
gaps sideways and keeps its height.
- A capture menu comment described a dialog that is gone.
The symptom form and the dashboard's medication intake asked "Discard?"
on every close, untouched. The sheet's fallback check read the input
Radix keeps for the intensity slider (value "0" before anyone moves
it) and the dose the form fills in from the preselected medication as
typed input.
The check now skips a slider's own input and counts a slider moved off
unanswered instead, and compares a field against the value its form
says it filled in (`data-seed`). A controlled input's defaultValue
follows its value, so it cannot tell the two apart. The dose field
names the medication's dose, one line more for the form's baseline.
Back on a sheet with unsaved input asks whether to discard it. A second
Back while that question was open put the sheet's entry back and asked
again, so the Android button could never get past the question. It now
closes the question and keeps the sheet and its entry, like Cancel.
Three ways the sheet's history entry could still cost a Back:
- A close that also writes the URL (the day's date picker, the life
event sheet opened by ?add=lifeEvent) let the router's answer to that
write land on the page entry with the closed sheet's mark. The next
sheet close then stepped one entry past the page, and on an installed
app opened from a notification that closed the app. A write that
carries the mark of a sheet that is gone now loses it.
- A sheet's entry started empty, and landing below it copied that empty
state onto the page entry, so an open day or document lost the key it
keys its own Back on. The sheet's entry now starts as a copy of the
page entry's state.
- The account form's Back guard heard the sheet's own Back first when
it was mounted before the first sheet of the visit, and asked to
discard the form's edits for closing the section list. The guard now
leaves a popstate the sheet history owns to it.
The browser tests run against Chromium's session history and fail
without each fix.
The new worker cleared a leftover app badge inside its activation and
waited for it. A browser that never settles that call keeps the worker
activating, so it never takes over its pages and every offline path is
gone. The badge is now cleared beside the activation, best effort.
The sign-in journey expects the page's own sentence for a wrong
password, not the server's text.
Moving the navigation into the shared helper took both sign-in hand-offs
out of what the headerless-transport guard walks, which only reads the
component and page trees. The helper now only clears the offline copies
with the capped wait, and each page navigates itself. The device
registration test checks the end of the no-device rest inside the test
that already pairs an iPhone instead of a copy of it.
Merging the previous release in kept this branch's older numbers for
telegram-webhook-handlers.ts (1335 lines, handleCallback 590), although
the file is unchanged at 1319 and 582, the numbers the previous release
recorded. The gap sat inside the ratchet's slack, so nothing flagged it.
The register refusal codes, the end of the no-device rest, the
no-recipient annotation and the sign-in shell comment had each grown a
function the size ratchet holds. The codes become constants, the rest
ends through a helper beside the one that starts it, and the
annotation moves into that helper. The settings Web Push card dropped
under the limit when its subscription post moved out, so its entry goes.
The sign-in page and the profile take-over page each cleared the offline
copies with the same capped wait before leaving for the identity
provider, with the limit written down twice. leaveForSingleSignOn does
it once for both.
A send with no browser subscribed, no iPhone paired or a transport the
operator has not set up no longer counts as a failure, but nothing else
replaced it. The mood reminder and the daily briefing retry every
minute until delivered, so such a channel wrote a skipped row each
minute, for good; the dispatch still counted as an attempt, so the
watchdog and status alerts took it as announced; and the status screen
said "active" with no hint why nothing arrived.
The channel now rests for an hour after such a send, with the reason on
its status, and the send does not count as an attempt. Subscribing a
browser or pairing an iPhone ends the rest at once.
The record fence journey called the page cache its positive control,
but signed-in pages are never stored, so the entry it counted was the
install precache, and the half inside the record repeated a claim that
holds by construction now that the worker keeps no API answers. The
comment now names the precache, and inside the record the journey reads
the IndexedDB dashboard snapshot, which must be missing or still the
caller's own.
Sign-out removes the push row first, through the raw fetch helper, which
has no default timeout. On a connection that is up but carries nothing,
the sign-out waited for the browser's own timeout before the logout
request even started. The removal now gives up after five seconds.
The dose tag comment still explained the clear push sent for a logged
dose, the session-end wipe comment still named the worker's API cache as
a live layer, and the worker's header and foreign-origin note did not
match what the worker stores and when it resumes.
The light-theme fix painted the strip behind the white status bar only
in the signed-in shell. The sign-in, registration and take-over pages
and the onboarding shell padded the strip without painting it, so after
a sign-out in the installed iOS app the clock was white on near-white.
They now use the same shell-safe-area band.
The Web Push test answered a refused push with `ok: false` and the
sender's `reason`, but the Test button reads `meta.errorCode`, so it
only said "Test failed". The answer now carries `upstream_rejected`, or
`connection_failed` when no push service answered.
The test push used the administrator's test title, which put the brand
first in several languages; it has its own title now.
The German sentence dropped its article, and the Spanish one read as if
the sign-out had been the session's own state ("Tu sesión sigue
abierta"); it now says the person is still signed in.
The hint for iPhone and iPad Safari ended with "turn on push here", but
the button reads "Enable push"; it now names the button in every
language. The hint is two sentences of steps, so it is shown as content
rather than muted, and the French one uses the typographic apostrophe
like its neighbours.
The offline strip joins two sentences, and both ended on "once you're
back online"; in French it ran to four lines on a phone, and the
Russian one was ungrammatical. The first sentence now only says that
changes can't be saved.
On Back, the shell set the list's saved offset again on every frame
while the list was still shorter than that offset. Someone who started
scrolling in that time was pulled back, and a list whose data took more
than a second ended up where the skeleton allowed. The restore now stops
at the first wheel, touch, key or pointer input and keeps trying for up
to three seconds.
Marc-André Bombeck