2026.8.3

Release Notes

This release is mostly about the Quickshell panel, which grew from three cards
to ten: it now opens on last night's window, draws carbs and insulin under the
trace, reads today against your own fortnight, shows a fortnight of time in
range one bar per day, and lets you tap an hour of the profile to see the
individual days the median hides. It can also act — snooze the alarm that woke
you, or log a treatment — and it says what the alarm actually did, including a
delivery that never arrived. The bar pill wears your theme's colour only when
something is wrong, and each of its parts can be switched off. Two fixes worth
naming: sugarrush config units mgdl used to be refused outright because it
reinterpreted your thresholds rather than converting them, and the panel
ignored the colourblind palette entirely.

Changed

• Views cross-fade instead of popping. One set of cards used to vanish and
  another appear in the same frame, which reads as a redraw rather than a move.
  animations false turns it off.

• The panel uses the width it has. It was one 380px column in a scroll
  view, on a desktop, and it has grown from three cards to nine — the Glucose
  view scrolled past its own statistics and the Profile view past its patterns.
  On a horizontal bar it is wider now, with the reading beside the last 24
  hours and the clinical summary beside the alarms; the charts keep full rows. A vertical bar keeps the narrow
  column, where it is the right answer, and wideLayout false turns it off.

• The stats card admits when its window is full of holes. A percentage
  computed over a day with half its readings missing drew exactly the same
  confident stripe as a complete one. The document now carries how many
  readings each figure stands on, and below 70% of a full window the card is
  marked "partial" and says how much of it is a gap rather than time in range.

• The alarm thresholds are one band, not four steppers. The Settings view
  draws all five zones with a handle at each threshold, in the colours the
  reading itself uses. The handles cannot cross, so the crossed-threshold
  configuration sugarrush config refuses is now unreachable rather than
  rejected. The four stepper rows are gone — a handle lands on a tenth of a
  mmol/L in about two pixels, so they were rows buying nothing.

• The panel draws itself while it loads. A cold open used to collapse the
  whole panel to the line "waiting for the first reading" for as long as
  Nightscout took, which looks like a failure rather than a wait. It now draws
  the cards it is about to fill.

• The reading rides in the panel's header. The wordmark was sized to half
  the panel width, and the number you opened it for scrolled off the top — on
  the Settings view it was not on screen at all. The mark is a third now, with
  the reading and its arrow beside it, so it survives scrolling and view
  switches.

• The panel's settings no longer claim a state they have not read. Every
  switch showed "on" until sugarrush config answered, so for a moment the
  Status bar card asserted four settings were enabled when two of them were
  not. Those cards are dimmed and inert until the file has actually been read.

• The Quickshell panel obeys your theme. Its chart, profile, time-in-range
  bar, sensor countdown and alarm chip all carried a hard-coded copy of the
  default palette, so anyone on the colourblind preset — the one setting whose
  entire purpose is that red and green are the wrong pair — got an accessible
  bar and an inaccessible panel. The snapshot now carries the resolved palette
  and the panel paints from it. Low and high are separate colours there too,
  as they always have been everywhere else.

• A stale reading in the panel looks stale. The age sat in the same grey
  pill at 3 minutes and at 40. Past your own stale_minutes, the reading greys
  out, the pill turns red and says "no reading for 26 min", and the caption
  gives the clock time it was last seen — at 3am the number is not wrong, it is
  absent, and the panel was presenting absence as a value.

• The panel's forecast stopped jumping sideways. The projection was
  anchored to the reading's rendered width, so gaining or losing a digit moved
  it. The reading now reserves room for the widest value its units can produce.

• The Quickshell pill wears the bar's colour until something is wrong. It
  used to paint the whole reading — number, unit, arrow, delta and trace — in
  the alert colour at all times, so a perfectly ordinary reading sat on the bar
  as a green light. Now only the number takes colour, and only when it is out
  of range or a forecast crossing is coming; everything else stays in the bar's
  own foreground. Stale data is carried by its ? marker alone.

• The panel's threshold lines are scale marks again. They were drawn in
  their alert colours, so a red rule ran across the chart whether or not the
  reading ever went near it. The line already turns amber or red where it
  crosses one; the rules are now the panel's foreground, with the urgent pair
  fainter than low and high so the target band reads first.

Added

• The Status bar card shows the pill it is describing. Four switches that
  changed something at the other end of the screen now draw it: the reading,
  the unit, the arrow, the delta and the sparkline, in the pill's own colours,
  updating as you flip them. You no longer have to look away to see what you
  did.

• Units are switchable from the panel. A mmol/L · mg/dL choice above the
  threshold band, which converts the thresholds with it.

• Last night, as one card. Nobody opens a CGM panel in the morning to
  browse six hours of chart — they open it to find out whether the night was
  fine. The Glucose view now leads with 23:00–07:00 in your own timezone: the
  shape of it, time in range, the lowest reading, and what the alarm did, with
  every excursion below range marked rather than only the worst one. Before
  07:00 it says "tonight so far" and stops at now, because reporting a night as
  finished before it is would be a claim about hours that have not happened.

• The panel shows what the alarm actually did. sugarrush alerts has been
  recording every episode and every delivery attempt since it was written, and
  nothing displayed any of it. The Profile view now lists recent alarms with
  how long each lasted — and, in red, any whose delivery never arrived. A push
  that failed means an alarm was raised and nobody was told, which was until
  now the least visible line in the whole system.

• Tap an hour of the profile to see the days behind it. A percentile band
  says the middle half of your evenings look fine; it cannot say that three of
  the fourteen went low, which is the finding that changes a basal rate. Tapping
  an hour marks it, draws each day's value for that hour as a dot, and says the
  median, the spread and how many days fell out of range. Tapping it again
  clears it.

• The panel keeps itself current while it is open. It used to fetch once on
  open and then sit there — watch a low come up and the panel quietly stopped
  being true while the pill behind it kept updating, the two disagreeing with
  the bigger surface being the wrong one. It now refetches at the pill's own
  interval for as long as it is on screen, and the header counts down to the
  next one: what a spinner should have said all along is not that something is
  happening, but when.

• The clinical summary is on screen before it is on the clipboard. The
  Profile view shows the same figures sugarrush export writes — mean, GMI, CV
  and the five bands with their boundaries spelled out — so you are not sending
  a clinician a block of text you have never read.

• The panel takes the keyboard. 1 2 3 jump to a view, ← → step
  between them, r fetches, d opens the dashboard, Esc closes, and ?
  shows the list. It ships to people running Hyprland, where the mouse is a
  last resort, and every view and settings row needed a pointer to reach.

• A fortnight of time in range, one bar per day. The Profile view gains a
  strip of stacked daily bars, oldest on the left, under the typical-day chart.
  The profile answers "what do my nights look like"; this answers the other
  question people track — whether it is getting better — and neither can be
  read off the other. Days grouped at local midnight, so a night is not cut in
  half by a UTC boundary. A day with too few readings to trust is drawn faded
  rather than as a full bar, since one thin morning would otherwise read as the
  best day of the fortnight.

• The panel says whether today is better than usual. "76% in range" is a
  number with no scale, and the comparison that changes what someone does is
  against themselves a fortnight ago. The stats card now reads time in range,
  time below range, mean and CV against the whole history window, with the
  change named and coloured — where there is a better direction. Mean has none:
  a lower average bought with lows is not an improvement, so its change is
  stated rather than judged.

• The panel's chart shows what you did, not only what happened. Carbs and
  insulin logged on the site are drawn in their own lane along the foot of the
  chart — a dot sized by the amount for carbs, a triangle for insulin, labelled
  where there is room — and the card title totals the window: Last 6 hours · 45g · 5.7u. A rise from 5 to 12 means one thing after 60g of pasta and
  another with nothing logged, and the chart could only ever show the rise.
  sugarrush had been fetching and discarding this all along.

• You can log carbs and insulin from the panel. A Log button beside the
  snooze pair opens two steppers; Review in terminal hands the amounts to
  sugarrush treatment, which prints what it is about to write and asks for
  the person's name before i...

2026.8.2

Release Notes

This release turns sugarrush from a dashboard you watch into something that
watches for you. sugarrush watch runs the alarm headless, so a nocturnal low
still wakes you when no terminal is open, and it now follows every configured
person at once — each with its own alert episode, its own snooze, and a record
of what the alarm actually did. You can log carbs and insulin back to
Nightscout behind an explicit confirmation, keep a private local history, and
export a clinical window as CSV plus a plain-language summary. Bars are
first-class too: one JSON line for Waybar, tmux, polybar and i3blocks, and a
full Quickshell widget for Omarchy 4 whose panel charts the last hours against
a typical day, the day's time in range, and the times of day where lows keep
happening.

Added

• sugarrush has a face. The running sugar cube is now the app icon
  (assets/icon.png), the wordmark opens the README and leads the Quickshell
  panel, and a silhouette of the cube rides in the bar beside the reading.
• The Quickshell widget has a popup panel. Clicking the bar pill opens a
  panel of labelled cards: the reading now, the last hours as a chart — your
  line over a typical day for the same hours, changing colour as it crosses a
  threshold, thresholds in their own colours, values and clock on the axes —
  the day's five time-in-range bands, and the pattern insights. Hovering the
  chart reads out the value and time under the pointer. Each card names
  the window it covers, so a 24-hour average is never read as a six-hour one — the parts of the dashboard worth a glance without opening
  the TUI. Right click still opens the full app. It fetches only while open,
  and reuses its last reading for a few minutes, so the multi-day history the
  patterns need is paid for only when someone is looking.
• sugarrush snapshot prints the whole picture as JSON. One document with
  the current reading, a series for a chart, time in range and patterns, in
  your display units. Built for the panel; useful to anything that wants
  sugarrush's numbers without scraping a bar line. --demo renders it from
  synthetic data with no site configured. The document also carries a
  typical-day band (median and 25–75% per quarter hour, over the requested
  history) for anything that wants to draw today against a usual day.
• A bar widget for Quickshell. Omarchy 4 replaced Waybar with a Quickshell
  bar, so quickshell/ now ships a widget for it alongside the
  Waybar examples: the reading, trend and delta in your theme's state colour,
  the usual tooltip on hover, and a click that opens the TUI (right click for
  its settings screen). Its README also covers the no-plugin route, running
  sugarrush waybar through the bar's own command module.

Changed

• The Quickshell bar widget prints the unit. The bar now reads
  10.5 mmol/L → -0.2 rather than 10.5 → -0.2: a number by itself names
  nothing, and the unit is the one word that says what was measured. Turn it
  off with showUnits on a crowded bar. The sugar cube that used to carry that
  job is now opt-in, behind showMascot.
• --format waybar is now --format json. The same JSON is read by
  Waybar, by the Quickshell widget, and by anything else that takes a JSON
  line, so the format is named for its syntax rather than for one bar.
  waybar and bar are accepted spellings, and sugarrush waybar is
  unchanged — no config needs editing.
• --format json reports the reading in parts. value, units,
  arrow and delta join the JSON, so a bar can compose its own line instead
  of parsing text apart. Waybar ignores them.
• The Quickshell bar widget no longer pops up a tooltip. Hovering the
  reading used to show a second copy of it with a text sparkline; clicking
  opens a panel with the real chart, so the tooltip was answering a question
  that already had a better answer.
• --format json also reports the state colour. The JSON now carries a
  color field holding the hex colour of the alert state, from your configured
  theme. Waybar ignores it and keeps styling the module from the CSS class;
  it is there for bars that style themselves from the payload instead of a
  stylesheet.

Fixed

• --help fits a terminal again. The new command signatures outgrew the
  fixed column, so every row past it ran to 138 characters and the description
  column stopped existing. Long signatures now put their description on the
  next line; the widest row is 97 columns.
• A lock left behind by a crash no longer disables treatment writes or the
  history cache permanently. Both used a lock file removed on clean exit, and
  a SIGKILL or a power cut leaves it behind — with no staleness rule, every
  later run waited out the timeout and failed, so one hard kill disabled the
  feature until someone found and deleted a file nothing had told them about. An
  abandoned lock is now reclaimed after 30 seconds, a live one is still
  respected, and the timeout message names the file and what to do about it.
• A successful treatment write is no longer reported as a rejection. The
  client read the array Nightscout returns on success — the documents it just
  stored — as a list of refusals. So a treatment that Nightscout accepted was
  reported as rejected, and following the retry advice wrote it again: two
  identical carb or insulin records for one dose, both reported as failures.
  Only a non-empty array now counts as accepted; every other response shape is
  reported as unknown, which tells you to check Nightscout and retry with the
  same operation ID rather than a new one — the one retry that cannot duplicate.
• Health and delivery joins survive person renames safely. Concurrent
  follower results and new delivery receipts now carry immutable site IDs;
  legacy name-only receipts remain readable without letting a reused display
  name inherit another person's current delivery status.

Added

• Unattended treatment writes now need a grant in config.toml.
  treatment --non-interactive could write to someone's health record with no
  human in the loop, on a flag combination alone — available to anything that
  can compose a command line. It now also requires allow_unattended_writes = true, off by default and deliberately not editable in the settings screen, so
  the grant is a decision made once in a file you own. sugarrush about reports
  whether it is on, and now also reports whether each site has a write token at
  all — the presence, never the value.

• Treatment submissions have a privacy-safe receipt view. treatments
  filters the bounded local audit by person and time, emits text/JSON/CSV, and
  exposes stable operation IDs for reconciliation without notes or credentials.

• Managed watchers now point to usable diagnostics on every platform.
  macOS and Windows services write to a private user-data log, install/status
  print its path plus the strict health command, and uninstall says that logs
  remain; platform session and independent dead-man limits are documented.

• Operational health is explicit instead of collapsing unlike guarantees.
  JSON now separates process, data, configured-channel, suppression, and known
  delivery status; --strict-delivery gives external monitors an opt-in
  degraded exit policy without claiming that an accepted alert was received.

• Multi-person exports can no longer silently choose the first person. A
  follower must select --site NAME or explicitly request --all; filenames
  and output identify their subject and the matching cache/timezone are used.

• Private cache storage is inspectable and selectively erasable. cache status reports each person's entry count, date span, and bytes without
  printing readings; confirmed clear commands target one person or everyone.

• People now have immutable internal identities. Renaming a display label
  no longer moves alarm episodes, snoozes, cached readings, or treatment audit
  receipts between people; legacy configurations receive a deterministic ID
  and newly added sites receive a generated UUID.

• Treatment writes are available behind an explicit security boundary. A
  separate per-site CarePortal token is masked in Settings and capability-
  checked before every confirmed treatment command; amounts, timestamps, and
  notes are validated and each accepted/rejected attempt is recorded in an
  owner-only audit without credentials or note text.

• Offline history is available as an explicit privacy choice. An opt-in,
  owner-only per-site cache gives instant startup and outage context without
  ever presenting cached data as a live fetch; retention is bounded to 1–90
  days and disabling it deletes the local record.

• Caregiver review now follows the person's context. Each site can set an
  IANA timezone used by AGP bucketing, pattern names, CSV offsets, and clinical
  summaries; alert history can be filtered with --site or emitted as
  structured JSON/CSV for private audit workflows.

• The always-on watcher now has native service management on every shipped
  desktop platform. One command installs, starts, inspects, or removes a
  systemd user service on Linux, a launchd agent on macOS, or a Task Scheduler
  task on Windows; the old --install-unit spelling remains an alias.

• Alarm delivery is now auditable and monitorable. Alert history records
  privacy-safe channel outcomes using careful accepted/rejected language, and
  sugarrush health --json exposes watcher liveness, per-site freshness,
  snoozes, alarm state, and the latest delivery attempt for external monitors.

• Caregiver actions now target the right person. The follower list has a
  stable selected row that opens that person's dashboard with Enter and
  snoozes only that person with a; the CLI accepts --site NAME and requires
  explicit --all when several sit...

2026.8.1

Release Notes

This release is about trusting what's on screen. The alarm path got a
correctness pass — the newest reading is now genuinely the newest, alerts no
longer re-fire every time a value hovers on a threshold, a flat glucose no
longer predicts a low, and a bad token says so and stops retrying instead of
looking like a network outage. Saving settings can no longer lose your
Nightscout token, the site URL and token are now editable in the app (masked,
and no longer echoed by the setup wizard), and an unencrypted http:// site is
called out rather than accepted silently. The stats panel turned clinical —
five-band time-in-range with time-below-range and CV variability, over a fixed
14-day window — and history is now fully navigable from the keyboard.

Added

• Keyboard navigation for the overview strip — H / L (or PgUp /
  PgDn) pan a whole window at a time and End jumps to the oldest edge of the
  overview, so moving through history no longer needs a mouse.
• Clinical time-in-range. The stats panel now splits readings across the
  five consensus bands (very low / low / in range / high / very high) instead of
  three, calls out the time below range percentage — the number that changes
  treatment — and reports CV, the standard measure of glycaemic variability,
  highlighted when it exceeds the 36% consensus target.
• Content-free notifications — a new Notification detail setting
  (notify_content) keeps desktop notifications free of your reading and alert
  state, for lock screens and shared displays. They still fire, and urgent ones
  are still critical.
• Unsaved settings are visible. Settings apply live but only w writes them
  to config.toml; the settings header now says · unsaved changes (w to save)
  so quitting can't silently discard them.
• Fix a bad site without leaving the app — the settings screen gains a
  Site section with the Nightscout URL and read-only token. Press Enter on
  a row to edit it; the app reconnects immediately, and w saves as usual. The
  token is masked while typing and never rendered back (the row reads
  set · ••••••).
• Unencrypted sites are called out. A plain http:// site sends your token
  and readings in the clear, so the dashboard footer now says so and the setup
  wizard asks you to confirm before accepting one. Loopback addresses
  (localhost, 127.0.0.1) are exempt — they never leave the machine.
• Push alerts are now a settings row (Alarm → Push alerts) — toggle the
  urgent-alert webhook on or off in the app without deleting the configured
  push_url. The row shows only the host, never the full topic/path, and the
  new push_enabled config key persists the choice.

Fixed

• Esc no longer quits the app. From the dashboard it returns to the live
  edge (and closes the help overlay or a prompt, as before) — quitting is q,
  which is what every other screen already did.
• The current-value pane keeps its range label when compact. On a short
  terminal the colour-independent label (LOW, in range) was pushed onto a
  line that got clipped, leaving colour as the only cue; it now sits on the
  headline next to the reading.
• The overview strip says why it's empty — "loading overview…" before the
  first fetch, "no readings in this window" after — instead of drawing a bare
  box.
• The audible alarm falls back to the terminal bell when no system audio
  player is available (headless boxes, minimal containers, a bare SSH login),
  instead of failing silently — silence is indistinguishable from "glucose is
  fine".
• The setup wizard no longer echoes your token to the terminal — it's masked
  as you type, so it doesn't survive in the scrollback.
• URLs are normalized instead of failing mysteriously. A bare
  mysite.example.com, a trailing slash, or a pasted
  …/api/v1/entries.json?count=10 now resolves to the right base URL, in the
  wizard, in the settings editor, and when loading an existing config.toml.
• A partly-failed refresh now says so. When the readings arrive but a
  companion fetch doesn't, the affected panels (IOB/COB, treatment markers,
  forecast, sensor age, history, overview) were left showing old values with
  the connection still reading as healthy. The footer now names what's stale.
• The connection recovers while you're browsing history. Backoff retries
  were skipped unless you were at the live edge, so an outage that started
  while you were panning stayed until you refreshed by hand. The backoff also
  now actually reaches its documented 60s ceiling (it stopped at 40s).
• No more "heading low in ~0 min" from a stale forecast. Uploader forecasts
  are timestamped when the pump published them; an old one is entirely in the
  past, and its points were being reported as an imminent crossing. Past points
  are now skipped.
• A delta too small to show no longer renders as -0.0 in the dashboard or
  the Waybar module — the sign follows the value as displayed, so a flat trend
  reads flat.
• The predict-horizon setting is honest about its reach — above 30 minutes
  it notes that the local (AR2) forecast only projects that far, so a longer
  horizon needs uploader predictions from Loop or OpenAPS.
• Saving settings can no longer lose your token. config.toml is now
  written to a temp file (created owner-only, flushed to disk) and renamed into
  place, instead of being truncated and rewritten — so a crash, a full disk, or
  a power cut mid-save leaves the old config intact rather than an empty file
  with the only copy of your Nightscout token gone.
• The setup wizard no longer builds config.toml by string interpolation —
  values are serialized by the TOML library, so a URL or token containing a
  quote or newline can't corrupt the file or inject extra settings into it. The
  file is also created with 0600 from the start, rather than being chmodded
  after the token was already written.
• The audible alarm no longer leaks processes. Each alarm spawns a system
  audio player; those were never reaped, so a long overnight urgent state piled
  up hundreds of zombie processes. Finished players are now cleaned up on every
  play.
• Alert thresholds can't cross any more — urgent-low ≤ low ≤ high ≤
  urgent-high is enforced while editing, so you can't silently disable a band
  by dragging one threshold past its neighbour.
• The newest reading is now always the newest reading — entries are sorted
  by timestamp on arrival instead of trusting the order Nightscout (or a proxy
  or mirror in front of it) happened to return, so the current value, delta,
  staleness check, and forecast can't silently key off an older reading.
• Alerts no longer flap on a threshold. A reading hovering on a boundary
  (69 → 71 → 69, well inside CGM noise) used to re-fire the notification, the
  audible alarm, and the push webhook every single time. Clearing an alert now
  requires moving 4 mg/dL past the threshold; raising one is unchanged, so a
  real low still alarms on the first reading that crosses.
• A flat glucose no longer predicts a low. The predictive alert and its ETA
  now follow the centre of the forecast rather than the edge of the uncertainty
  cone, which widens with the horizon and so crossed a threshold even when
  glucose was perfectly steady.
• A bad token or URL now says so and stops retrying instead of being
  retried forever as if it were a network outage. After three consecutive
  authentication (401/403) or not-found (404) responses, automatic fetching
  pauses with an explanatory message; press r, or switch/edit the site, to
  resume.
• Stats are now clinical, not cosmetic — time-in-range, mean, and GMI are
  computed over a fixed window of the last N days (the AGP days setting,
  default 14 — the clinical standard) instead of whatever slice of the graph
  happened to be on screen, so panning or zooming no longer changes them. The
  stats panel title names the window (e.g. stats · 14d).

2026.7.3

Release Notes

This release hardens the safety-critical alarm path and reworks the graphs. The
audible alarm can no longer stall silently — the Nightscout client now times
out, a total sensor dropout raises a Stale alarm, sensor-error codes no longer
fire a false urgent-low, and a wrong token says so instead of reading as
"offline". Visually, both the AGP and the short-term forecast now render as
filled percentile/uncertainty bands, chart shading lines up with the axes, the
colourblind palette renders on every terminal (including tmux/SSH), and there's
a ? keybinding overlay. It also adds a first-run units prompt and an in-app
"not a medical device" reminder.

Added

• The AGP view now renders as a filled percentile fan (a shaded 5–95 and a
  brighter inter-quartile band) with a single bright median line, and its title
  shows the target range — much closer to a clinical AGP than the previous flat
  lines.
• A keybinding help overlay — press ? for a full cheatsheet. The footer
  now falls back to a terse hint set on narrow terminals (it previously clipped
  silently, hiding settings/site/snooze) while always keeping ? help visible.
• The dashboard footer now shows a snooze indicator with a countdown while
  the audible alarm is silenced, so it's clear the alarm is off and for how long.

Changed

• The forecast cone on the timeline is now a filled low–high band (matching
  the AGP fan) with the centre line drawn on top, instead of two dim edge lines,
  and it emanates from the last reading (the AR2 fallback's initial jump no
  longer leaves the fan floating above the current dot).
• In-app safety note — the "not a medical device" reminder now appears in
  the running app (a dim note in the header, and up front in the first-run
  wizard), not only in the README and about.
• Alert banner now follows the theme/colourblind palette (it was hardcoded
  red/yellow), and the selected settings row is highlighted full-width.
• Dashboard polish — the range bar now shows all four zones (urgent-low /
  low / in-range / high / urgent-high) and uses integer mg/dL labels; IOB and
  COB stand out from the dimmed device line; and the graph labels its carb and
  bolus markers with a small legend.

Fixed

• Chart background tints now line up with the chart. The in-range band, the
  AGP fan, and the forecast fill are painted onto the terminal buffer, and the
  plot-rect geometry didn't match ratatui's — it was two rows too tall and used
  the wrong left gutter, so the shading spilled past the axes and sat offset from
  the lines/points it shades (the forecast fan looked detached from the current
  reading). The geometry now replicates ratatui's chart layout exactly.
• Graph theming — the shaded in-range band is now derived from the in-range
  palette colour (it was hardcoded green, so it broke the colourblind preset),
  and the y-axis shows integer mg/dL values instead of spurious decimals.
• Setup — the first-run wizard now asks for the display unit (mmol/L or
  mg/dL) instead of always defaulting to mmol/L, so mg/dL users aren't dropped
  into the wrong unit.
• Accessibility — the colourblind-safe palette now uses named ANSI colours
  instead of truecolor hex, so it renders correctly on 16/256-colour terminals,
  tmux, and SSH sessions that lack truecolor (where it previously collapsed
  silently). The current reading is also exposed as plain text alongside the
  big-number glyphs, so screen readers, tmux copy, and braille displays can read
  it.
• Clearer connection errors — a wrong or non-readable token now reports
  "authentication failed — check your read-only token (not API_SECRET)" instead
  of a generic "offline", both at runtime (in the header) and during first-run
  setup; unreachable hosts and HTTP errors are also distinguished.
• Alarm responsiveness — a sensor gap now escalates to a Stale alarm within
  seconds (re-checked on the alarm tick) instead of waiting for the next full
  refresh, and a failed escalation push (dead push_url) is surfaced instead of
  swallowed silently.
• Alarm reliability — the audible alarm could stop working silently in
  several cases, now fixed: the Nightscout client had no request timeout (a
  stalled connection froze input and the alarm), a total sensor dropout read as
  "in range" instead of a sensor gap, Nightscout sensor-error codes (0–12) were
  read as a real reading and could fire a false urgent-low, and predictive
  alerts evaluated the previous refresh's forecast. Failed data fetches no
  longer pile up doomed follow-up requests.

2026.7.2

Release Notes

This release is a dashboard glow-up. The graph now colour-codes readings by zone
with a shaded in-range band and dashed threshold rails, adds a zoned range bar
under the current value, and gains a switchable AGP (ambulatory glucose
profile) view alongside the 3h/24h timelines. The stats panel picks up a
time-in-range bar and a mean sparkline, and short-term forecasts now render as
an uncertainty cone — a high/low band — instead of a single line.

Added

• Graph view tabs (Tab / Shift+Tab) — switch the graph pane between a
  3h or 24h timeline and an AGP (ambulatory glucose profile) that folds the
  last N days of readings onto a 24h clock as a percentile band (median +
  25/75 + 5/95). The number of days is configurable in settings (AGP days).
• Dashboard graph glow-up — readings are colour-coded by zone
  (low/in-range/high) with dashed reference rails at the low/high thresholds,
  the in-range region is shaded as a band behind the trace, and a zoned range
  bar under the big current value shows where it sits between the thresholds.
• Stats upgrade — time-in-range is drawn as a stacked zone bar, and the
  mean gets an inline sparkline of recent readings.

Changed

• Forecast is now an uncertainty cone — predictions render as a widening
  high/low band (the plausible range) instead of a single line; the
  time-to-low/high ETA warns on the worst plausible path.

2026.7.1

Release Notes

First public release. A fast, keyboard-driven terminal UI for viewing
self-hosted Nightscout CGM data.

⚠️ Not a medical device. Don't use it for treatment decisions.

Try it instantly on synthetic data with sugarrush --demo (no config, no
network).

Added

• Dashboard — big current value with trend arrow, delta, and a colour +
  text range label; stats panel with time-in-range, mean glucose + GMI,
  insulin-/carbs-on-board, and device status (battery, sensor age, last seen).
• History & forecast — braille/dot graph you can pan (h/l), zoom
  (+/-, 1h–24h), and jump to a date (g); a 24h minimap you click or drag;
  a short-term forecast overlay (uploader predictions or a local AR2 fallback)
  with a "now" line, a time-to-low/high ETA, and predictive alerts; carb and
  bolus markers on the graph.
• Alerts & safety — in-TUI banner plus cross-platform desktop notifications
  (Linux/macOS/Windows); an audible alarm for urgent lows/highs with snooze,
  per-level tones, quiet hours, and unacknowledged-alarm escalation (optional
  phone push); clear offline vs. sensor-gap states with backoff retry.
• Configuration — an in-app settings screen (s) grouped into sections,
  editing units, refresh, thresholds, alarms, and theme live and saving back to
  config.toml; configurable colours with a colourblind-safe preset; multiple
  Nightscout sites (n to switch); a first-run setup wizard.
• Elsewhere — a Waybar module (sugarrush waybar) with a sparkline tooltip
  and click-through; sugarrush --demo to try the app on synthetic data with
  no config or network.
• Distribution — published to crates.io, the AUR (sugarrush-bin), and a
  Homebrew tap; prebuilt binaries + shell/PowerShell installers via cargo-dist.