Repository navigation
Watch mode
A station check grades a station. A watch observes one.
Ninety seconds answers is this station healthy now. It cannot answer does it stay healthy, and that is the question behind most real complaints: the base that drops for forty seconds every twenty minutes passes every spot check anyone ever runs on it.
Tap Watch station on the station screen — every run starts and stops there, and each run button says beneath its name which tier answers and roughly what it costs: the check gives a verdict in about two minutes; the watch adds stability and needs ten minutes of evidence before it says anything at all.
The run keeps going until you stop it — through the screen going dark, and with a notification for its whole life so the system leaves the connection alone. While it runs, the verdict banner says so: its subtitle reads watch 4 min · evidence 98/600 s and its progress bar climbs toward the ten-minute stability floor; once the floor is passed, the bar retires and the subtitle carries the stability verdict instead.
Everything else behaves as during a check: the eight rows keep updating, and the analysis views keep filling — the Analysis screen is the viewing room, not the cockpit; it starts and stops nothing.
Below the eight checks, a watch fills the Stability card: six metrics over the whole run — availability, frame integrity, signal level, satellites held, ionosphere, delivery rate — each with the figure that decided it.
Its words are deliberately not the check's. A station can be fit this minute and have been unstable all week, or the reverse; so the card says STABLE, DEGRADED or UNSTABLE, and STATION OK stays the check's word alone.
INSUFFICIENT EVIDENCE is an answer, not an error. The card needs ten minutes of stream before it judges anything, and until then it says exactly how much is still owed. During a plain check — which ends long before ten minutes — the card says so and points you at the watch. Windows are measured in the stream's time, which is what makes a replayed capture report exactly what the live run did; and a report whose stream has stopped moving declines to keep judging, rather than publishing STABLE from evidence that stopped being about now.
The longer the watch, the more the verdict means: STABLE over 0.2 h is a start; STABLE over 4.0 h is a sign-off.
It appears on the main screen once a watch is running.
| Line | What it means |
|---|---|
| Watching for 3 h 07 m | How long this watch has been going |
| Healthy 99.4% of judged time | The share of time the verdict was OK. "Judged" excludes the opening seconds before there was enough evidence to judge at all |
| Current healthy streak 42 min (best 2 h 15 m) | How long it has been clean, and the best it managed |
| 3 degradation(s); worst FAILED | How many times it left OK, and how bad it got |
| At 3 h 12 m: FAILED for 6 s. KPI 4 failed | Each drop, newest first: when in the watch, how bad and for how long, and which checks were failing, warning or restarting. The last eight are kept in detail |
| No degradation yet; worst OK | Nothing has gone wrong so far |
| 2 reconnects — the stream dropped and came back | The connection was lost and re-made. Shown only when it happened |
A degradation is the event, not the duration. Three short drops and one long one both read as degradations; the streak lines and each drop's own line are what separate them. A drop names its checks by number, as the main screen numbers them: KPI 1, 5 failed after a dropped connection is the stream and the satellites that went with it; restarting means a check went back to waiting for evidence. Each check is named once, at the worst it reached. The shared report carries the same lines.
Read the reconnect line together with the degradation line. Three degradations with reconnects is a link that dropped — a tunnel, a cell hand-over, a caster restart. Three without is a station that faltered while the connection held. Those are different faults, and only the pair tells them apart.
A dropped stream is not the end of a watch. The app waits a second and tries again, doubling the wait to a minute at most, and goes on doing that for as long as the watch runs — so a phone that walks into a building, hands over between masts, or meets a caster being restarted picks the stream back up without being asked. A stream that expires is treated as a drop, and reconnected the same way.
This cannot turn a bad station into a good one. The verdict's sustain clock resets on the gap, so OK has to be re-earned from the reconnection onwards, exactly as it does after any degradation. What reconnecting buys is that one bad minute does not cost you the other eight hours.
Both editions do this; it is not a paid capability. What it leaves behind is the count on the Watch card, the same number under check 1, and a break in the satellite trails where nothing was measured.
- An hour catches a station whose satellite geometry or interference varies through the day.
- A working day is what an acceptance report wants: availability over a period the customer cares about, with the worst state named.
- Overnight is the honest way to catch a scheduled restart, a maintenance window, or a caster that sheds clients under load.
⚠ Android 15 sets a ceiling on this. A phone running Android 15 or later allows an app about six hours of background streaming in a day, and then stops the service itself. The watch ends there and says so — "stopped by Android's six-hour limit" — and everything measured up to that point is kept and reported, so an interrupted overnight watch still yields the hours it did manage. To cover a longer period, start a fresh watch afterwards. Android 14 and earlier have no such ceiling.
A watch ends in one of three ways: you stop it, the link fails, or Android stops it at the six-hour ceiling above.
Capture stops; the picture stays. Whichever ending it is, nothing measured is thrown away. The eight rows, the sky view with its trails, the signal bars and the C/N0 scatter all hold what they had at that moment, and can still be read, compared and shared. What ends is the measuring: no new data arrives, and the numbers stop moving. When it is Android's ceiling, a dismissible notification says so, and the main screen reads "stopped by Android's six-hour limit" rather than borrowing the word for what the Stop button does.
What survives is in the app, not on the phone. No capture is written to storage — the only thing this app keeps on disk is your connection settings. A capture survives the watch ending, rotating the phone, the screen going dark, and leaving the app for something else. It does not survive swiping the app away, restarting the phone, or Android reclaiming the app's memory while you are elsewhere.
So get a capture worth keeping out of the app before the phone goes back in a pocket, and there are two ways out: the share control in the top bar sends the report from the main screen and the picture from a plot, and Pro's Export statistics… (in the ⋮ menu) writes the numbers themselves as a file — the full snapshot as JSON, or the same CSV columns the desktop tools produce, for a spreadsheet. The export works during a run and after it; only a capture that was never taken cannot be saved.
A fresh watch is a fresh capture. The plots belong to one run and are cleared when the next one starts, which is why a period longer than the ceiling is covered by consecutive watches rather than by one continuous picture. Note when each one started; the app cannot join them for you.
Neither plot fills up in any period you are likely to watch:
| Satellite trails (Pro) | A day per satellite, at a point a minute. A longer watch keeps the most recent day and drops the oldest points as it goes |
| C/N0 against elevation | Everything, however long the watch. Samples are counted into the plot's own cells rather than kept one by one, so nine hours and nine minutes cost the same memory |
| The eight rows and the Watch card | The whole watch: availability, degradations and streaks are running totals |
The numbers to quote are availability, degradations and worst state. A station at 99.9 % with one FAILED degradation is telling you something different from one at 99.9 % that never left CAUTION.
- It holds a network connection open for the whole watch. On mobile data, that is measurable — a typical station runs at 1–3 kB/s, so roughly 4–10 MB per hour. It is measurable at the other end too: a multi-hour watch is a multi-hour client on somebody's caster, so watch a station you are entitled to watch (Getting started).
- Keep the phone on power for anything longer than an hour or two.
- If a watch stops unexpectedly, check battery optimisation for this app. The notification should prevent it, but some manufacturers are aggressive regardless.
- The verdict during a watch is the same eight checks. A station that degrades and recovers will show the recovery: the sustain clock restarts, so OK is re-earned rather than remembered.