v3.8.8 — the car in the sea puts itself back, and a trip stops waiting on a cloud that went quiet

The car in the sea puts itself back, and a trip stops waiting on a cloud that went quiet.

Fixed

• 🗺 A car plotted in the wrong hemisphere now corrects itself at startup. The remembered
  north/south and east/west sign was a single stored value, and before v3.8.6 one frame that
  arrived with its minus dropped overwrote it. From that moment every restart primed from the
  poisoned copy, so the car stayed mirrored for good — the guard added in v3.8.6 stops the sign
  being learned wrongly, but it cannot un-learn one already stored.

  Mate now re-derives that sign from the car's own position history instead of trusting the
  stored value blindly. You cannot have driven a fortnight of kilometres somewhere you have never
  been, so the history outvotes any single frame by construction — and it is already on disk, which
  is why this needs no button and asks you nothing. Restarting Mate is the repair.

  Measured on @rop12770's install (#232): eighteen trip starts logged at longitude −7.2 between
  1 and 5 August, against a stored sign of +1 written by one bad frame after the last of them.
  That is why his trips read correctly while only the marker sat out at sea — the poller had the
  right positions on disk and the web was applying the wrong sign over the top of them.

  🔑 The history is counted in places, not rows. While the car sleeps the cloud re-serves one
  frozen frame and Mate still records it every 30 seconds: he banked ~1900 identical rows over
  sixteen hours. Counted as rows, one stuck frame would outvote real driving 60:1 and confirm the
  very bug this undoes. A sign is only adopted on a near-unanimous majority, so a car that genuinely
  moved across the line is left alone rather than guessed at.

• 🚗 A trip no longer stays open because the cloud stopped talking. Ending a trip needs the
  cloud to say gear P, six times running. When it instead freezes on a frame that says D — it
  re-serves the last frame it holds, indefinitely — those readings never come: the car is parked in
  the drive and Mate reports Driving for the rest of the day, with the parked hours swallowed
  into the trip. Reported by @riri19 (#233): "stuck in Drive mode for hours, sometimes all day."

  After half an hour of one repeated frame, the trip is now given up on and closed.

  🔑 The threshold was measured, not chosen. On his 40 246 polls: while the car is genuinely moving
  a fresh frame arrives every 18 seconds (median), 36 s at the 95th percentile and 9.4 min at
  the 99th — even on a link as poor as his — and of the 17 frozen stretches longer than ten
  minutes, all 17 ended with the odometer on the exact value it started at. Not one was a car
  still driving. Replaying his twelve days through the fix: the longest trip falls from 11 hours
  24 minutes to 61 minutes, the three multi-hour trips disappear, and the total distance moves by
  1 km out of 793.

• 📍 A trip that opened late now starts where the car set off, not where the signal came back.
  v1.16.5 already moved the distance and the energy back over kilometres driven while the cloud was
  dark (#130). The start POSITION was deliberately left behind, on the grounds that a frozen frame's
  GPS is often 0,0 and would plant the trip in the Gulf of Guinea. That threw away the good
  coordinates along with the bad: @riri19's 19 km drive opened 5 km from home while the frame in
  Mate's hand still said parked at home and was 32 minutes stale. Mate now anchors to the last
  position it actually held — and simply never stores a zero as one, so the Gulf of Guinea case is
  unchanged.

• 🕰 "Last seen" now says when the CAR reported, not when Mate wrote a row. The poller records a
  position every 30 seconds while parked, and it keeps recording them while the cloud re-serves one
  frozen frame — so the row is always seconds old, whatever is in it. The Overview map popup and the
  status card both showed that row time: "22 seconds ago" over a marker that had not moved in
  sixteen hours. It is the figure @rop12770 was looking at in #232 while asking why his car was in
  the sea, and on a link like @riri19's (#233) the true age is 78 minutes at the median.

  Nothing new is computed — the honest age was already measured from the frame's own clock on every
  poll, and simply never reached the page. The amber mark that flags a link lost mid-drive stays,
  now as a colour on that one figure rather than a second number beside it.

• 🔁 The Trips and Statistics pages no longer reload themselves out from under you. Pick last
  month, read for half a minute, and the idle auto-refresh put you back on the current one — with
  the statistics period reset too. Reported by @michapr (#236).

  The month and the period are held in the page, not in the address, so location.reload() could
  never restore them. The idle guards added for @pdifeo in beta #22 could not help either: they
  protect a field being typed in or a mode switched on, and a choice already made and now being
  read looks exactly like an idle page. Both pages show history, so there is nothing live for a
  reload to fetch and the refresh button is in the header. Every other page keeps its 30 seconds.

Changed

• 🩺 The diagnostics bundle now shows the sign of each GPS signal, not just that it arrived.
  The line tested only for "not empty", so a signal that arrives as zero — the axis saying
  nothing — was listed as present alongside real ones. It reads 2:−, 3:+, 3724:+ now, with zero
  called out by name. One character per axis: it narrows the car to a quadrant, which the remembered
  sign printed directly beneath it already did, and no magnitude ever appears. Chasing #232 through
  the old line cost two hours and produced a diagnosis that was wrong.

---

Update the add-on: Add-on Store → ⋮ → Check for updates → Update.