You stop to fix a tyre. Half an hour, the rig motionless, you do not move a metre.

Last week we fixed what your page showed: instead of forty-seven three-second lines, it shows your two real stops. That was a reading fix, and we said so at the time.

Here is what it did not fix. Underneath the display, the app still believed you had stopped forty-two times — and at each false resume it handed a few seconds back to your moving clock. Across that half hour it credited you 28:02 of stopped time instead of 29:09. A minute and seven seconds of rest counted as riding, and a "moving average" wrong by the same amount.

This was not a display to fix. It was the measurement.

What the receiver says while you stand still

We exported the run from the phone: 7,343 positions, one per second, with no gap at all. Through the thirty minutes of repair, the rig did not move. The receiver did not stop talking.

Over those thirty motionless minutes it reported speeds of up to 2.30 km/h. Twenty-four times it said "at least 2 km/h".

That is not the problem. The problem is what we had measured a month earlier, on two bike rides: when you push your rig by hand out of a genuine stop, the receiver reads from 1.26 km/h upward.

Put the two side by side:

Pushing sits entirely inside standing still. There is no number, no threshold, no setting that separates those two situations. It was not a botched calibration: it was the wrong quantity.

A different question

Resuming a run therefore no longer asks "how fast is it going?" but "did it move?". Until the rig has covered ten metres since the stop began, no fast fix, no jolt, no receiver noise restarts the clock.

Ten metres is about two seconds of a moving team. A genuine departure goes straight through.

And the number was not chosen on the run that raised the problem — that is the surest way to be wrong everywhere else. It was measured against every real trace this project keeps:

What this does not do

Your stop list itself will not change much. It has shown your two stops since last week. What changes are the DURATIONS: the longest stop on that run goes from 26:56 to 28:03, and the amber band drawing it on the curve widens accordingly, since it is drawn to the scale of time.

Your past runs do not change. A run's stops are computed when you press STOP, and written once. We do not rewrite what your app has already recorded — not even to improve it. This is not "not yet": it is a rule. Your logbook says what it said the day you closed it.

On foot, nothing changes — and that is deliberate. We tried: with this guard, a walking trace recorded in snow went from 48 to 82 seconds of stopped time. A musher on foot stops without ever covering ten metres, so the guard would turn every halt into an endless stop. It is therefore off on the on-foot profile, and it stays off until a real walking trace tells us what setting would make sense there.

Walking beside your rig still counts as moving. You cover ground, so the guard lets you through. That is the wanted behaviour, not an oversight — and it is also why this guard does not, by itself, settle the case of a musher pushing alongside their dogs.

What we did not do, and why

Faced with "the app sees stops that never happened", the reflex is to change the threshold. We did not, and that is the most important decision in this release: the measurement shows the two ranges overlap, so any new threshold would have moved the error without ever removing it. Three tuning routes had been open for weeks; the measurement refuted them one by one before a single line of them was written.

We also accepted a cost: a genuine departure takes about two seconds longer to be recognised, and those two seconds are counted as stopped time. Across the real traces that amounts to between zero and seven seconds per run. That is the price of a half-hour stop no longer being cut into forty-two pieces.