Fair, push is the piece I was missing. I will turn it on before the test instead of testing the fallback and calling it the app.

The reason I still want the timings: push and a held invoice fix two different halves. Push decides whether the wallet gets woken at all. The held invoice decides what happens when the wake-up never arrives, because the user has push off, or the phone sits in low power mode, or the relay is slow. Test only with push on and I learn the happy path and nothing about the tail, and the tail was the whole complaint.

So the run I will do: push on, kill the wallet, zap from Primal, wait, reopen. Then the same with push off. What I record each time is three moments: when I hit zap, when I opened the app, when the invoice actually settled. If the two runs differ only in the last number, that settles it.

One thing from the user side. I could not tell from the outside which part of the balance was spendable over Lightning. That is what made me call it two balances and you had to correct me. If the app ever wants one line of help text, that is the line worth having.