Why task success needs a reliability context
A completion rate can look stable while the experience becomes slower or less predictable for a smaller audience. Segmenting by device, release and performance condition can expose hidden service debt. Start by pairing one meaningful outcome with one technical condition, then ask whether the relationship is consistent.
The danger of a single health score
A composite score is attractive because it makes comparison easy. It can also conceal opposing movements and arbitrary weightings. If you use one, keep the underlying measures visible, document how it is calculated and define which decisions it should never make alone.
Operational demand is app evidence
Support contact, manual workarounds and repeat attempts often reveal issues event tracking misses. Agree a small set of operational categories that can be matched to journeys without collecting unnecessary personal detail.
Measure releases as interventions
Before launch, write down the intended mechanism of improvement and possible side-effects. This provides a more honest basis for evaluation than searching afterwards for whichever metric moved.