Playbook: the daily-driver push

scope: generic · severity: technique · confidence: proven · subsystem: method

Goal: the device is good enough to carry.

Done when: you carry it. There is no other test, and every proxy for it lies.

What actually blocks it

In the order they bit on taimen, which is not the order anyone expects:

Suspend. A phone that does not sleep is a phone that is dead by lunch. This dominates everything else. 40-suspend
Stability under real use, not under test. Statistical bugs that a deliberate test never triggers.
Performance. App launch, scroll, wake latency. Measure two runs per arm and know your noise floor — critical-chain-shows-the-longest-path-not-the-floor is the cautionary tale.
The radios. 50-wifi-bt-modem
Camera. Almost always last, almost always the hardest.

The discipline that makes this phase work

Measure before and after, with a noise floor. A performance change without a noise floor is a story. Two identically-configured boots on taimen differed by 0.35 s, which is larger than most of the “improvements” anyone proposed.

Audit rather than assume. tools/ph-daily-audit.sh prints evidence, not verdicts — deliberately. A tool that prints a verdict invites you to skip reading the evidence.

Keep a support matrix with evidence attached. “WiFi works” is not a status. “WiFi: associated, survived 40 rekeys and 20 suspend cycles on , zero restarts” is.