# Systematic Playbook (No Guessing)

## Goal

Move from ad-hoc trials to controlled experiments where only one variable changes per run.

## Rule Set

1. One variable per run.
2. Always execute via experiment runner (creates reproducible artifact bundle).
3. Do not interpret UI without technical verdict + user observation in same run folder.
4. Stop a branch after 2 consecutive regressions.

## Command

Use this command for every run:

./tools/experiment-run.sh <profile> <udid> ReBreak

Optional timeout (seconds):

AB_SERVE_TIMEOUT_SEC=120 ./tools/experiment-run.sh <profile> <udid> ReBreak

## Run Order

1. current
2. partial-lite
3. partial-safe
4. legacy

## Decision Logic

1. If verdict is BLOCKED_FMI: disable Find My iPhone, re-run same profile.
2. If verdict is BLOCKED_PAIRING: re-pair with idevicepair, re-run same profile.
3. If verdict is BLOCKED_SERVE_TIMEOUT: keep same profile, lower noise, retry once.
4. If verdict is TECH_SUCCESS: evaluate UI from user-observation.md only.

## Promotion Criteria

Promote profile to default only if all are true:

1. TECH_SUCCESS in 2 consecutive runs.
2. Sprache + Region preserved.
3. Apple account preserved.
4. No registration failure screen.
5. Fewer or equal setup clicks compared to current.