- Decision
- ADR-176
- Date
- 2026-10-18
- Status
- ProposedProposed (owner request 2026-10-18; patch
frameworks-base/0009host-tested, 35 of 35 mutants killed; landed on the build server 2026-10-18 as3b0677c8531a; module build ofservices,SystemUIandSettingspassed; inlynximage2026102101, flashed 2026-10-21;GRB-018partial on 2026-10-25: a set-down from about 3 cm locks, so the threshold is too strict for daily use and a redesign is open, seestatus-privacy-and-identity.md). Amended by ADR-195: an impact after a short free fall is a set-down and does not lock, and the strictness is a setting). - Part
- Grab detection
- Mentions
- ADR-169, ADR-195
- Mentioned by
- ADR-195, ADR-260
1Decision
Amend ADR-169: add an impact route to grab detection. A hard acceleration peak (default 30 m/s², key grab_impact_peak_dms2, 0 turns it off) while the phone is unlocked and interactive triggers the same lock and 60-second restart at once, with or without touch and whatever the phone does afterwards. The pull rule of ADR-169 is unchanged and stays the only route for a peak below the impact threshold. A drop onto a bed or a hard set-down locks the phone too.
2Why and evidence
The owner reported that a grab with no finger on the screen, or one after which the thief stands still, never locks, and that a lock after a drop is no harm. [measured] on lynx, 2026-10-18 (GRB-004, GRB-005): 8 staged grabs with the other hand all failed only the continued-movement condition (energy 0.18 to 1.50 against 1.5); 6 of 10 pulls without touch locked only through the catching hand.
Of 59 untouched peaks logged that day 39 were at most 27.0 m/s² (handling, pickups, gentle pulls); the 20 above it were bed drops (33 to 114), hard set-downs (29 to 171) and one unexplained 54.3. The staged grabs with a hand on the screen were 31 to 124.
Rejected: a hold-state rule (movement in the hand against at rest), because rest on a table measured 0.018 to 0.43 against 0.10 to 1.7 in the hand, so the two overlap; a free-fall check, because a lock after a drop is accepted.
3Consequences
A new route in GrabDetector and one key; no new permission, Binder interface or sensor. The margin is thin (27.0 against 30) [measured]. Whether a real snatch peaks above 30 is [assumption]: nobody else grabbed the phone. More locks in daily use are expected from a hard impact in a bag or pocket with the screen on and from running with the phone in the hand; a false lock costs a PIN entry, or a restart after 60 seconds without one. The recorder samples at 200 Hz and the service at SENSOR_DELAY_GAME, so only the service's own log decides the threshold.
4Review when
Daily use on stallion shows more than one unexplained impact lock a day; a real snatch peaks below 30; a trusted-device route is designed.
Read from docs/governance/decision-log.md, entry ADR-176.