diff --git a/docs/review/device-qa/README.md b/docs/review/device-qa/README.md index 253eadf..f981e37 100644 --- a/docs/review/device-qa/README.md +++ b/docs/review/device-qa/README.md @@ -364,5 +364,6 @@ Each note records one fact about bytes or draw calls already on main. None of th | [socket-lists-2026-11-06.md](socket-lists-2026-11-06.md) | The socket lists, not a rejection | | [medal-directory-2026-11-06.md](medal-directory-2026-11-06.md) | The medal directory, DailyMedals | | [patrol-spawn-2026-11-06.md](patrol-spawn-2026-11-06.md) | The patrol spawn, without fairness | +| [retest-2026-11-06.md](retest-2026-11-06.md) | Debug autopilot, then the owner marked stutter and the picture failed | Next session: follow [retest-checklist.md](retest-checklist.md) (D-105 allows device checks on any build). diff --git a/docs/review/device-qa/retest-2026-11-06.md b/docs/review/device-qa/retest-2026-11-06.md new file mode 100644 index 0000000..3cf5bad --- /dev/null +++ b/docs/review/device-qa/retest-2026-11-06.md @@ -0,0 +1,47 @@ +# Device retest 2026-11-06 + +Debug autopilot on a physical iPhone 17 Pro, then one human play of the restored Release build. GitHub #117 is closed. This record does not pass the retest and does not reopen the issue. T903, T904, and T905 stay open. + +The tree is runtime `1e9ac9d`. The binary was a Debug build of that tree plus one uncommitted change in `DebugAutopilot.swift`: `-SSAutopilotProfile` selects `competent`, `stealth`, `loud`, or `firstRun`. Release does not compile the autopilot. After the runs, the phone was put back on the Release build of `1e9ac9d`. + +Device: iPhone 17 Pro, product `iPhone18,1`, iOS 26.3.1. Developer mode was on. The cable was attached. Launch arguments were `-SSQA`, `-SSAutopilot run`, `-SSAutopilotUpgrade ricochetPulse`, `-SSAutopilotProfile` as named below, `-SSDaily`, and `-SSMute`. The daily seed was `14304826796902850030`. Sound was muted. No person played. Shipped art from that tree was in the build. Benchmark review art was not installed. + +Each launch is a new process, so the consecutive-success count starts at 0. The two successes are not three consecutive completes. Atlas bytes are nil. Memory warnings were 0. Serious or critical thermal samples were 0. + +## Runs + +| Style | Profile | Ticks | Outcome | Medals | Takedowns | Gate C | +|---|---|---|---|---|---|---| +| Natural | competent | 15142 | failure | none | 10 | false | +| Sneaky | stealth | 16443 | success | GHOST, SWIFT | 17 | false | +| Loud | loud | 18326 | success | SWIFT | 2 | false | + +None is SURGICAL. Resident bytes were 127238144, 125386752, and 130908160. + +The last `-SSQA` sample on each run: + +| Profile | n | p50 | p95 | p99 | worst | +|---|---|---|---|---|---| +| competent | 17399 | 16.669500 | 16.669667 | 39.740958 | 235.779792 | +| stealth | 18599 | 16.669500 | 16.669667 | 41.143833 | 166.693333 | +| loud | 20098 | 16.669500 | 16.669667 | 32.963042 | 131.191708 | + +`frameTimePassesGateC` is false on each device-run record. Those samples are the Debug autopilot with the console attached. They are not a measurement of the Release session below. The damage export is large and is not attached. Competent wrote 46 damage records and 15142 commands. Stealth wrote 56 and 16443. Loud wrote 31 and 18326. + +## Owner mark + +The phone was on the Release build of `1e9ac9d`. The owner played it and marked the game unplayable. Stutter is FAIL. The picture is FAIL as a whole. The four visual rows were not marked one by one. + +## Checklist rows + +| Row | Mark | +|---|---| +| Camera pieces | not marked separately; the picture as a whole is FAIL | +| Invisible blocking | not marked separately; the picture as a whole is FAIL | +| Naming enemies | not marked separately; the picture as a whole is FAIL | +| Damage with no visible source | not marked separately; the picture as a whole is FAIL | +| Stutter | FAIL | + +RS-025, RS-026, RS-030, and RS-031 were not performed. + +The human rows in [retest-checklist.md](retest-checklist.md) stay failed.