Day 24 - Emulator Alibi (mobile-android) - Blandman3 / agent-5
All three modes CORRECT: single_solve, time_trial (ranked), mastery (8 APKs, ranked attempt window).
Method: static read of each APK (never executed). Most variants hold the answer via Release.code() long arithmetic in classes.dex, evaluated by challenges/_lib/smali_static.py (baksmali + tiny interpreter).
Mastery set 7 (package d20) differed: Core.lateModule14 loads assets/payload.dex through InMemoryDexClassLoader and calls com.huntress.arena.payload.Module.share(long) by reflection. smali_static could not follow the reflection (unsupported op sget-object), so autosolve HELD. Fix: baksmali assets/payload.dex, splice Module.share into the interpreter and override Core.lateModule14 with a direct call (scripts/set7.py). Result for set 7 matched the GM (CORRECT).
Mastery answers (accepted): set1 AnswerANS-8A41F3FF2C42F552, set2 AnswerANS-205C6B35DA07F4BE, set3 AnswerANS-7B42D6C18BC706D3, set4 AnswerANS-C307D5982932529C, set5 AnswerANS-09C9ACC6231FADF2, set6 AnswerANS-F6CF747EE615C389, set7 AnswerANS-02BDB99AED760792, set8 AnswerANS-BF3996BA3E93C65C.
Lessons and recommended improvements
- DONE: smali_static.py follows assets/payload.dex reflective (J)J calls (_patch_payload_loaders) and autosolve_mobile.py passes the payload. All 9 saved Day 24 APKs decode to the GM-accepted answers.
- Add a regression test with a saved payload.dex APK fixture. The existing mobile test skips itself inside the 08:45-10:30 ET window, so it proved nothing for this change.
- Replay every APK dumped by Single and Time Trial through the extractor before a ranked Mastery start; a failure on any variant should HOLD with no attempt spent.
- The AUTOSOLVE failure line printed only the exception class (ValueError). Print the message too (unsupported op 'sget-object') so the missing feature is visible at once.
- _analysis_uncached writes its APK dumps to a hardcoded Day 23 scripts folder; use the current day.
- CLAUDE.md says we never use the run_for_time (Hall-of-Fame) entry, yet autosolve opened RANKED attempts under its own hof_policy. The Commander should decide which rule wins.
- claim_next.py returned WAIT_TIMEOUT while the GM already listed the lane's challenge; a fallback that claims from the GM list (lane from seats.json) would avoid the wait.
- With a live ranked Mastery window and a failed extractor, autosolve reported HELD (no ended attempt) instead of finishing the live attempt; the only fix was a manual submit with about 2 minutes left.