Pengo E014: a target cell is skipped
Two fresh native-speed reproductions of the unchanged E013 upward-route tape confirm that the existing deterministic controller can miss a cell decision point. The sixth MU starts at7383 expecting player y48. The observed player and sprite coordinates agree at the checkpoints: y49 at7398, y47 at7399. Neither observation is aligned. The controller waits for exact y48 equality, keeps UP pressed, reaches y32 at7414 and declares blocked at7533.
Full declared RAM/sprite/driver traces equal the accelerated E013 reference and both native replicas. Sampled native pixels and exposed CPU/driver states also agree between replicas. Both native wall times124.609s; batch125.031s; 15,066 physical frame evaluations. No observed death or life fault occurs in the replayed ACT3 suffix7306..7533. No new input was chosen or appended. Complete-machine identity remains UNKNOWN.
This confirms one specific limitation of the exact-equality controller. It does not establish a general speed rule, show that every cell can be a stable decision point, validate a new controller, verify egg/diamond mechanics, complete ACT3 or improve the best trajectory. The historical controller and all121 pinned sources are unchanged. Original E003 remains incomplete.
Before any progression trial, a separate opt-in controller must explicitly handle target crossing, input polling, alignment/recovery and safe stopping. Its tests must include this unchanged reproducer and ordinary move/push cases. Fixed timing variants must be preregistered before emulator outcomes; do not turn diagnostic adjustment into unreported adaptive score search.
Evidence: data/pengo/e014/controller-crossing-v1/registration.json and result.json; source tape SHA256 9b3ba202b9b45ed01b452bb6dd907b9af6d0afd019c98c0aa6747f0b0a1b914f; observed dense trace SHA256 83f0c35f1c8803114aec69824f9e0853a93909e6ad8651f4d1befde5ece992ca. Native screenshots, renders, symbolic board/object checkpoints and private RAM diff7398/7399 reside in the two owned E014 job folders on doppia-3090. No raw RAM or ROM is to be published.