# E015 — Final fixed latency sample v2

The preregistered extension is terminal:32 additional recipes,64 fresh native-speed processes,316,944 physical frame evaluations,658.015seconds. All pairs pass observed trace/sample pixel/exposed device equality; no failures or retries. Combined native cost:65 recipes,130 processes,647,650 frames.

Strict observed-prefix/ACT/actor deduplication yields37 distinct matching contact episodes overall. The preregistered first30 in fixed old-then-new recipe order contain **23 complete attributable timelines and7 censored by player death**. The remaining7 contacts remain in the complete dataset; they are not silently discarded. The minimum20-usable-event gate passes within the declared30-event sample.

Observed contact-to-counter delays in the sample are97,98,99,100,101,104,105,107,112,116,117,118,119,120,121 physical frames. The data directly reject a constant117-frame delay in these tested contexts. They do not isolate every cause of variation or establish independent boards: timing variants share eight roots and the inherited fixed board. Labels/contact contexts are privileged and scoped to the current ROM/DIP environment; whole-machine identity remains UNKNOWN.

Every selected event retains contact, crush animation, first disappearance if observed, actor retirement, score update and official counter clocks, counter association, competing events, phase termination and prefix hashes. Score changes are logged, never used to choose roots, recipes or outcomes. Counter decrements are delayed acknowledgements, not new kills. Censored events are not complete timelines.

This closes the latency prerequisite only. E0056→0 still requires the opt-in controller, validated objective/event reader, healthy stable root and final executable preregistration. E017 fixed stop pulses are now started as the next finite diagnostic. No scoring search, N1b or historical search resumption occurs.

Evidence: `data/pengo/e015/native-latency-v2/result.json`, its registration/manifest and per-case receipts; all first-batch and census versions are retained. The public first-batch video remains explicitly scoped to its earlier17-contact/14-usable snapshot. This extension does not change that video or any historical artifact.