Host tooling and console behaviour are different
The repository combines host-side build and packaging work with console-side research. Producing an artefact on a development machine can check format and build assumptions; it does not establish how that artefact behaves on a particular console.
Retries change the measurement
A single-attempt success rate and a session with repeated attempts describe different outcomes. A theoretical cumulative rate assumes independent attempts with a stable probability. Shared timing, stale state or correlated failures can violate those assumptions, so the arithmetic is a prediction rather than hardware evidence.
Recovery must be observed
The research direction includes bounded retries, watchdogs and hardware-aware timing. Their useful acceptance criteria are observable: whether a stalled run is detected, whether state is restored and whether another attempt actually proceeds. Intended recovery behaviour should remain distinct from a measured successful recovery.
Keep targets and results separate
A reliability target belongs beside repeatable hardware observations, failure counts and test conditions. Build completion and documentation claims do not provide those observations. This field note keeps the focus on that engineering distinction without publishing operational exploit instructions.