E55: a step ledger on both sides of the boundary catches and locates every fault in what crosses, with no reference run, across machines too

[tactino]

The question

E54 found two checks that together catch every boundary fault it injected, but each has a gap:

• byte identity needs a fault-free run on the same machine;
• the env-side record of episodes cannot see faults in observations or actions.

A boundary fault is a difference between what the environment took and gave and what the trainer sent and received. Does a ledger kept on both sides, compared step by step, catch and locate every fault in what crosses? Can it do so with no reference run, and across machines?

The answer

Yes. The environment-side ledger is a gymnasium wrapper (LedgerEnv) outside the client's code. The trainer-side ledger is E54's trace. ledger.py compares them env by env, step by step.

	runs	client faults caught and located	false alarms	ledger or record
one machine, seeds 0-2	216	189 of 189	0 of 27	207 of 207
across machines (laptop client, guangzhao SB3)	24	21 of 21	0 of 3	23 of 23

Every check and prediction holds (results/verdicts.txt).

• Each first difference was the fault's first changed value, in the field the fault touches.
• The ledger counted exactly the values each fault changed.
• Across machines, the fault-free run ended on d7a19cc0, not guangzhao's ec5b8226. Byte identity would have called it faulty; the ledger found no difference.
• Across machines, the status rule caught 0 of the 24 runs and E54's record caught 8.
• The wrapper changes nothing that crosses: none still ends on E50's weights.

The faults in the trainer's own log change nothing that crosses, so they stay with E54's record. Checking a digest of each step inside the protocol, as the run goes, is the follow-up.

The per-run traces and ledgers stay on guangzhao, including the laptop's side as cross-laptop-side.tgz. results/ and results/cross/ carry every verdict.