Same solution and quality, but fuel differs by thousands of times — is this allowed?

General
1

While checking benchmark 0fdcd53cb6d2c4d809fad283a91ae002, nonce 53, for c005_a031, I reproduced the same final solution and quality using the official algorithm artifacts.

The benchmark declares aws_g4dn, while my replay used an RTX 5060 Ti, so the hardware environments were different.

Results:

  • Quality: MATCH
  • Solution SHA-256: MATCH
  • Protocol solution signature: MATCH
  • Submitted fuel: 38,940,396
  • Replay fuel: 168,907,578,475
  • Fuel ratio: approximately 4,338×
  • Runtime signature: MISMATCH
  • Submitted commitment: MATCH

I understand that commitment consistency does not establish correctness of the execution accounting.

Could you clarify the protocol’s reproducibility requirements?

  1. Must fuel_consumed and runtime_signature match a reference execution? If so, under which hardware/runtime conditions and tolerances?
  2. Are alternative execution paths, modified runtimes, or computation outside the standard instrumentation permitted, provided the final solution and quality are reproducible?
  3. Could the hardware difference explain a fuel discrepancy of this magnitude, and which reference environment should I use to make a valid comparison?

I am asking for clarification and investigation, not claiming a protocol violation.

image
image2126×752 94 KB