D049 - obscene-tool crack recovers names from NIDs by guessing, and says what a miss is worth

Status: decided.

A NID is the first eight bytes of SHA-1(name ‖ suffix). Hashing is one way, so the only route back is to hash candidates and look for a match. That is the whole tool.

It is not decode, and the two are easy to confuse. decode reverses the encoding - eleven printable characters back into the eight bytes they stand for. Both directions of that are cheap and neither touches the hash. After decoding you have the hash and you are exactly as stuck. The confusion is worth naming because the command that sounds like it does this does not.

Hits are proof; misses are nothing. A match is certain, because the hash agrees. A non-match says only that the candidate list did not contain the name - never that no such name exists. Presenting the two alike would turn "we did not guess it" into "it is not there", which is the kind of confident wrongness this project keeps having to design against. The output states how many candidates were tried so a reader can weigh a miss.

The generator is measured, not assumed. --known takes pairs already established and reports how many the candidate list reproduced. A list that cannot regenerate names already known is not ready to be believed about unknown ones, and the tool warns on standard error when it falls short. data/nid-corpus.txt holds 389 such pairs.

The output carries its own provenance. Suffix, candidate count, reproduction rate, recovery count - because a generated table outlives the run that made it and gets trusted long after it stops being right. Same argument as provenance on a check (D044).

Why it did not exist until now. Every check so far started from a name already known, so nothing needed the reverse direction. It becomes necessary the moment coverage starts from the machine's list rather than one this project wrote, which is what the enumerator (D043) introduced. It was blocked behind that, not overlooked.

Why it lives here and not in the emulator. The hasher, the corpus and the validation set are all here, and an emulator never cracks anything - it resolves NIDs by table lookup at load time. The complementary job, which only a loader can do, is recording which NIDs it failed to resolve. That list is the ranking worth spending candidates on, and it does not need names to be useful.