The Okhrana put surveillance men on Rasputin and they wrote down where he went, who came to the flat, what time the lights went off. The logs called him "the Dark One." Nobody kept those logs to settle an argument about whether he was a saint or a fraud — they were filed for a superior who, as far as anyone can tell, never acted on them. And that is exactly why they win. Every memoir about Rasputin was written by someone with a stake: the killers wanting to look brave, the believers wanting a martyr, the court wanting to forget. The watch register wanted nothing. It just recorded. Moynahan's whole method leans on this: when the bored cop's daily log disagrees with the dramatic confession, trust the log.
First framing: a record made for an indifferent recipient is more trustworthy than a record made for an audience. Intention to persuade is the contaminant. The Okhrana logbook is clean because it was never meant to convince anyone of anything.
Second framing: this is a byproduct principle. The most reliable evidence is the stuff generated as exhaust from some unrelated process — bureaucratic, automatic, incidental — not the stuff produced to be evidence. A memoir is testimony built to be read. A surveillance log is residue. Residue doesn't flatter.
Third framing: the failure mode is obvious once you name it. The principle holds only while the record-keeper has no stake. The second surveillance itself becomes contested — the second someone knows the log will be read in a trial, or wants the target framed — the watch becomes just another confession wearing a uniform. The cleanness isn't in the form (a log), it's in the indifference.
A standalone cross-domain concept page — "byproduct evidence outranks testimony" — that travels to far cases: ships' logs vs captains' memoirs; server access-logs vs a defendant's account; a calendar app's automatic timestamps vs someone's recollection of where they were; metadata vs statements. Each case is the watch beating the confession. The essay version is a methodologist's tool: before you weigh two accounts, ask which one was made to be read.
Promote to WORKBENCH when it survives two non-Rasputin test cases (one historical, one contemporary/forensic) and when the failure boundary is stated sharply enough to falsify — i.e. a named case where a "byproduct" record was in fact motivated and misled. If the indifference condition can't be operationalized, it stays a spark.