What Makes a Second Source Independent
Why three systems reading one badge are one source, the four kinds of independence that matter, and how to count sources honestly before relying on them.
Five records of one arrival, sorted by what they actually depend on
Reads the badge.
Reads the same badge.
Reads the same badge again.
Saw a face; genuinely independent.
A different credential, on a shared machine.
Three of the five are one source: the badge. Only the colleague is independent of the credential, and the sign-on is independent of the badge but not of a person. This is one employer's own file.
Three systems reading the same badge are one source. If the badge was presented by somebody else, all three are wrong in the same way, and their agreement tells you only that the badge moved through the building — which was never in doubt.
The evidential discipline in “What Makes a Second Source Independent” should also govern workforce technology. A team evaluating what teams should check in how to detect mouse jigglers for how to detect mouse jigglers can use time and project records as operational context, but should preserve the original record, document access and let the employee correct a misleading entry before it supports a conclusion.
Counting them as three is the most common error in corroboration, and it is encouraged by the fact that they appear in a report as three rows from three systems.
The California privacy regulations offers another lens on the issue raised in “What Makes a Second Source Independent”. Compare its principles with the actual record, ownership model and review route rather than importing a generic checklist unchanged.
The four kinds of independence
Independent of the credential: does this record depend on the same card, code or device? A colleague who saw a face does not. A second reader does.
Independent of the clock: were these timestamps produced by the same clock, or by clocks whose relationship has been measured?
Independent of the person: could the person have produced this record at will? A self-reported check-in could; a customer's signature could not.
Independent of the observer: does this come from somebody with a stake in the outcome?
A record that is independent on all four is strong. Most are independent on one or two.
Counting sources honestly
Write them out, with what each depends on. The exercise takes ten minutes and regularly reduces five records to two genuine sources.
That is not a weakening of the case. It is the case, accurately described — and a case built on two acknowledged sources holds up considerably better than one built on five that collapse into two when questioned.
The strongest available independent sources
A person who saw a person. A customer or third party who recorded something. A physical output that exists or does not. A system the employee does not control and did not know was recording.
The third is underused. Work that was done leaves traces that are not timestamps — a repaired machine, a completed form, a delivered item — and those are independent of everything in the time system.
Self-reported records
Anything the person themselves entered — a timesheet, a check-in, a ticket — is evidence of what they recorded rather than of what happened.
That does not make it worthless. It makes it a different kind of record: good evidence of what they said at the time, which matters if their account changes, and poor evidence of the underlying fact.
Independence of the clock, briefly
Covered in its own right elsewhere here. The short form: two records from systems whose clocks have never been compared cannot corroborate each other to the minute, because the apparent agreement or disagreement may be entirely clock drift.
If a case turns on minutes, the clocks have to have been measured.
Negative corroboration
The strongest kind and the least used: a record that would exist if the account were false, and does not.
If somebody says they were on site and the vehicle tracker shows the van at the site, that is corroboration. If the tracker shows it elsewhere, that is contradiction. Both are more useful than another reading of the badge, and both should be checked whichever way they are expected to go.
Counting them on paper
The count takes ten minutes and belongs in the file as a table rather than as an assertion about how many systems agree.
| Record | Depends on | Independent of the badge? |
|---|---|---|
| Door reader | The badge | No |
| Time terminal | The badge | No |
| Car park barrier | The badge | No |
| Colleague's account | A face | Yes |
| Workstation sign-on | A different credential | Yes, but shared machine |
Written out, five records resolve to two genuine sources. That is the case, accurately described, and it is far more durable than a claim that five systems confirm anything.
Looking for the source that would settle it
Having counted what you have, the more useful question is what would actually resolve the matter — and then whether it exists.
Frequently it does: a customer who signed something, a colleague who was there, an output that happened or did not. Asking what would settle this is a better use of an afternoon than analysing what is already in hand.
Writing the count down
In the file: each source, what it depends on, and whether it is independent of the others. Four lines in a table.
That table is what prevents the overstatement that undoes these cases — "five separate systems confirm" — and it is also what directs the investigation towards the one kind of evidence that would actually settle the question, which is usually a person rather than a machine.