Keep a Log of Your Fairness Checks

If you ever check your AI system for unfair outcomes — and you should — write down the check. Not just the conclusion. The check itself.

"We looked at it and it seemed fine" is not evidence. It's a memory, and memories don't survive audits. What survives is a log entry: on this date, we tested this system, this is how we tested it, this is what we found, and this is what we did about it.

You don't need a statistics degree. The checks can be simple: do approval rates look consistent across groups? Are errors clustering somewhere unexpected? Would you be comfortable explaining the pattern to the people affected? Write down the question, the answer, and the date.

Here's why the log matters twice. First, it proves diligence — to auditors, clients, and yourself. Second, it creates a baseline. Six months from now, you can compare and see whether things are getting better or drifting. Without the log, every check starts from zero.

Fairness isn't a one-time verdict. It's a habit with a paper trail. Check, write it down, repeat.

I explain frameworks; I don't give legal counsel.