Show the Facts Behind the Result

Every record now carries an explanation. The team is pleased with it, because it reads well:

This submission was classified as a strong fit given the company's enterprise scale, the clear alignment between their stated needs and our core capabilities, and the professional tone of the enquiry.

A seller reads it, agrees, and works the record. It goes nowhere. The company turned out to be well below the size GTM Lab sells to.

Nothing in that paragraph was checked against anything. It was produced after the outcome was chosen, by asking for a summary of why the answer might be right. It describes enterprise scale without naming a figure or where the figure came from. It is fluent, plausible, and unfalsifiable, which makes it worse than no explanation at all, because a seller can act on it.

An explanation shows what was consulted. A rationalization argues for the answer.

The distinction is the whole chapter, and the test for which one you have is blunt: could this explanation be wrong in a way somebody could notice?

“Enterprise scale” cannot be wrong, because it does not commit to anything. “Revenue on the matched account record: above threshold, observed 2026-08-14” can be wrong, and anybody looking at the record can catch it.

A generated summary is not evidence. Keep Your CRM Data Trustworthy made the point about generated fields, and it applies with more force here: a generated explanation is more dangerous than a stale one, because it is fresh, fluent, and it asserts things.

Show the inputs, with their values and their sources

The evidence display carries, for each input the decision actually used: what it is, the value it had at decision time, where that value came from, and when it was last observed.

For the record above, that display would have shown a matched account present at decision time, annual revenue below threshold from the account record and observed 41 days ago, and the message and use case judged by the interpretive step. It would have caught the failure instantly. The revenue was below threshold and on the record the whole time. The prose explanation talked past it.

Note the observation time. A value observed forty-one days ago is a different quality of evidence from one observed at decision time, and a reader can only weigh that if you show it.

Show what was not used

Two absences are worth displaying, and both are cheap. Required evidence that was missing, because a reader needs to know the decision ran on partial information. And evidence deliberately excluded, where the decision card marked something prohibited. This one looks like overkill until the first time somebody asks whether the system is using something it should not be, and you can answer by pointing rather than by investigating.

Carry this into your business

Find an automated decision at your company that writes an explanation and check whether any sentence in it could be proven wrong by looking at the record. If not, it is a summary of the answer rather than a reason for it, and it is teaching people to trust something that was never checked.