AI Development Test Record

A short record for one increment: what you checked, what should have happened, what actually happened, and what you did about the difference. It turns "I had a look and it seemed fine" into something you can point at.

To keep a copy, use your browser's print option — the page is formatted to print without the site navigation.

What to record

One row per acceptance check. The checks come from the behaviour and clarifications you accepted before the work started — if you have used the behaviour definition template, they are already written.

Condition checked Expected behaviour Observed behaviour Result Correction and recheck
 
 
 
 
 
 

What each column is for

Condition checked

The specific case, including the input. "Add an item with an empty name" rather than "adding items".

Expected behaviour

What should happen, taken from what you accepted before the work started — not decided now, in the presence of the result.

Observed behaviour

What actually happened. Write it even when it passes: "item appeared at the bottom of the list" is more useful later than a tick.

Result

Pass or fail. Nothing in between — "mostly" is a fail with an explanation missing.

Correction and recheck

For a failure: the smallest correction requested, and the outcome when the same condition was checked again. A correction that has not been rechecked is not finished.

A completed example

Condition checked Expected Observed Result Correction and recheck
Add item "Milk" Appears at the bottom of the list; input clears Appeared at the bottom; input cleared Pass
Add with empty input Nothing is added An empty row was added Fail Asked for empty input to be ignored, changing only the add handler. Rechecked: nothing added. Pass.
Add "Milk" twice Both entries present Both present Pass
Existing items after adding Order unchanged Order unchanged Pass
Page loads with no items Empty-state message shown Message shown Pass

Note the fourth and fifth rows. Neither was part of the change; both were checked because a change to the list could plausibly affect them. Those rows are where regressions get caught, and they are the ones most often left out.

When to bother

Not for every change. A one-line adjustment to a personal tool does not need a written record, and insisting otherwise turns a useful habit into an obstacle.

It earns its place when the work matters to someone other than you, when you are likely to be asked later what was checked, when a defect has already appeared once in the same area, or when you are working over several sessions and will not remember. In those cases the record is not paperwork — it is the difference between believing the software works and being able to show it.