The tool said nothing
We ran a comparison-group before/after study on Barry Road, Campbellfield. An earlier version of our engine had minimum crash-count floors. The study fell under them, so the engine printed a single sentence — low power: not a conclusion — and no numbers.
Hiding a result is also a claim
A refusal looks careful, but it withholds information a decision-maker is entitled to, and it lets the reader assume whatever they like. Recognised practice is to publish the estimate and let the interval carry the weakness. We rebuilt the engine so every valid study returns numbers, labelled by their uncertainty.
What the data say
What we will not claim
We will not say the treatment cut crashes by 25%. The result is consistent with a reduction, but not evidence of one: the interval runs from a 70% reduction to an 85% increase. More after-period data will narrow it.