How we work · principle

Evidence before recommendation

Separate announcements from observed behaviour and tested guidance, then show the source, scope and review date.

A new claim may be interesting immediately, but it becomes guidance only when the evidence is strong enough for the decision.

Use it on your own work

Bit asks what you are working on and applies this principle with you.

Ask Bit

Questions to ask

Is this announced, observed or tested?
Can a reader open the primary source?
What conditions and limits apply?
Would this evidence change a real recommendation?

0 of 4 answered

See the difference

Good boundary

A provider announcement is labelled Announced and linked to its original release. The product waits for repeatable tests before changing a workflow recommendation.

What not to do

A social post about a model result is paraphrased as a universal capability claim with no source, test conditions or date.

Watch for this failure

The pressure to appear current turns early impressions into advice and quietly erodes the trust the evidence layer is meant to create.

How it works

Official releases are the primary source for what a provider says changed. Independent reporting and practitioner observations can reveal behaviour, limits and context, but should not be presented as equivalent evidence.

Recommendations must state what is known, what remains uncertain, who the guidance applies to and when it was reviewed. Material corrections should remain visible rather than silently rewriting history.

The Human Bit method →

Build the capabilityRead the related AI 101 lesson →