What was tested?
Task, data, prompt, budget, and evaluation setting belong beside the claim.
A reference in progress / 00
Read the result. Then read the conditions.
A careful guide to benchmark evidence: what was measured, under which conditions, and where the conclusion stops.
No fictional scoreboard. The frame around a result is part of the result.
Evidence mapContext required
What changed the outcome?
Comparisons only help when their choices and constraints can be read.
Where does it stop?
Uncertainty and transfer limits are evidence too.