Reading the match % and its reasoning trail
The match percentage is not a single model score. It is a weighted roll-up of four component scores, and every one of them is inspectable. If a number looks wrong, the reasoning trail will usually show that the job description, not the candidate, is the problem.
What goes into the number
Skill coverage carries the most weight by default, at 45%. Relevant experience contributes 30%, measured as time spent in roles the model judges adjacent rather than identical. Trajectory — whether the candidate has been moving toward this kind of role — is 15%. The remaining 10% covers explicit requirements you marked as mandatory, and a failure there caps the total rather than reducing it.
Check this first
A candidate showing 48% with every skill matched almost always failed a mandatory requirement. Look for the red cap marker at the top of the reasoning trail before investigating anything else.
Changing the weights
Weights are set per job, not per workspace, under Job settings → Ranking criteria. Changing them re-scores the existing batch in place; it does not require a re-upload, and the previous scores are kept in the audit log so you can show a regulator what changed and when.