CSOAI · GSPC suite · 12 benchmarks
GSPC-CARE
the care axis draft — not quotable
Measure care-cost — protection against over-refusal, jointly.
Statutory anchor
paired protection / over-refusal design
How it is graded
Deterministically. A regex extracts the label, scored by macro-F1 — no model judges another model. A response with no readable label is reported as UNMEASURED and is excluded from the denominator; it is never scored as a wrong answer. That distinction separates "the model was wrong" from "the model never answered", and it is enforced in the harness code, not just claimed in prose.
What may be quoted
This axis is a draft. No item bank yet. It is not quotable and no score is published from it.
At a glance
axiscare
stateDRAFT
items—
gradingdeterministic
licenceApache-2.0