simple-scienceconditionalweight 0.5adapter 8 of 17SimpleScience Evals.
The SimpleScience adapter runs the SimpleScience baseline: multiple-choice science questions scored deterministically. Like SimpleMath, it establishes a reasoning floor and carries a reduced weight so baseline checks corroborate rather than dominate the composite.
What it measures.
Answer accuracy
Share of multiple-choice science questions answered correctly.
Deterministic scoring
Fixed answer key, no model-judged grading.
Component keys.
simple-science.scoreNormalized correctness score on the 0–100 component scale.
When it counts.
Conditional — counts when SimpleScience results exist for the subject.
When signals are missing.
Absent results leave the component missing and out of the computation.
How it contributes.
{
"component": "simple-science.*",
"scale": "0-100",
"mode": "conditional",
"weight": 0.5,
"contribution": "component x 0.5 / total weight"
}The adapter’s components are averaged into the composite as a weighted mean: every contributing adapter’s component is multiplied by its weight, summed, and divided by the total weight of contributing adapters. The result is published as trustScores.total with a config version and timestamp so anyone can recompute it.
Questions about SimpleScience Evals.
What does the SimpleScience adapter measure?
It measures correctness on the SimpleScience multiple-choice evaluation and normalizes it into the simple-science.score component.
How is SimpleScience different from SimpleMath?
SimpleScience covers factual, multiple-choice science questions while SimpleMath covers exact arithmetic. Both are deterministic floor checks with the same reduced weight.