@ledgelm/evals

Run any user-owned function, assess its output, and buffer a normalized LedgeLM result.

import { evaluate } from '@ledgelm/evals';
import { flush } from '@ledgelm/reporter';

await evaluate({
	name: 'support-answer',
	input: { question: 'Can I return this?' },
	run: ({ question }) => myAgent(question),
	assess: ({ output }) => ({ passed: output.includes('30 days') })
});

await flush();

Models, credentials, test data, and judging remain in the caller's process. evaluate() reports one result and never flushes the shared reporter buffer.