@irisrun/evals

Reproducible evals — because determinism makes scoring repeatable. Run the same scenario and it replays byte-identically from its journal, so a score is a fact about the run, not a roll of the dice. The arbiter is reproducibility, not taste: the same case + scorer re-runs identically; a swapped tactic scores differently, but reproducibly.

What it is

runEval runs a deterministic scenario (a fresh store + scripted performers per run) on the core runTurn, then scores the recorded session via @irisrun/inspect; runSuite runs a set; reproduce re-runs a case to confirm byte-identical replay. This is faithful record-replay of captured effects — not a claim that a live model is deterministic. Depends on @irisrun/core + @irisrun/inspect.

Use it

iris eval ./evals/suite.mjs    # reproducible scenario scoring

See docs/Audit & reproducible evals.

Part of Iris — own, portable, verifiable state.