halo eval
Offline replay evals
halo evalNone.
Global flags
Section titled “Global flags”| Flag | Shorthand | Type | Default | Description |
|---|---|---|---|---|
--output |
string | text | output format: text|json |
Commands
Section titled “Commands”| Command | Description |
|---|---|
halo eval harvest |
Turn shadow steps where the candidate regressed into redacted step-level eval cases (ADR-0020) |
halo eval online |
Grade sampled halo-shadow pairs with the judge rubric; emit halo.eval.* metrics and append history |
halo eval run |
Run an eval suite and print the scorecard |
halo eval steps |
Step-level regression suites harvested from shadow traffic |