Skip to content

Judging

Blind code review by LLM judges.

judging

LLM judges: blind code review of finished runs.

review

review(plans: Any = None, judges: Union[Profile, str, Sequence[Any]] = (), *, runs: Any = None, budget_usd: float | None = 1.0, concurrency: int = 4, rubric: list[Any] | None = None, guidance: str | None = None, include_static: bool = False, max_output_tokens: int | None = None, skip_reviewed: bool = True, wait: bool = True, quiet: bool = False) -> Union[Results, Job]

Blind LLM code review of finished runs by one or more judge profiles (cloud or local), then fresh Results whose code-quality score includes the reviews. Review whole plans, or only some runs with runs=. budget_usd caps the judges' spend. rubric and guidance replace the defaults (see judge_defaults(): a rubric is a list of {"key", "label", "description"} dimensions); include_static shows judges the static-analysis findings; skip_reviewed=False reviews again runs a judge already reviewed. wait=False returns the Job.

judge_defaults

judge_defaults() -> dict[str, Any]

LLMDyno's default judging setup: the rubric (dimensions with descriptions), the guidance judges get, and the default budget, concurrency and output limit.

reviews

reviews(plan: Any = None, run: Any = None) -> list[Any]

Reviews of a plan's runs or of one run (overall score, per-dimension scores, summary, issues, judge).

delete_reviews

delete_reviews(judge: Any, plan: Any = None) -> int

Delete one judge's reviews (a judge profile or its id), everywhere or only for one plan; returns how many.