Applies to: ... line, and behavioral guidance. A proposal records a change set against a specific accepted base revision; a null change deletes a file. Ordinary fresh executions receive accepted rules, while candidates are visible during authoring and explicit trials.
Task completion in an execution report describes runtime execution; use an independent verifier for business success. A reported rationale is agent-authored evidence, not proof of improvement.
Records are frozen Pydantic models with unknown fields rejected.
model_dump() / model_dump_json() serialize them; model_validate() reconstructs them. Validators shown below enforce record shape, not behavioral correctness.
On this page
EvaluationOutcome, ExecutionReport, FailureNotice, ProposalOutcome, Rule, RuleProposal, TrialArm, TrialCase, ValidationVerdict.Example
EvaluationOutcome
Fields and properties
ExecutionReport
Fields and properties
FailureNotice
Fields and properties
ProposalOutcome
Fields and properties
Rule
Fields and properties
Methods
safe_path
safe_path
applicable_markdown
applicable_markdown
RuleProposal
Fields and properties
TrialArm
Fields and properties
Methods
normalized_scores
normalized_scores
TrialCase
Fields and properties
ValidationVerdict
Fields and properties
Signatures and fields are generated from the SDK source. See the import index for available entry points.

