cayu.evals.workflow_recovery

Public Cayu API. Definitions are grouped by their public source module, and definitions from private implementation modules appear on their package’s page; adding a symbol does not move existing links. The import shown at each entry is a supported public namespace. Read the Python API overview and install the required extra.

capture_workflow_eval_attempt

from cayu.evals import capture_workflow_eval_attempt

Recapture an exact completed attempt without invoking application callbacks.

Supply the original target (including its original capture policy) and exact input. Omit output to use the private retained projection, or supply the trusted original projected output. bounds controls this revision. To reject changes since a prior successful capture, supply its evidence_sha256. Persist the returned document separately from the original EvalRun.

async capture_workflow_eval_attempt(target: WorkflowEvalTarget, source_trial: EvalTrialResult, *, messages: tuple[Message, ...], output: WorkflowEvalResult | None=None, bounds: SessionTrajectoryBounds, expected_evidence_sha256: str | None=None) -> SavedWorkflowEvalCapture

Release source · Runtime guarantees · Capability guides

import_workflow_eval_attempt

from cayu.evals import import_workflow_eval_attempt

Explicitly adopt a legacy report’s saved completion, without rewriting that report.

The caller supplies the original target/input and an explicitly projected result, plus the exact persisted attempt and completion IDs. This import establishes an anchor now, not a claim of an earlier full-evidence seal. Keep source_run unchanged.

async import_workflow_eval_attempt(target: WorkflowEvalTarget, source_run: EvalRun, *, case_id: str, trial_number: int, attempt_id: str, completion_event_id: str, messages: tuple[Message, ...], output: WorkflowEvalResult) -> EvalTrialResult

Release source · Runtime guarantees · Capability guides

SavedWorkflowEvalCapture

from cayu.evals import SavedWorkflowEvalCapture

An independently inspectable capture revision linked to the original report trial.

Bases: BaseModel.

Fields and attributes

Defaults below are source expressions; Field(...) includes schema/validation settings and default_factory creates a fresh value. Inherited declarations are included. Constructor behavior follows the model or dataclass shown in the bases/decorators.

Name	Type	Default
schema_version	Literal[1]	1
capture_id	str	Field(default_factory=lambda: str(uuid4()))
captured_at	datetime	Field(default_factory=lambda: datetime.now(UTC))
source_trial	EvalTrialResult	required
source_attempt	WorkflowAttemptAnchor	required
bounds	SessionTrajectoryBounds	required
evidence_sha256	str	required
trajectory	Trajectory	required

Release source · Runtime guarantees · Capability guides

SavedWorkflowEvalScore

from cayu.evals import SavedWorkflowEvalScore

Separate deterministic scoring revision; source execution cost is not incurred again.

Bases: BaseModel.

Fields and attributes

Defaults below are source expressions; Field(...) includes schema/validation settings and default_factory creates a fresh value. Inherited declarations are included. Constructor behavior follows the model or dataclass shown in the bases/decorators.

Name	Type	Default
schema_version	Literal[1]	1
score_id	str	Field(default_factory=lambda: str(uuid4()))
scored_at	datetime	Field(default_factory=lambda: datetime.now(UTC))
source_capture_id	str	required
target_revision	str	required
evidence_policy_revision	str	required
pricing_profile_fingerprint	str | None	None
source_evidence_sha256	str	required
assertion_revisions	tuple[str, ...]	required
assertions	tuple[EvalAssertionResult, ...]	required
score	float | None	required
model_calls	Literal[0]	0

Release source · Runtime guarantees · Capability guides

score_workflow_eval_capture

from cayu.evals import score_workflow_eval_capture

Freshly validate a sealed attempt, then run deterministic portable assertions only.

Model judge specs are rejected before store reads or scoring. Environment probes absent from the saved evidence remain unavailable; no tools run to recreate them.

async score_workflow_eval_capture(target: WorkflowEvalTarget, capture: SavedWorkflowEvalCapture, assertions: tuple[AssertionSpec, ...]) -> SavedWorkflowEvalScore

Release source · Runtime guarantees · Capability guides