Module evaluation 

Provider-neutral execution facts and auxiliary evaluation primitives.

This module is an evaluation substrate, not a product reviewer. It records bounded execution facts, reads a consistent evidence window, and runs an isolated host-supplied evaluator. Rubrics, thresholds, findings, and business audit policy stay outside Code Core.

Re-exports

pub use crate::execution_identity::ExecutionIdentityV1;

pub use crate::execution_identity::ExecutionResultOutcomeV1;

pub use crate::execution_identity::ExecutionResultReceiptV1;

pub use crate::execution_identity::EXECUTION_RESULT_MAX_BYTES;

pub use crate::execution_identity::EXECUTION_RESULT_RECEIPT_SCHEMA_V1;

Structs

AuxiliaryCapabilityProfileV1

Explicit capability profile for an auxiliary run. The service never grants capabilities implicitly; a host can additionally provide a parent ceiling and the requested profile must be a subset of it.

AuxiliaryRunContextV1

AuxiliaryRunHandle

A cloneable capability to observe or cancel exactly one auxiliary run.

AuxiliaryRunOutputV1

AuxiliaryRunSnapshotV1

AuxiliaryRunSpecV1

EvaluationDispatch

EvaluationPlanV1

EvaluationRecordV1

EvaluationResultV1

A host-defined outcome token. Core validates shape and provenance only; it deliberately does not enumerate reviewer/business dispositions.

EvaluationSupervisor

Connects an append-only fact journal to a host policy, evidence reader, and auxiliary-run service. The policy is the only component that decides when an evaluation is useful; Core only enforces identity, bounds, and cleanup.

EvaluationWireEnvelopeV1

A strict, versioned envelope carrying one evaluation substrate value.

EvaluationWireKindDescriptorV1

One source-of-truth descriptor used by SDK artifact generation and parity tests. The payload type is documentation metadata; Rust validation still uses the concrete type in the match below.

EvaluationWireTypeV1

Canonical string constants for the version-one wire catalog.

EvaluationWriteOutcomeV1

EventCursorV1

Cursor for a run-local append-only event/fact stream.

EvidenceArtifactV1

EvidenceEventV1

EvidenceLimitsV1

EvidenceReadRequestV1

EvidenceRunStateV1

EvidenceSnapshotV1

ExecutionFactInputV1

Input accepted by the journal. Payload bytes are counted and represented by a digest; raw model/tool content belongs in an explicitly authorized artifact store or evidence reader.

ExecutionFactPageV1

ExecutionFactRecorder

Small adapter for runtimes that already receive RunEventRecord values. It makes the frame binding explicit at construction and keeps event ingestion out of product-specific evaluator code.

ExecutionFactSnapshotV1

ExecutionFactV1

ExecutionFrameV1

Runtime frame that records parentage without taking ownership of Cloud checkpoint/fork lineage. A child evaluation may point at its parent run; hosts remain responsible for business lineage and authorization.

ExecutionTargetV1

Session/run identity used by all evaluation records.

FactAppendOutcomeV1

FileEvaluationDispatchLedger

FileEvaluationResultStore

A durable implementation of EvaluationResultSink.

InMemoryAuxiliaryRunService

InMemoryEvaluationDispatchLedger

InMemoryEvaluationResultStore

In-memory CAS result sink. A durable host can implement the same trait with an append-only object store and an external retention policy.

InMemoryExecutionFactJournal

Bounded in-memory implementation used by the native harness and tests. Hosts may implement ExecutionFactJournal over a durable append log.

MemoryEvaluationDispatchLedger

RunEvidenceReader

Reader over the native in-memory run journal and optional artifact/fact stores. It captures state and the event window from one RunStore observation generation.

StructuredAuxiliaryExecutor

Adapter for the provider-neutral structured-output engine. The request factory is host-owned so Core does not prescribe a rubric, prompt, or result vocabulary.

Enums

AuxiliaryModeV1

AuxiliaryRunError

AuxiliaryRunStateV1

EvaluationBoundaryV1

EvaluationDispatchClaimOutcome

EvaluationDispatchLedgerError

EvaluationDispatchOutcome

EvaluationProtocolError

Errors produced while decoding or validating an evaluation wire message.

EvaluationStoreError

EvaluationWireKindV1

Closed top-level payload kinds in evaluation wire version 1.

EvidenceContentModeV1

EvidenceError

ExecutionFactKindV1

IdentityError

JournalError

SupervisorError

Constants

AUXILIARY_MAX_OUTPUT_BYTES

AUXILIARY_MAX_STEPS

AUXILIARY_OUTPUT_SCHEMA_V1

AUXILIARY_RUN_SCHEMA_V1

AUXILIARY_SNAPSHOT_SCHEMA_V1

EVALUATION_DISPATCH_LEASE_GRACE_MS

EVALUATION_DISPATCH_LEDGER_DEFAULT_MAX_RECORDS

EVALUATION_DISPATCH_LEDGER_MAX_BYTES

EVALUATION_DISPATCH_LEDGER_SCHEMA_V1

EVALUATION_DISPATCH_MIN_LEASE_MS

EVALUATION_MAX_COOLDOWN_MS

EVALUATION_MAX_ID_BYTES

EVALUATION_MAX_PENDING

EVALUATION_PLAN_SCHEMA_V1

EVALUATION_PROTOCOL_MAX_MESSAGE_BYTES

Maximum encoded envelope size accepted at a process boundary.

EVALUATION_PROTOCOL_SCHEMA_V1

Stable schema identifier for the cross-process evaluation envelope.

EVALUATION_PROTOCOL_VERSION_V1

Version carried by the evaluation wire envelope.

EVALUATION_RECORD_SCHEMA_V1

EVALUATION_RESULT_SCHEMA_V1

EVALUATION_RESULT_STORE_DEFAULT_MAX_RECORDS

EVALUATION_RESULT_STORE_MAX_BYTES

EVALUATION_RESULT_STORE_SCHEMA_V1

EVALUATION_WIRE_KIND_DESCRIPTORS_V1

Ordered catalog known by evaluation wire version 1.

EVIDENCE_MAX_ARTIFACTS

EVIDENCE_MAX_ARTIFACT_BYTES

EVIDENCE_MAX_EVENTS

EVIDENCE_MAX_EVENT_BYTES

EVIDENCE_MAX_PROMPT_BYTES

EVIDENCE_MAX_RESULT_BYTES

EVIDENCE_SNAPSHOT_SCHEMA_V1

EXECUTION_FACT_SCHEMA_V1

EXECUTION_FRAME_SCHEMA_V1

EXECUTION_TARGET_SCHEMA_V1

Traits

AuxiliaryExecutor

AuxiliaryRunService

EvaluationDispatchLedger

EvaluationPolicy

EvaluationResultSink

EvidenceReader

ExecutionFactJournal

Functions

digest_bytes

digest_json

validate_digest