cayu.evals.execution_profiles

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.

`EvalExecutionCandidateIdentityV1`

from cayu.evals import EvalExecutionCandidateIdentityV1

Safe current candidate identity selected by runtime preparation.

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
`agent_name`	`StrictStr`	`Field(min_length=1, max_length=EVAL_EXECUTION_PROFILE_MAX_TEXT_CHARS)`
`provider_name`	`StrictStr`	`Field(min_length=1, max_length=EVAL_EXECUTION_PROFILE_MAX_TEXT_CHARS)`
`model`	`StrictStr`	`Field(min_length=1, max_length=EVAL_EXECUTION_PROFILE_MAX_TEXT_CHARS)`
`environment_name`	`StrictStr \| None`	`Field(default=None, min_length=1, max_length=EVAL_EXECUTION_PROFILE_MAX_TEXT_CHARS)`
`runtime_execution_profile_schema_version`	`StrictInt`	`Field(ge=1, le=64)`
`runtime_execution_profile_fingerprint`	`StrictStr`	`Field(min_length=64, max_length=64)`

validate_text (method)

EvalExecutionCandidateIdentityV1.validate_text(value: str | None, info) -> str | None

validate_fingerprint (method)

EvalExecutionCandidateIdentityV1.validate_fingerprint(value: str) -> str

Release source · Runtime guarantees · Capability guides

`EvalExecutionProfileBindingV1`

from cayu.evals import EvalExecutionProfileBindingV1

Durable server-prepared runtime identity for one admitted eval run.

Bases: BaseModel.

Fields and attributes

Name	Type	Default
`schema_version`	`Literal[1]`	`1`
`profile_revision`	`StrictStr`	`Field(min_length=71, max_length=71)`
`runtime_execution_profile`	`ExecutionProfileIdentity`	`required`

validate_schema_version_type (method)

EvalExecutionProfileBindingV1.validate_schema_version_type(value: object) -> object

validate_profile_revision (method)

EvalExecutionProfileBindingV1.validate_profile_revision(value: str) -> str

copy_runtime_profile (method)

EvalExecutionProfileBindingV1.copy_runtime_profile(value: object) -> object

Release source · Runtime guarantees · Capability guides

`EvalExecutionProfilePolicyV1`

from cayu.evals import EvalExecutionProfilePolicyV1

Trusted application declaration layered around runtime execution identity.

The policy contains no callback or browser-provided authority. A generated or otherwise undeclared profile uses the conservative one-by-one defaults and claims only fresh Cayu session state, not reset external application state.

Bases: BaseModel.

Fields and attributes

Name	Type	Default
`schema_version`	`Literal[1]`	`1`
`fixture_strategy`	`Literal['none', 'application_managed']`	`'none'`
`reset_strategy`	`Literal['fresh_session_only', 'application_managed']`	`'fresh_session_only'`
`effect_posture`	`Literal['ordinary_application_authority', 'isolated_application_authority']`	`'ordinary_application_authority'`
`isolation_revision`	`StrictStr \| None`	`Field(default=None, max_length=71)`
`max_trials`	`StrictInt`	`Field(default=1, ge=1, le=EVAL_CORPUS_MAX_TRIALS)`
`max_concurrency`	`StrictInt`	`Field(default=1, ge=1, le=EVAL_MAX_CONCURRENCY)`

validate_schema_version_type (method)

EvalExecutionProfilePolicyV1.validate_schema_version_type(value: object) -> object

validate_isolation_revision (method)

EvalExecutionProfilePolicyV1.validate_isolation_revision(value: str | None) -> str | None

validate_scale_contract (method)

EvalExecutionProfilePolicyV1.validate_scale_contract() -> EvalExecutionProfilePolicyV1

safe_default (method)

EvalExecutionProfilePolicyV1.safe_default() -> EvalExecutionProfilePolicyV1

Release source · Runtime guarantees · Capability guides

`EvalExecutionProfileV1`

from cayu.evals import EvalExecutionProfileV1

Bounded public snapshot of one currently executable server profile.

Bases: BaseModel.

Fields and attributes

Name	Type	Default
`schema_version`	`Literal[1]`	`1`
`revision`	`StrictStr`	`Field(min_length=71, max_length=71)`
`profile_id`	`StrictStr`	`Field(min_length=1, max_length=128)`
`label`	`StrictStr`	`Field(min_length=1, max_length=EVAL_EXECUTION_PROFILE_MAX_TEXT_CHARS * 2)`
`source`	`Literal['generated', 'explicit']`	`required`
`target_key`	`StrictStr`	`Field(min_length=1, max_length=128)`
`application_release_id`	`StrictStr`	`Field(min_length=1, max_length=EVAL_EXECUTION_PROFILE_MAX_TEXT_CHARS)`
`app_manifest_fingerprint`	`StrictStr`	`Field(min_length=64, max_length=64)`
`candidate`	`EvalExecutionCandidateIdentityV1`	`required`
`target_material`	`EvalExecutionTargetMaterialIdentityV1`	`required`
`fixture_strategy`	`Literal['none', 'application_managed']`	`required`
`reset_strategy`	`Literal['fresh_session_only', 'application_managed']`	`required`
`effect_posture`	`Literal['ordinary_application_authority', 'isolated_application_authority']`	`required`
`isolation_revision`	`StrictStr \| None`	`Field(default=None, max_length=71)`
`evidence_policy`	`EvaluationEvidencePolicySpec`	`required`
`ceilings`	`EvalExecutionResourceCeilingsV1`	`required`

validate_schema_version_type (method)

EvalExecutionProfileV1.validate_schema_version_type(value: object) -> object

validate_revisions (method)

EvalExecutionProfileV1.validate_revisions(value: str | None, info) -> str | None

validate_manifest_fingerprint (method)

EvalExecutionProfileV1.validate_manifest_fingerprint(value: str) -> str

copy_evidence_policy (method)

EvalExecutionProfileV1.copy_evidence_policy(value: object) -> object

validate_revision (method)

EvalExecutionProfileV1.validate_revision() -> EvalExecutionProfileV1

comparison_revision (property)

EvalExecutionProfileV1.comparison_revision() -> str

Return the execution-semantic identity used between application releases.

Admission must use revision, which binds every published byte. Result comparison deliberately excludes release and presentation metadata while retaining the complete candidate, runtime, target-material, isolation, evidence, and resource contract.

create (method)

EvalExecutionProfileV1.create(*, profile_id: str, label: str, source: Literal['generated', 'explicit'], target_key: str, application_release_id: str, app_manifest_fingerprint: str, candidate: EvalExecutionCandidateIdentityV1, target_material: EvalExecutionTargetMaterialIdentityV1, fixture_strategy: Literal['none', 'application_managed'], reset_strategy: Literal['fresh_session_only', 'application_managed'], effect_posture: Literal['ordinary_application_authority', 'isolated_application_authority'], isolation_revision: str | None, evidence_policy: EvaluationEvidencePolicySpec, ceilings: EvalExecutionResourceCeilingsV1) -> EvalExecutionProfileV1

Release source · Runtime guarantees · Capability guides

`EvalExecutionResourceCeilingsV1`

from cayu.evals import EvalExecutionResourceCeilingsV1

Complete server-owned limits that HTTP may only narrow.

Bases: BaseModel.

Fields and attributes

Name	Type	Default
`max_cases`	`StrictInt`	`Field(ge=1, le=EVAL_CORPUS_MAX_CASES)`
`max_trials`	`StrictInt`	`Field(ge=1, le=EVAL_CORPUS_MAX_TRIALS)`
`max_concurrency`	`StrictInt`	`Field(ge=1, le=EVAL_MAX_CONCURRENCY)`
`max_timeout_seconds`	`StrictInt`	`Field(ge=1, le=EVAL_CORPUS_MAX_TIMEOUT_SECONDS)`
`max_bootstrap_messages`	`StrictInt`	`Field(ge=0, le=CORPUS_EXECUTION_MAX_BOOTSTRAP_MESSAGES)`
`max_total_input_chars`	`StrictInt`	`Field(ge=1, le=CORPUS_EXECUTION_MAX_TOTAL_INPUT_CHARS)`
`max_compiled_input_chars`	`StrictInt`	`Field(ge=1, le=CORPUS_EXECUTION_MAX_COMPILED_INPUT_CHARS)`
`max_steps`	`StrictInt`	`Field(ge=1, le=MAX_STEPS)`
`run_limits`	`RunLimits`	`required`

copy_limits (method)

EvalExecutionResourceCeilingsV1.copy_limits(value: object) -> object

Release source · Runtime guarantees · Capability guides

`EvalExecutionTargetMaterialIdentityV1`

from cayu.evals import EvalExecutionTargetMaterialIdentityV1

Opaque commitment to candidate inputs owned by one trusted target.

Runtime execution authority is represented separately by ExecutionProfileIdentity. This identity covers the remaining target material consumed while compiling each fresh request: the request base, bootstrap messages, and complete corpus-execution limit object.

Public-safe material has a restart-portable structural digest. Material crossing the application’s workload-secret boundary uses a process-keyed HMAC instead, so it remains exact within the serving process without exposing an offline digest of private values. A restart then changes the process scope and safely rejects previously admitted work.

Bases: BaseModel.

Fields and attributes

Name	Type	Default
`schema_version`	`Literal[1]`	`1`
`kind`	`Literal['structural_sha256', 'process_local_hmac_sha256']`	`required`
`fingerprint`	`StrictStr`	`Field(min_length=64, max_length=64)`
`process_scope`	`StrictStr \| None`	`Field(default=None, min_length=64, max_length=64)`

validate_schema_version_type (method)

EvalExecutionTargetMaterialIdentityV1.validate_schema_version_type(value: object) -> object

validate_fingerprints (method)

EvalExecutionTargetMaterialIdentityV1.validate_fingerprints(value: str | None, info) -> str | None

validate_process_scope (method)

EvalExecutionTargetMaterialIdentityV1.validate_process_scope() -> EvalExecutionTargetMaterialIdentityV1

Release source · Runtime guarantees · Capability guides