nemo_gym.rollouts.harness_capabilities.behavior

Checks against controlled observations; independent of artifact validation and process exit.

Module Contents

Functions

Name	Description
`_mapping`	-
`_objects`	-
`_terminal_error_reported`	Compare the scripted provider error with the response received by the EnvironmentServer.
`inspect_behavior`	Check saved evidence and observations against explicit test expectations.
`model_checks`	Compare canonical fields to the independently scripted endpoint exchanges.
`tool_checks`	-

API

nemo_gym.rollouts.harness_capabilities.behavior._mapping(
    value: object
) -> dict

nemo_gym.rollouts.harness_capabilities.behavior._objects(
    value: object
) -> list[dict]

nemo_gym.rollouts.harness_capabilities.behavior._terminal_error_reported(
    attempts: list[dict],
    responses: list[dict],
    expected_status: int
) -> bool

Compare the scripted provider error with the response received by the EnvironmentServer.

nemo_gym.rollouts.harness_capabilities.behavior.inspect_behavior(
    witness: dict | None,
    record: dict | None,
    http_errors: tuple[int, ...],
    terminal_error: bool,
    tool_steps: int,
    expected_reward: float,
    fingerprint: collections.abc.Callable,
    judge_verifier: bool = False
) -> list[dict]

Check saved evidence and observations against explicit test expectations.

nemo_gym.rollouts.harness_capabilities.behavior.model_checks(
    record: dict | None,
    attempts: list[dict],
    fingerprint: collections.abc.Callable,
    available: bool = True
) -> list[dict]

Compare canonical fields to the independently scripted endpoint exchanges.

nemo_gym.rollouts.harness_capabilities.behavior.tool_checks(
    record: dict | None,
    witnessed: list[dict],
    finish_id: str | None = None,
    applies: bool = True,
    available: bool = True
) -> list[dict]