Research APIs

• Benchmark documents pllm.benchmark(...) and pllm.deployment_benchmark(...).
• Assure documents pllm.assure() and the limits of its result.
• Benchmark records explain the immutable evidence returned by the SDK.
• Assurance records explain what an assurance result can and cannot support.
• pllm benchmark run documents the separate real-role loopback diagnostic.

For provenance and implementation boundaries, use method implementations. For measurement scope and claim limits, use research evidence.

Python SDK example

import pllm

report = pllm.assure()
findings = {finding["id"]: finding for finding in report.to_dict()["findings"]}
assert report.schema_version == "pllm.assurance_report.v1"
assert findings["mask_reuse"]["outcome"] == "refuted_in_scope"

API: pllm.assure

This runs only the package's scoped deterministic assurance fixtures.