Assurance

An assurance record identifies the exact plan, component versions, threat model, protected values, visible information, corruption and collusion assumptions, numeric policy, environment, procedure, result, and limitations.

Reference tests, formal analysis, cryptographic review, transport hardening, deployment inspection, and leakage tests answer different questions. Missing assurance is not a failed test, but it is not a pass. Evidence for participants that follow the protocol does not establish security against malicious behavior.

A publication claim must cite the relevant research evidence and keep its limitations. See reading research and evidence for the conceptual boundary and publications and claims for the review requirements.

Python SDK example

import pllm

document = pllm.assure().to_dict()
findings = {finding["id"]: finding for finding in document["findings"]}
assert document["ideal_uniform_control"]["outcome"] == "proved_in_model"
assert findings["affine_label_reuse"]["outcome"] == "refuted_in_scope"

API: pllm.assure

Report scope is bundled assurance fixtures, not universal privacy proof.