zanii-underwrite

Insurance evidence for AI agents: build and verify the signed formats of Zanii Underwrite in Python. Pure functions, no I/O, no network. Alpha (0.x): formats may change between minor versions.

pip install zanii-underwrite

from zanii_underwrite import Refusal, verify_pack

try:
    verdict = verify_pack({"pack": pack, "documents": documents, "receipt": receipt, "decisions": decisions,
                           "pq": pq, "purpose": purpose, "recipient": recipient, "at": at,
                           "trusted_signers": trusted, "seen_nonces": [], "require_pq": False})
except Refusal as r:
    print("refused:", r)  # the first failing rule id

What it covers:

• Risk: exposure records, loss runs, completeness proofs, insurability and readiness packs, monitoring events, renewal diff, framework crosswalk, AI-use statement, pool and portfolio views.
• Response: case logs, incident exports, playbook reports, timelines, claim packs.
• A zanii-uw-py CLI with the same contract as the TypeScript zanii-uw.

Every format is canonical JSON (RFC 8785), integers only, and unknown values are None with a reason. The TypeScript library @zanii/underwrite on npm gives the same bytes for the same input.

Zanii never insures, quotes, places or advises on cover; a licensed partner carries every regulated step. The framework crosswalk is an evidence map awaiting a compliance reviewer, not a certification. Seeds and keys named TEST ONLY are published on purpose for test vectors; a production verifier must never trust them.

License: Apache-2.0. Contact: info@zanii.agency.