Summary

Add the canonical Python/PDF evidence for the RC23 GetEntropy policy:

• prove a fresh initialized, PIN-protected, locked device still requires confirmation;
• prove an uninitialized device receives exactly 8 x 8192 bytes = 64 KiB without confirmation;
• check exact response lengths, block uniqueness, byte diversity, dominant-symbol bounds, and a broad monobit bound;
• prove the next request restores ButtonRequest confirmation;
• keep the protection-level test meaningful by clearing the session before its confirmation assertion;
• update C27 to describe the exact evidence and explicitly avoid claiming statistical certification.

CI contract

The new C27 case is gated by KK_EXPECT_ENTROPY_BUDGET=1. Upstream Python CI uses the published emulator and skips the unreleased firmware policy. The repaired firmware PR sets the marker against its exact-head emulator, where C27 is mandatory and PDF catalog validation must pass.

Base and scope

Targets reconcile/upstream-sync@9c58e677135e24e59b5d867348073d47dae61785, immediately after the canonical Taproot staging merges #204 and #206. Test/report only; no protocol or package publication change.

Local validation:

• git diff --check
• python3 -m py_compile tests/test_msg_getentropy.py tests/test_protection_levels.py scripts/generate-test-report.py
• targeted pytest collection with the repository's pure-Python protobuf compatibility mode