Vaults.add_keys and kv add skip null content keys instead of failing the batch; is_null_key moves to core.vault. A manifest snapshot restore has one 32 MiB inflate budget. Tests reset the speed limiter and config between runs.
The server CDM stub names only the manifest's KIDs; sibling keys stay content keys. API and HTTP vaults retry a 429, and a vault that is down or still rate limited costs one timeout per run.
Map each KID to its fragments from the ciphertext moov/moof (tenc, seig, ISM duration clock) and decode up to three windows per KID that never cross into another KID, so a clear lead cannot hide a wrong key and only the failing KID is flagged. Decode every frame (wrong-key HEVC keyframes decode clean), re-judge mid-GOP starts on keyframes, time out every FFmpeg run, and refuse to store a flagged pair again in SQLite.
The licence response lists KIDs a server vault supplied, the client proves them with the decode check, and a bad pair goes to the new session keys/bad route so the server flags its own row. Client vault keys go first over --remote, and the dashboard health warns when no SQLite vault can store flags.
Vaults.add_key and add_keys now push to every vault at the same time, so one unreachable vault costs its own timeout instead of the sum. MySQL defaults connect_timeout to vault_timeout.
imSp4rky