Expand description
Prepared conditioning request snapshot and exact-token admission gating.
Enforces:
- Dual limits: both
max_bytesandinput_tokensverified against the exact final rendered stream. - Separate output reservation:
actual_input_tokens + reserved_output_tokens <= supported_context_tokens. - Full metadata snapshot: binds profile ID, version, evidence revision, access scope, model instance, adapter hash, and rendered token digest into the prepared request.
- Generation of a canonical
InferenceCacheKeycompatible with the Tier-1 prefix cache.
Structsยง
- Prepared
Conditioning Snapshot - Snapshot of a fully validated and admitted conditioning request ready for execution.