Observed Evidence Capture Status

Current coverage

promoted frameworks: 18
source records: 18 of 18
individual pages: 18 of 18
benchmark mappings: 18 of 18
non-authorizing fixtures: 18 of 18
priority capture queue: 7 of 7
executable capture harnesses: 7 of 7
captured external outputs: 0 of 18
independently replayed outputs: 0 of 18

Priority capture harnesses

Framework	Harness posture	Runtime posture	Evidence state
Open Policy Agent	dedicated deterministic harness	pinned OPA executable required	awaiting_capture
Cedar Policy	dedicated implementation-neutral harness	exact Cedar implementation required	awaiting_capture
Model Context Protocol	reusable command harness + manifest	exact client/server and transport required	awaiting_capture
Agent2Agent Protocol	reusable command harness + manifest	exact agent implementation required	awaiting_capture
Guardrails AI	reusable command harness + manifest	exact package, validator, and model context required	awaiting_capture
Llama Guard	reusable command harness + manifest	exact model digest and runtime required	awaiting_capture
NeMo Guardrails	reusable command harness + manifest	exact package, rails configuration, and model context required	awaiting_capture

Evidence progression

fixture_ready
-> awaiting_capture
-> captured_unverified
-> observed_partial
-> replay_ready
-> replay_confirmed
-> interoperability_candidate

No framework has advanced beyond awaiting_capture on this surface. Harness installation is structural readiness, not observed behavior.

Boundary

capture harness != external execution
captured output != compatibility
single run != replayability
matching replay != execution authority
policy decision != current delegation
classifier result != admissibility
protocol response != standing

The exact version, timestamp, execution environment, input, output, policy or configuration, authority context, freshness context, hashes, limitations, and replay instructions must be retained before any state advancement.