Enterprise validation
Public validation record connecting readiness claims to runtime evidence, proof-path behavior, and aggregate benchmark results.
What this page validates
Read this page when the buyer asks what evidence backs the product claims. This page connects readiness language to machine-derived operating evidence.
Current public validation snapshot
96executable traces built from24protected seed templates across4runtime familiesPCAA runtimeroute quality: exact accuracy1.000, macro-F11.000, severe recall1.000, block precision1.000- review burden: explicit review
0.292, simulate-first0.208, hard block0.250 - proof path: manifest stability
1.000, replay readiness1.000, receipt completeness0.516 - ablation sensitivity: removing externality lowers exact accuracy to
0.875; removing approval-enforceability handling lowers it to0.917; removing the integrity lane drops manifest stability to0.000 - CAVA semantic pattern detection is now treated as an internal CAVA capability: canonical actions can be mapped to hidden externality, public persistent egress, security-control weakening, credential exposure, delegated authority mismatch, and workflow sink risk before policy profiles route the final posture.
- CAVA locale-aware risk hints are now treated as secondary evidence: Japanese and Chinese operator text can raise conservative review signals for sensitive data, public export, external sharing, production mutation, or approval gaps when runtime structure is incomplete. Runtime-provided tool structure remains the primary control signal.
- Runtime Coverage is reported in three tiers: reference adapters for reproducible open examples, managed OSuite adapters for supported product lanes, and enterprise custom adapters for private customer systems. The tier is a disclosure field, not a marketing label.
How to read the evidence
The public evidence is meant to answer three questions:
- does the control path preserve route quality across heterogeneous runtimes
- does the review burden stay structured instead of collapsing into blanket blocking
- does proof stability fail differently from routing when integrity support is removed
- which runtime coverage tier produced the evidence, and whether incomplete evidence caused abstention rather than a false confident classification
The benchmark is executable, but the public reporting surface is aggregate by design. The raw scenario list, exact thresholds, exact heuristic weights, and connector-specific routing rules are intentionally withheld.
What the numbers do not mean
These numbers do not mean runtime governance is solved, nor do they imply that every deployment exposes identical pre-execution control. They show that the current certificate-centered control path behaves coherently on the protected benchmark and degrades in informative ways under ablation.
The public reference parser-pack coverage should not be read as OSuite's production coverage ceiling. It is a reproducible baseline. Managed adapters and enterprise custom adapters are evaluated as customer runtime coverage, with their own evidence posture and disclosure requirements.
Why this page matters
If validation is weak or vague, the rest of the readiness record becomes harder to trust. This page exists so that claims about approvals, replay, disclosure, and buyer packets can point back to one reviewable validation surface.
For governance reviews, the validation record should also show whether shared context survives across workflow boundaries and whether approval closure remains visible when an action moves from recommendation to release.