← benchmarksrAIlsApps →
Reliability Lab · precomputed evidence

Cross-app decisions, replayable as evidence

This lab publishes a deterministic rAIlsApps benchmark bundle: governed cross-app decisions, independently checked final states, immutable episode digests and learning eligibility. It is read-only; live deployments are provisioned separately and torn down after test runs.

2decision episodes
50%verified success
50%correct hold
100%learning eligible
100%decision accuracy
100%fault recovery
0double applies
100%trace complete
PASSreplay gate
HOLDlearning gate
PENDINGplatform gate
4/11app cores exercised
Current bundle: synthetic Lane A revenue-operations cases from the cross-app benchmark. One case verifies a correct HOLD when an invoice dispute is open; one verifies governed collection when no dispute exists. The artifact is intentionally small while the runtime and benchmark contracts are in review, and it reports limitations rather than hiding unsupported app cores.
bundle digest: loading...
decision provider: loading...
durability: loading...
explainability: loading...
learning: loading...
cross-platform: loading...
audit artifacts: loading...
live evidence request: loading...
live evidence runbook: loading...
assessment: loading...
baseline comparison: loading...
policy promotion: loading...
implementation status: loading...
errata register: loading...
blog brief: loading...
app expansion: loading...
publication integrity: loading...
publication decision: loading...

App Coverage

Lane A scope: this bundle exercises the revenue-operations path and keeps non-exercised cores visible as unsupported for this lane rather than counting them as passing.
coverage: loading...

Episodes

What This Proves

Independent outcome readback

The verdict is derived from an oracle reference, not the mission's own success claim.

Correct holds are first-class

A HOLD on open dispute is counted separately from completed execution, not averaged away.

Learning is gated

Episodes carry behavior policy, propensity, readback and observed outcome before they become eligible training evidence.

Generated from redevops-benchmarks/workloads/crossapp/apps_e2e_export.py against the runtime contracts in agentic-os. Public bundle contains synthetic benchmark evidence only.