MISSION-Bench Reproduction Registry

[github-actions[bot]]

MISSION-Bench exists to test intelligent systems under failure, not merely under happy-path benchmark conditions.

This thread is the public registry for independent reproductions, contradictory runs, ports, and extensions of MISSION-Bench.

When posting a result, include:

• exact Aftergraph source commit or release,
• workload/task set and failure-injection mode,
• runtime, model/provider/version, hardware, dependencies, and relevant configuration,
• sample size and exclusion criteria,
• VSR, FCR, recovery, unauthorized-action, cost/CPVO, latency, and control-plane overhead where applicable,
• raw results, scripts, logs, traces, and analysis artifacts,
• deviations from the published procedure and known limitations.

Negative results are first-class contributions. A reproduction that fails is more useful than a green screenshot with no provenance.

Canonical research source: https://github.com/Aftergraph/intelligence-systems-research

The target is not agreement with Aftergraph. The target is a result another team can inspect and repeat.