Offline fixture experiments, provider-neutral translation, and reproducible grading.
| Framework Home | Cross-Pillar Governance | AISM | NEXUS | Dashboard |
Start with the command reference and runnable workflow. The design contract explains the wire format. The Challenge Lab owns study protocols and evidence standards; this directory is the installable Python implementation, not a separate framework.
| Module | Responsibility |
|---|---|
protocol.py, runner.py |
Pin the six-case protocol and run three inert fixture treatments |
bundle.py |
Orchestrate the one-command starter and verify portable bundles without executing submitted code |
grading.py |
Recompute state-based outcomes without accepting producer scores |
adapters.py |
Translate generic or illustrative TENIR exports, preserving originals |
compare.py |
Refuse incompatible conditions and expose agreement, gaps, and conflicts |
model.py, integrity.py |
Validate evidence semantics, detect changes, and verify optional signatures |
io.py |
Bound local artifact I/O and reject unsafe paths and overwrites |
report.py |
Render human-readable Markdown and self-contained HTML Decision Cards |
No live agents or provider code are executed. Synthetic translation agreement is not independent replication. Framework conformance and AISM maturity remain unassessed; real deployments require the Lab’s separate live-evidence process.
AI SAFE² v3.1 · Cyber Strategy Institute