ai-safe2-framework

AI SAFE² Challenge Evidence Runtime

Offline fixture experiments, provider-neutral translation, and reproducible grading.

AI SAFE² Module

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