Runnable and readable artifacts backing the project's flagship claims — every major claim in the README has one here.
| Directory | Claim it demonstrates |
|---|---|
findings/ |
"Every tool returns one normalized schema" — a sample finding per server, schema-validated in CI |
transcripts/ |
"A small model drives a full triage" and "a model can never write alone" — annotated sessions, including both fail-closed paths of the write gate |
personas/ |
"One evidence base, four altitudes" — the same findings rendered by all five persona renderers, reproducible with real code |
mcp/ |
Client wiring — MCP config snippets for Claude Desktop / Claude Code / generic stdio clients |
Fastest live demo (offline, no GPU, no tenant):
uv run python scripts/demo_mock_findings.py — see docs/demo.md.