We measure AI systems against the rules that govern them, sign the result (Ed25519), and publish what we cannot yet measure. Not a certifier. Not an enforcer. No accreditation chain — an independent measurement instrument.
- 13 GSPC axis benchmarks — governance, safety, provenance, continuity, conformance, openness, machinery, care, cross-reality, detector-interop, art5-safeguard, swarm, affect (EU AI Act Art 5 + Art 50). → banks on HuggingFace and Kaggle.
- Live MCP endpoint —
https://csoai-gspc-mcp.nicholastempleman.workers.dev/mcp(streamable HTTP, tools:measure+verify). Free to verify, metered to issue. - Durable measurement boards —
csoai/gspc-boards: deterministic multi-model scoring, Wilson intervals, tail-risk (CVaR + correlated-failure), honest UNMEASURED where the floor (n≥30) isn't met. - Layer-0 compliance MCPs — EU AI Act (410 verbatim articles), DORA, CRA Annex IV, NIS2, C2PA/Art 50 watermarking. See pinned repos.
Deterministic gold labels · no model judges another model · unparsed counts incorrect · nothing quoted below n≥30 · every published number recomputable from its rows · corrections published, never silently edited.
Independent verification — cards are Ed25519-signed + time-anchored (OpenTimestamps against the Bitcoin calendar); anyone can recompute from the published rows.
🌐 councilof.ai · 📊 boards · 🤖 MCP · ✉️ nicholas@csoai.org