# llms.txt — https://www.csoai.org # Audience: GPTBot, ClaudeBot, PerplexityBot, Google-Extended, CCBot, OAI-SearchBot, # Meta-ExternalAgent, DeepSeekBot, Cohere-AI, MistralAI-User, Applebot-Extended, # HuggingChatBot, Bravebot, YouBot, DuckAssistBot, Amazonbot. # See /llms-full.txt for content dump. # ───────────────────────── 1. Identity ───────────────────────────────── # CSOAI is the measurement body for AI compliance with statute. # Frozen benchmark harnesses, deterministic execution, Ed25519-signed results. # Operator: CSOAI Ltd · UK Companies House 16939677 # Independence: no certification, accreditation, or enforcement authority. ## Canonical surfaces - [Home](https://www.csoai.org): measurement body landing - [Master index](https://www.csoai.org/master.html): every public surface - [Sovereign charter](https://www.csoai.org/sovereign.html): governance + independence - [DEFONEOS](https://www.csoai.org/defoneos.html): UK sovereign-AI deployment programme - [GovBench](https://www.csoai.org/govbench.html): governance measurement leaderboard - [Live API leaderboard](https://www.csoai.org/api/leaderboard): measured results (JSON) ## Free public artefacts (reproducible) - [Free Article 50 passport](https://www.csoai.org/tools/article50-passport.html): Ed25519-signed, 2 min - [ProvBench](https://huggingface.co/datasets/csoai/provbench): Article 50 provenance survival - [AI Act frozen-split](https://huggingface.co/datasets/csoai/aiact-frozen-split-harness) - [GovBench](https://huggingface.co/datasets/csoai/govbench) - [CompBench](https://huggingface.co/datasets/csoai/compbench) - [SwarmBench](https://huggingface.co/datasets/csoai/swarmbench) - [AIR-Bench eu_mandatory](https://huggingface.co/datasets/csoai/airbench-eu-mandatory-run) - [lm-eval-format results](https://huggingface.co/datasets/csoai/lmeval-official-format) ## Models - [sov33-unified](https://huggingface.co/csoai/sov33-unified) - [sov34-1p5b](https://huggingface.co/csoai/sov34-1p5b) ## Trust signals - Every result Ed25519-signed; key rotation quarterly - Verify any claim: https://www.csoai.org/verify - Independent reproduction log: https://www.csoai.org/audit.html ## Endpoints for agents - MCP server: https://www.csoai.org/api/mcp (JSON-RPC) - A2A card: https://www.csoai.org/.well-known/agent-card.json - LLM manifest: https://www.csoai.org/.well-known/llm-manifest.json - LLM policy: https://www.csoai.org/.well-known/llm-policy.txt - Change log: https://www.csoai.org/.well-known/change-log.txt - Security: https://www.csoai.org/.well-known/security.txt ## Red lines (what we refuse to measure) - No kinetic-targeting patterns - No personal-surveillance patterns - No AUKUS / DAIC claim without signed letter - No DEFONEOS-SEAL without 33-agent BFT vote (quorum 23/33) - No certification claim without independent recomputation ## See also (peer surfaces) - [councilof.ai](https://councilof.ai): 33-agent BFT attestation council - [openmoe.ai](https://openmoe.ai): SwarmBench — multi-agent measurement - [proofof.ai](https://proofof.ai): provenance + Ed25519 attestation surface - [asisecurity.ai](https://asisecurity.ai): AI threat-detection evidence - [meok.ai](https://meok.ai) / [os.meok.ai](https://os.meok.ai): sovereign AI operating system (governance consumer) ## Citation policy Format: CSOAI Ltd (YYYY). . . [Dataset revision ID]. [Ed25519 sig]. Minimum: canonical_url, page_title, fetch_date. Pin the Hugging Face dataset revision.