{
 "@context": "https://csoai.org/llm-context.json",
 "type": "LLMPageSummary",
 "url": "https://csoai.org/research.html",
 "title": "Research — papers, instruments, and open evidence | CSOAI",
 "description": "CSOAI research: the GSPC psychometric instrument for AI-governance evaluation, ProvBench, twelve measurement axes with honest states, and open datasets on Hugging Face and Kaggle. Every result signed and recomputable.",
 "headings": [
  "Research",
  "Papers",
  "The instrument: twelve axes, honest states",
  "Method commitments",
  "Open contributions",
  "Practice guides",
  "Corrections and retractions"
 ],
 "text": "Research. Everything here is open: the item banks, the graders, the raw runs, and the signed attestation chains. Papers: GSPC: A Psychometric Instrument for AI Governance Evaluation (working paper, 2026-08-05, doi.org/10.5281/zenodo.21806092). What Twelve AI-Governance Benchmarks Do Not Report — methods paper, arXiv submission in preparation (cs.CY). ProvBench: Measuring Content Credential Survival Across Real-World Transforms — preprint and live harness at /provbench. The instrument: twelve axes, honest states. Measured: governance (237 public items), safety (GSPC-AGI v2), provenance, conformance, openness, continuity. Draft: machinery, care. Spec: cross-reality, detector-interop, Art-5 safeguard. Planned: swarm. Datasets at huggingface.co/csoai and Kaggle; leaderboard harness GovBench (govbench.pages.dev). Method commitments: deterministic grading (98.9% agreement vs 92 hand-labelled responses, zero false serves); certified dead-item analysis (35 of 90 apparent-dead at fleet 8, 8 at certified fleet 22); refusal to quote below usable-n >= 30; signed evidence (Ed25519 / ML-DSA-65, in-toto/DSSE), verify at proofof.ai/verify. Open contributions: NVIDIA NeMo labs-OO-Agents PR #75 under review; inspect_evals registration follows the arXiv posting. Practice guides: ten sector white papers at proofof.ai/whitepapers. Corrections and retractions: five retractions published 2026-08-05, each naming the artefact that disproved the claim. CSOAI Ltd, UK company 16939677. Measurement, not certification.",
 "text_truncated": false,
 "register": {
  "role": "measurement_and_attestation_support",
  "csoai_certifies_systems": false,
  "csoai_is_a_notified_body": false
 }
}