gspc-papers / README.md
Nicholastempleman's picture
Papers index — 11 artifacts
b96ee7a verified
|
Raw
History Blame Contribute Delete
1.91 kB
metadata
license: cc-by-4.0
pretty_name: GSPC / SOVOS  Papers, Methods & Roadmap
tags:
  - ai-governance
  - benchmark
  - psychometrics
  - methods
  - eu-ai-act
  - provenance
language:
  - en

GSPC / SOVOS — Papers, Methods & Roadmap

By CSOAI, an independent AI-measurement body. Every number tagged MEASURED / SOURCED / SPEC / CAVEAT. csoai.org

Papers

File What
GSPC_METHODS_PAPER.md + arxiv/gspc_methods.tex What 12 AI-gov benchmarks don't report — psychometrics + deterministic grading. The arXiv candidate.
SOV_SIGNAL_METHOD_PAPER.md Statute-judged operator-vs-frontier disagreement as a training signal.
SOVOS_FLYWHEEL_PAPER.md One signed instrument across every dated regulatory obligation.
GSPC_CARECOST_PAPER.md Grounded operators beat a size-matched base (control-armed protect×help).

Measurement protocols (unowned commercial axes)

File Obligation
AGISAFE_ART5_SAFEGUARD_METHODOLOGY.md 2 Dec 2026 — Art 5 NCII/CSAM safeguard effectiveness (safety-bounded: corpus handled only by authorised holders).
POAI_DETECTOR_INTEROP_METHODOLOGY.md 2 Feb 2027 — watermark-detector interoperability matrix (≈190 signatories, no standard yet).

Concept, code & data

File What
XRAIV_CONCEPT.md Cross-reality AI validation — validate an agent by what it DOES in a world vs the law.
xraiv/gspc_xraiv.py XRAIV v0 — a runnable inspect_ai agent that consults the statute then commits a verdict, graded by the law.
regulatory_clock.json 19 geocoded, dated, sourced AI-governance obligations to 2030.
SOVOS_LEVELED_ROADMAP.md The L2–L4 roadmap, owner-tagged.

The 6 benchmark datasets: coai-bench · agisafe-bench · poai-bench · asisec-bench · mcp-scoreboard · omai-bench. Live arenas: kaggle.com/benchmarks/tasks/nicktempleman. Arena: SOV Space (Cesium globe).