Skip to content
Design Bench · Holding AGI accountable

Epistemic-defense evaluation harness

A heavily formal evaluation harness for an epistemic-defense intelligence derived from Thirsty's Standard V3 — a SHA-256-hashed manifest, a canonical test catalog, and a reviewer-trust schema, with zero executed evaluations run against it yet.

Design boardinternal designation — Q
Design board

Named and scoped. Nothing built yet, or too early to show.

In more detail

The specification is extensive: an independent evaluator script, a formal reviewer-trust schema, a canonical test catalog. What it doesn't have yet is any output to evaluate — its own manifest reports runtime_enforcement as not_implemented and canonical_execution as blocked_no_subject_runtime_outputs. That's design-board in the most literal sense the taxonomy on this page allows: fully scoped, nothing to point to yet.

Concrete specifics

  • ·SHA-256-hashed manifest and canonical test catalog
  • ·Own manifest self-reports: runtime_enforcement: not_implemented
  • ·Own manifest self-reports: canonical_execution: blocked_no_subject_runtime_outputs
Go deeper