Score your agent's context on seven axes as a preflight gate, before you ever look at behavioral metrics
This July 15 paper treats context quality as an independent leading indicator of agent reliability and scores it with multi-juror consensus across role clarity, guardrail coverage, instruction consistency, tool schema quality, grounding sufficiency, injection hardening, and token efficiency. The reported mappings are specific enough to act on: grounding sufficiency predicts hallucination resistance, guardrail coverage predicts manipulation resistance, instruction consistency predicts instruction-following, and tool-schema quality predicts tool-use effectiveness. Running this as a deploy-time preflight — separate from behavioral evals — gives you a non-circular signal about which of the four failure modes you're about to ship; the harness (ProofAgent-Harness) is open source.
↳ Follow the thread