Agent conformance report
Checked 2026-08-20. Generated by scripts/build_conformance_report.py — do not hand-edit.
Third-party scoring of the published agent cards by Agenstry, against its open nine-criterion rubric (methodology, CC-BY-4.0).
| Agent | Grade | Score | Uptime | Protocol |
|---|---|---|---|---|
| agenda-intelligence | A | 91/100 | 97% | v1.0 |
| agent-output-verification | A | 91/100 | 100% | v1.0 |
| middle-corridor-deal-risk-gate | A | 91/100 | 100% | v1.0 |
| vassiliylakhonin.github.io | A | 91/100 | 99% | v1.0 |
| agentic-interaction-trust | B | 87/100 | 100% | v1.0 |
| cis-secondary-sanctions | B | 87/100 | 100% | v1.0 |
| corridor-sanctions-assistant | B | 87/100 | 100% | v1.0 |
| gulf-maritime-exposure | B | 87/100 | 99% | v1.0 |
| kazakhstan-market-entry-readiness | B | 87/100 | 98% | v1.0 |
What this number is
Conformance measures how the agents are published and whether they answer: card validity, declared protocol version, JWS signature, endpoint liveness, uptime across the scorer’s probes, skill declaration, registry-matched identity, freshness, and declared security.
It is not a usage, adoption, or quality measurement. Recorded invocations on these listings are zero. The grade says the endpoints are correctly published and answer reliably; it says nothing about anyone using them, and it must not be restated as traction on any surface of this site.
Where the points go, from the per-criterion breakdown read on 2026-08-20 for seven of the nine listings: card validity, protocol version, JWS signature, endpoint liveness and uptime score full on every one. Each loses five points on identity, for want of an authoritative-registry match (GLEIF, Companies House and equivalents) — a legal-entity question, not a technical one — and three on security declaration, the score for a deliberately public endpoint. The B-grade listings lose a further four on skill declaration, which rises with the number of declared skills rather than with anything the endpoint does.