What are we mismeasuring? Help us harden the Agent-Tool Discoverability Standard

[shigeki7777]

The Agent-Tool Discoverability Standard grades public MCP servers A–D against 10 criteria, each bound to a spec/measurement (not opinion). It is v0.x and almost certainly wrong somewhere.

A measurement layer that mis-scores a healthy server is worse than no layer at all, so we want to be told where the instrument is off. Questions:

• Which of the 10 criteria produce false positives / false negatives in your experience?
• What's a real readiness signal we don't measure yet (and how would you measure it deterministically)?
• Where is a criterion measuring opinion dressed up as a spec?
• Any server that got a grade you can defend as flat wrong? Link it.

Concrete proposals can graduate into a PR against STANDARD.md (see the good-first-issues). Adversarial review is the most valuable contribution here — that's the whole moat.