What one agent-readiness scanner cannot tell you

Most agent-readiness checks run one scanner and stop. The report reads well, the grade looks final, and the site moves on. That grade says the site fits one vendor's model of what an agent needs. It says nothing about what that model left out.

A checklist is not a proof

A scanner is built around a fixed set of checks and a fixed weighting between them. It can only fail a site on something it looks for. When a category sits outside a scanner's model, a real gap in that category passes clean, because nothing in the checklist asked about it. A high grade from a single tool is easy to over-read as finished, while the site can still stop an agent cold on the one thing that tool never checked.

Where two scanners disagree

turva.dev is scored on two independent scanners with different category models, isitagentready.com and startuphub.ai. isitagentready.com carries no Quality category and marks Commerce as optional. startuphub.ai grades six categories, Discoverability, Content, Access Control, Capabilities, Commerce, and Quality, so both of the categories the first scanner treats as thin get a full reading in the second. Running a site through both at once means a gap sitting in one model's blind spot still shows up in the other's report, before a buyer or an agent finds it the hard way.

What this changes about an audit

Every audit here checks a site against both scanners, and a claim about the result carries the date it was verified and the categories the report named. A score nobody re-ran after a change is a guess wearing a number. Two readings pointed at the same site is the cheapest way I know to stop fooling yourself about what "done" means, and it is the same discipline that runs on turva.dev itself before any change ships.

For an agent-readiness audit that checks a site against more than one scanner, contact info@turva.dev.

Originally published at https://turva.dev/blog/two-scanner-audit-method