Fidget Labs

What does your architecture look like to an AI agent?

Submit a domain. We run a battery of deterministic checks and publish a scored report to a shareable URL. Agent readiness, performance, security posture, architectural governance, and three competitors scored against you.

Request your report

Free · No sales call required · Delivered as a URL and a PDF

Three scored axes, one composite

ARI
Agent Readiness IndexWhether AI agents can reach, read and act on the site. Crawler access, bot cloaking at the edge, server-rendered content, entity schema, and machine-readable interfaces.
CWV
Core Web VitalsReal-user field data where it exists, lab measurement where it does not. Always labelled so you know which you are reading.
SEC
Security PostureTransport, headers, cookie flags, email spoofing exposure, and a fixed set of commonly-exposed paths.

What else is in the report

Governance
Serverless, API and MCP surface, composable architecture. Assessed from evidence, not from a questionnaire.
Market
Trend and news signal for the category, with every claim traced to a retrieved source.
Competitors
Three named competitors scored on the same rubric, with the delta against you and a recommendation.

Why this is worth measuring

12.4%

of the Fortune 1000 have valid entity schema for AI

5-10%

of what AI search draws from is a brand's own website

31%

of the US population will use generative AI search this year

2026

ADA enforcement and WCAG pressure both increased this year

Sources: Fortune 1000 entity schema analysis, 2026 · AI search source attribution research, 2026 · US generative AI search adoption estimate, 2026 · ADA digital accessibility enforcement trend, 2026

Request a report

Four fields. The report arrives by email as a shareable link and a PDF.

We assess the domain you submit. We do not store the contents of anything we find. Your details go to Fidget Labs and nowhere else.

How it is scored

Every score is a pure function over archived evidence. No language model can move a number. Models write the prose, and they only ever see figures that were already computed. Re-scoring the same archived run produces the same result, and there is a test that proves it.

A check we could not run is reported as unmeasured, never scored as zero. Scoring a site badly because our own network failed would make the report worthless, so the protocol treats the two states separately and the report tells you which happened.

Read the full scoring protocol