Fidget Labs

The FAR protocol

Version 1.0.0

Fidget Architecture Readiness. This is the whole scoring method, published so you can check our work rather than take the number on trust.

The determinism guarantee

Given the same archived evidence and the same protocol version, re-scoring a run produces identical scores, bands and findings.

Three rules make that hold. No language model sits in the scoring path, so no model output can move a number. Probes perform the observation and scoring is a pure function over what they observed, so scoring never touches the network. Everything a probe saw is archived, so any score can be re-derived later. There is a test that re-scores saved evidence and asserts the output is byte-identical, and it runs on every change.

A check we could not complete is reported as unmeasured, and its axis is renormalised across the checks that did run. It is never scored as zero. Marking a site down because our own network failed would make the report worthless, so the protocol treats those two states separately and every report shows which happened.

Naming

ARI, Agent Readiness Index. Deterministic. Measured from published signals, not a live agent run. It measures published signals: crawler access, edge behaviour toward AI agents, server-rendered content, entity schema, machine readable interfaces, and document semantics.

AXIS, Agent Experience Score. Live agent run, judged transcript. Available on the full assessment. It is a different measurement, so it carries a different name. This report will never present a deterministic score as though a live agent produced it.

Axes and weights

AxisWeight in compositeSignals
Agent Readiness0.4013
Core Web Vitals0.306
Security Posture0.3011

Agent readiness carries the most weight because it is the dimension nothing else in this market scores, and because it is the one with a live distribution consequence.

Bands

RangeBandMeaning
90 to 100ExemplaryReference class. Cite this site to others.
75 to 89StrongSolid posture with targeted gaps.
60 to 74AdequateWorks, but leaves value on the table.
40 to 59At RiskMaterial gaps with commercial consequences.
0 to 39CriticalActively losing visibility, trust, or both.

Every signal

All 30 scored assertions, with the weight each carries inside its axis.

Agent Readiness

SignalIdentifierWeight
Answer-surface AI crawlers are allowedari.robots.tier1_access10
Training and index crawlers are allowedari.robots.tier2_access3
Site serves AI agents the same content as browsersari.cloaking10
Content is present without JavaScriptari.ssr_content9
llms.txt is published and spec-validari.llms_txt5
Organization entity schema is publishedari.organization_schema8
Structured data covers key content typesari.structured_data_breadth4
Main content is machine-extractableari.extractability6
Document structure is semanticari.semantics5
Sitemap is valid and discoverableari.sitemap4
Machine-actionable interfaces are discoverableari.machine_surface5
No agent-hostile patternsari.agent_hostile4
No critical accessibility violationsari.a11y_semantics6

Core Web Vitals

SignalIdentifierWeight
Largest Contentful Paintcwv.lcp10
Interaction to Next Paintcwv.inp8
Cumulative Layout Shiftcwv.cls7
Time to First Bytecwv.ttfb4
Lighthouse performance scorecwv.lighthouse_performance6
Real-user field data is availablecwv.field_data_present2

Security Posture

SignalIdentifierWeight
HTTPS is enforcedsec.https10
HSTS is configuredsec.hsts7
Content Security Policy is enforcedsec.csp9
Baseline security headerssec.headers_baseline6
Cookies are correctly flaggedsec.cookies5
No publicly exposed sensitive pathssec.exposure10
TLS certificate is comfortably validsec.tls_expiry5
Domain is protected against email spoofingsec.email_spoofing6
No software version disclosuresec.version_leak3
No mixed or insecure contentsec.mixed_content4
security.txt is publishedsec.security_txt2

Competitor recommendations

Competitors are scored on this same rubric, at the same protocol version, so the comparison is like for like. The recommendation is a rule on the delta, not a judgement. A model writes the sentence explaining it. A model does not choose it.

DeltaRecommendation
10 points or more behindClose the gap
Within 9 points either wayTake the lead
10 points or more aheadYou already lead. Spend elsewhere.

What this protocol does not do