Scoring model

How CanAgentUse calculates report scores

CanAgentUse scoring is intentionally layered. Versioned check packages decide their internal step scoring, while the published assessment suite decides where each check belongs and how category, subcategory, and check weights roll up into the report.

1. Scoring Principles

Separated ownership

Checks own their internal step weights. The assessment suite owns category, subcategory, and check weights. That keeps versioned check behavior isolated from suite composition.

Active denominators

Informational and not-applicable results are displayed when useful, but they are removed from the scoring denominator so irrelevant checks do not punish a page.

Rounded report scores

Each weighted average is rounded to a whole number and clamped between 0 and 100 before it is shown in the report.

2. Formula Chain

A report score rolls up from the smallest scored unit to the final score. The chain below mirrors the current runtime behavior.

1

Step value

pass = 100, warning = 50, fail = 0

Informational and not-applicable steps do not enter the active denominator.

2

Check score

sum(step value x step weight) / sum(active step weights)

The versioned check package owns the step list and step weights.

3

Subcategory score

sum(check score x check weight) / sum(scored check weights)

The suite owns which checks belong to each subcategory and how much each check counts there.

4

Category score

sum(subcategory score x subcategory weight) / sum(scored subcategory weights)

Subcategories without scored checks are omitted from the category denominator.

5

Agent readiness

sum(category score x category weight) / sum(scored category weights)

The published suite manifest owns category weights.

6

Overall score

agent readiness x 0.80 + performance/accessibility x 0.20

The combined technical score is desktop Performance x 0.65 plus desktop Accessibility x 0.35.

3. Suite Category And Subcategory Weights

The active suite is can-agent-use-agent-readiness@2026.06.10. Category weights are relative contributions to agent readiness, not standalone percentages of the final score. Agent readiness itself contributes 80% of the overall report score.

CategoryCategory WeightSubcategoriesChecks
Discoverability0.2
Crawl Discovery 1
4
Content Readiness0.2
Page Structure 1Structured Entities 1
7
GEO, AIO & AEO0.15
AI Answer Readiness 1
3
Bot Access Control0.25
Bot Policy 1
6
Security & Trust0.15
Browser Trust 1
5
API0.04
API Discovery 1
2
Auth0.04
Auth Discovery 1
3
MCP0.04
MCP Discovery 1
2
Skill Discovery0.04
Agent Capability Discovery 1
4
Agent Commerce0.04
Commerce Protocols 1
4

Check weights live one level deeper in the same suite manifest. For example, this suite pins 40 runnable check packages, each with its own suite-level check weight and isolated package-level step weights.

4. Technical Score And Final Overall

Performance and Accessibility

desktop performance x 0.65 + desktop accessibility x 0.35

The final overall score gives the combined Performance and Accessibility score 20% weight. Inside this section, Performance contributes 65% and Accessibility contributes 35%. Optional-mode fallback: when this technical section is omitted, agent readiness carries 100% of the overall score.

Final Overall

agent readiness x 0.80 + performance/accessibility x 0.20

The result is rounded to a whole number and shown as the main report score.

Optional-mode fallback formula: agent readiness x 1 + performance/accessibility x 0.

SEO is still shown as its own report row. In suite 2026.06.10, displayed as its own report score row. This version does not blend the SEO score into the final overall score.

5. Readiness Levels

The final rounded score maps to a readiness label. The label is descriptive; it does not change the underlying numeric score.

Score RangeReadiness Label
90-100Level 5, Agent-Ready
75-89Level 4, Mostly Ready
50-74Level 3, Bot-Aware
25-49Level 2, Agent-Limited
0-24Level 1, Not Agent-Ready
Open the exact suite registry