# How the Lantad score is calculated

> The full method: what is fetched, how prose parity is computed, how access is probed per crawler, and every weight in the composite.

- Canonical page: https://lantad.co/methodology
- This file: https://lantad.co/methodology.md

## Key facts

- **Composite weights:** prose parity 50, access 25, structure 15, schema 10, out of 100
- **Grade bands:** A at 90, B at 75, C at 60, D at 40, F at 0
- **Entity Confidence:** A separate diagnostic scored out of 100 and reported as a tier of not declared, declared or anchored. It never feeds the composite.
- **Crawler tokens:** 15
- **Tokens with no published user agent:** anthropic-ai, Google-Extended, Applebot-Extended

## The four sub-scores

| Sub-score | Weight | What it measures | How it is obtained |
| --- | --- | --- | --- |
| Prose parity | 50 | How much of the page text survives without JavaScript | Raw HTML fetch compared against a real browser render, shingled |
| Access | 25 | Whether named AI crawlers may fetch the page | robots.txt per token, plus a live probe per published user agent |
| Structure | 15 | Whether a machine can find the answer inside the page | Heading density and block length over the extracted text |
| Schema | 10 | Whether JSON-LD identifies who published it | JSON-LD parsed from the raw HTML |

## Grade bands

| Grade | Composite score |
| --- | --- |
| A | 90 to 100 |
| B | 75 to 89 |
| C | 60 to 74 |
| D | 40 to 59 |
| F | 0 to 39 |

## Why a probe can be impossible

| Situation | What the report says | What it does NOT say |
| --- | --- | --- |
| Token publishes a user agent | Probed as that agent, result reported | Nothing inferred |
| Token publishes no user agent | robots.txt verdict only, marked not probed | Never implies a live probe happened |
| Page could not be fetched | The outcome, and what was not measured | Never a grade |

## Why the weights are public

A score whose weights are hidden cannot be argued with, and a score that cannot be argued with is a marketing number. They are published here and read from one constant in the codebase, so this page and the scorer cannot disagree.

## The limit that matters most

None of this measures what an AI system does with a page. Nobody outside those systems can observe that. What Lantad measures is whether the content is THERE to be read, which is a precondition rather than a prediction, and the report says so in those words.

## Related

- https://lantad.co/ (markdown: https://lantad.co/index.md)
- https://lantad.co/bot (markdown: https://lantad.co/bot.md)
- https://lantad.co/glossary/prose-parity (markdown: https://lantad.co/glossary/prose-parity.md)
- https://lantad.co/blog/why-we-withhold-a-grade (markdown: https://lantad.co/blog/why-we-withhold-a-grade.md)

---

Lantad measures whether AI crawlers can actually read a page: it fetches as a non-rendering
crawler, renders as a browser, and reports the gap. Free scan, one URL, no signup.

Method and weights: https://lantad.co/methodology | All pages as markdown: https://lantad.co/md | Crawler policy: https://lantad.co/bot