Skip to main content

How it works

Eight steps. The evidence is kept at every one.

Seymour is built so that every number can be traced back to a page it captured.

Seymour, a yellow Labrador guide dog in harness, lying down alert with a purple and white magnifying glass held to his nose. He is the detector: he reads the page so everyone can see more.

The steps

From a register to a rate.

  1. DiscoverSeymour starts from official registers, then follows sitemaps and links to find public pages.
  2. FetchIt reads robots.txt first and keeps to one request at a time per site, at most one every 2 seconds and usually slower, slowing down at once if a site struggles. A kill switch stops every crawl at once.
  3. CaptureEach page is stored exactly as served and fingerprinted, so a finding can always be checked against what Seymour saw.
  4. ExtractText, footnotes, and images with their exact position on the page.
  5. DetectRule-based, with no AI. It is fast and repeatable: the same page gives the same result every time.
  6. ClassifyIs the phrase a quotation, history, guidance, a legal name, or the organisation’s own voice? Context changes what a match means.
  7. Human reviewPeople confirm, reject or amend candidates. Review never overwrites the evidence underneath.
  8. MetricsVersioned counts always carry the number of pages they came from, so rates can be compared fairly.

Principles

Four things Seymour holds to.

  • Coverage is evidence

    “Found nothing” is different from “did not look”. Seymour reports both.
  • Evidence and interpretation stay apart

    What the page said is stored separately from what we think it means.
  • No combined score

    Language, accessibility and disclosure are reported on their own. Nothing is blended into one grade.
  • Public information only

    Seymour reads published pages. It never logs in and never reads anything private.

Politeness

A guest on your site, and he behaves like one.

  • Reads robots.txt first

    If you say no, Seymour does not visit.
  • One request at a time

    Per site, at most one every 2 seconds and usually slower. A kill switch stops every crawl.
  • Says who it is

    It always identifies itself as SeymourBot/1.0 (+https://seemeplease.com/bot).

To opt out or slow Seymour down, read the opt-out guide. Seen it in your logs? See the crawler information page.