Method / 01

Most are refused.

Orbcrawl reconciles UK public records into single businesses, runs deterministic checks on what a visitor sees, and declines the large majority — most of them because the site is already fine. A conclusion is only useful when its source and date survive with it.

1,066,536 businesses on record, and 79% of the websites checked so far refused · snapshot of 28 September 2026

01

Ingest

We read UK public records. Every source row is stored before anything is resolved.

02

Resolve

Records are reconciled into single businesses without pretending weak matches are certain.

03

Inspect

Each business’s website is found, verified and checked with deterministic, observable facts about what a visitor sees.

04

Refuse or rank

Site quality stays separate from commercial fit. A sound site is refused; a blocked crawl gets no verdict; a genuine problem may be ranked.

The whole loop

What happens to one business, from public record to decision.

Each stage below is built. Each says whether it runs today, at partial coverage, or is exercised internally — because the difference is the whole story, and a roadmap in the present tense is how a claim stops being worth anything.

  1. 01

    Read UK public records

    UK public records are read, and the source row is stored before anything is resolved, so every later claim can answer “where did this come from” with a row and a date.

    Partial coverageRead UK-wide
  2. 02

    Work out who is who

    One business can appear several times under different names. The listings are reconciled into a single business, with the chain of evidence kept behind it.

    Partial coverageReconciled, provenance kept
  3. 03

    Find the domain

    A business’s website is identified and verified against what the page itself says before it is accepted. A guess is never the answer.

    Partial coverageDiscovery still in progress
  4. 04

    Audit the site

    The site is fetched and judged by deterministic checks on what a visitor sees. No model is asked for the verdict, so the result is reproducible and can be argued with.

    Partial coverageAudit still in progress
  5. 05

    Refuse most of them

    A sound site is left alone and a blocked crawl gets no verdict rather than a bad one. What survives is ranked by likely fit, which is a different question from how poor the site looks.

    Partial coverageMost are refused
  6. 06

    Assess the social profile instead

    A business with no website has nothing to crawl, but may keep a public social profile. A profile can be assessed in the same spirit, with each weakness named alongside a fix. The stage is built; nothing feeds it automatically yet.

    Built, not yet fedNot yet automated
  7. 07

    A named person approves

    Downstream of the qualification decision, a one-page preview can be prepared from a business’s own public material, with every claim traced to its source. Nothing goes further unless a named person approves it: a human approves every message, and any contact follows PECR. The gate is the point, not a stopgap.

    Gate on, by designHuman approval required
  8. 08

    Read what comes back

    Outcomes are recorded so they can be learned from. We intend to use them to calibrate ranking once there are enough of them; until then the ranking rests on judgement.

    BuiltNot yet calibrated

Two scores, deliberately

“Bad website” and “worth contacting” are different questions.

A dead site may belong to a closed business. A dated site may still do its job. Orbcrawl keeps technical quality separate from trading liveness and reachability so one score cannot disguise the other.

Blocked does not mean broken. Unknown does not mean absent. Those distinctions are enforced in the system rather than added as caveats later.

Known limits

What the current system does not claim.