Skip to content
Mark Warren
Menu

All work

Scored audit engine

A scored audit engine with an honest "could not run"

Deterministic checks that score how ready a website is for search and AI answer engines, built so a failed check never looks like a failed site.

How it fits together

  1. Inputs

    • Website to audit

      A request names the site to check.

    • Run budget

      Each run starts with a spend estimate and a hard budget.

  2. Engine

    Audit engine

    A library with its own test suite, compiled into n8n code nodes so the tests and production run the same logic.

    • n8n
    1. Run scored checks

      Crawler access, structured data, content structure, and conversion readiness.

    2. Separate could-not-run

      A check that cannot run leaves the denominator instead of counting as a failure.

    3. Stamp the version

      Every output names the engine version that produced it.

  3. Outputs

    • Scored report

      Scores cover only what was measured, with disclosure text that tests enforce.

    • Partner reporting

      AI-visibility reporting for a marketing agency partner.

Guardrails

  • Calibration fixtures with known answers
  • A lock against running a stale engine
  • The build fails if compiled output drifts from source
  • Per-run spend budget with an estimate mode

The problem

Audit tools often count a check that crashed as a failure, which punishes the site for the tool's own bug. A score is only useful if the reader can tell what was actually measured.

What I built

  • Scored checks across crawler access, structured data, content structure, and conversion readiness.
  • A "could not run" state that leaves the denominator instead of counting as a failure.
  • Calibration fixtures with known answers, and a version stamp in every output so a report always says which engine produced it.
  • A per-run spend budget with an estimate mode, and a lock that refuses to run a stale engine.
  • Disclosure text that tests enforce, and a build check that fails when the compiled output drifts from the source.

How I built it

The engine is a library with its own test suite, compiled into workflow code nodes so the same logic runs in the tests and in production. It also supports AI-visibility reporting for a marketing agency partner.

What it shows

I build measurement so that its limits are visible, and I test the tool before I trust its output.