Home / Method
How we grade
The rubric,
in the open.
This is our website audit methodology, weights and all: how the pipeline runs, and where we draw the line between what we measured and what we estimated. If you are going to act on a grade, you should be able to check how it was made.
Scan. Score. Synthesize. Recommend.
A method is only worth trusting if it runs the same way every time. Ours has four stages. Each does one job and hands a labeled result to the next.
On this page
The pipeline: scan, score, synthesize, recommend The 5 axes, the 26 metrics and the Reputation lens Measured vs. estimated, and the promise behind it Why we grade before we build Who we are, and how to check us Frequently asked questionsScan
Read signals off the live site with Lighthouse, axe-core, headers and TLS. Nothing is copied or stored.
Score
A weighted model grades 5 axes and 26 metrics, and every reading carries its own confidence level.
Synthesize
The grades become one plain-language read on strengths and risks, with the trade-offs spelled out and the confidence labels kept.
Recommend
Fixes in priority order: biggest impact first, weighed against effort, each with its expected score lift.
No stage runs on opinion. Two people who put the same site through it under the same conditions get the same result. That repeatability is what turns "this site feels slow" into a number you can defend in a board meeting. Anything observed in the scan is a measured fact. Where a real measurement can't be taken, the gap is recorded rather than quietly filled, and anything used in its place carries an Estimated label.
Five axes, 26 metrics, and a sixth lens.
The website quality score is 26 distinct metrics grouped into five axes. Each axis is scored on its own, then weighted into the overall grade.
Performance
How fast the page loads and responds. Core Web Vitals and real load behavior, measured with Lighthouse.
Accessibility (ADA / WCAG)
WCAG violations flagged by axe-core, and the business and legal risk each one carries, explained without the acronyms.
Findability (SEO & AEO)
Whether search engines and AI answer engines can read the site, index it and quote it, from structure to metadata.
Security & Privacy
TLS, response headers and trackers, read directly from the live site. These are the gaps attackers and regulators notice first.
Trust & Compliance
The credibility signals and obligations a serious site is expected to meet, framed as business risk. This is not legal advice.
Reputation, the sixth lens
Beyond the five scored axes sits Reputation: what the wider web says about you, from reviews to sentiment. We treat it as a lens rather than a hard-scored axis because much of it can't be measured directly off your site. Everything in it is labeled by confidence, and anything resting on estimate rather than observation is marked Estimated, so nobody mistakes it for a verified reading.
We label our confidence. Always.
Every lens carries a confidence tag. We would rather tell you a number is an estimate than dress a guess up as a fact.
A direct reading from a live tool (Lighthouse, axe-core, header inspection). Reproducible, and reported as fact.
Worked out from observable signals while full live measurement rolls out. Flagged, and never blended into measured scores.
One limit: the Grader you can run today still estimates some signals instead of scanning them live. That is why the labels exist. As the real-scan backend (Lighthouse and axe-core behind a live /api/scan) rolls out, more of every grade moves from Estimated to Measured, and you will see it happen on the page.
We will never present an estimate as a verified scan. If a number is estimated, it says so.
Please note
This assessment does not constitute legal advice.
Why we grade before we build.
Most dev shops guess at quality and ask you to take their word for it. We measure first, the same way every time, and let the evidence set the agenda. The grade is the brief. That is what "evaluation-led development" means, and it is why the method is published here instead of kept in a drawer. We don't guess. We grade.
Built by people who stress-test their own grade.
Houtos Labs is an evaluation-led web, app and web3 development firm founded by Ryan Bonelli, who also leads Peregrine Precision Systems. The grading engine, the Peregrine Precision Evaluator, is named for that lineage. The Grader is our flagship and our calling card: a website-audit tool anyone can run, which puts the claim we make to every client where you can test it.
We hold ourselves to the bar we publish. Accuracy comes first, and the engine is stress-tested against real sites. Every lens has to come with a plain-language explanation of what it means and what fixing it would change; that is a rule, not a style choice. When the deterministic model hits its ceiling, we say so and build the real scanner instead of dressing up the number.
We publish how the grade is made because we want it judged on its method, not on how sure it sounds. If you find a way to break it, tell us. That is how it gets sharper.
Questions about the method.
How accurate is the website grade?
It depends on how each signal was gathered, which is why every metric is labeled Measured or Estimated. Measured signals (speed, accessibility violations, security headers) come from running real tools against the live site and are as accurate as those tools. Estimated signals are used only when a measurement is unavailable, and they are never reported as verified facts. We state the limits of any estimate rather than inflate the number.
What's the difference between a measured and an estimated signal?
A measured signal is one we observed directly by scanning your live site, which happens once live scanning reaches that lens: a WCAG violation flagged by axe-core, say, or a Largest Contentful Paint time from Lighthouse. An estimated signal is worked out from the structural evidence available when a direct measurement could not be taken. Measured signals carry a Measured chip and estimated ones an Estimated chip. We never present an estimated figure as a verified scan.
What tools do you use?
Performance is measured with Google Lighthouse, the industry-standard performance and best-practices auditor. Accessibility is measured with axe-core, the open-source WCAG rules engine used widely across the accessibility industry. We also read security response headers, TLS configuration and page structure directly. Where a real measurement is not available yet, we fall back to an estimate and label it as one.
Can I trust the score?
Trust it as far as its labels say, and no further. The method is repeatable: grade the same site twice under the same conditions and you get the same result, so the number is defensible rather than a matter of taste. Every figure traces back to its evidence and carries a confidence label, and we will always tell you when a result is an estimate rather than a verified scan.
How often should I re-grade my website?
After any meaningful change (a redesign, a platform migration, new third-party scripts, a lot of new content), and otherwise on a schedule, monthly or quarterly. Because the method is repeatable, comparing grades over time shows whether a fix moved the metric it targeted and whether anything slipped backward.