Methodology
Instruments first. AI second. Scores locked in between.
Plenty of tools will paste your website into a chatbot and call the answer an audit. We don't. A Nova Insight audit is a measurement pipeline: real instruments collect the evidence, deterministic rules compute the scores, and only then does AI help explain what the numbers mean. Here's how it works — openly, minus only the recipes.
Instruments measure first
Before any analysis happens, real instruments collect evidence: Lighthouse performance runs, a headless Chromium browser rendering your pages the way a visitor's does, and a battery of security, SEO, and accessibility probes. Our crawler is polite — it identifies itself, respects your server, and never probes beyond what any visitor could see.
01 lighthouse performance runs
02 headless-browser rendering
03 tls / header / robots probes
04 structured-data + accessibility checks
05 crawler: NovaInsightAudit/2.0
Scores are locked before AI sees anything
Your scores are computed by deterministic rules from the evidence — pure arithmetic, no judgement calls, no AI. Only after the numbers are locked does any AI model get involved. That's why the same site gets the same score, every time. An audit whose score changes with the weather isn't an audit.
AI explains — within strict guardrails
We use frontier AI models to do what they're actually good at: reading the evidence and explaining what it means for your business, in plain language. Every AI response is validated against a strict schema before it reaches your report — if the analysis doesn't conform, it's rejected and re-run. The AI can describe your scores; it can never move them.
The methodology is versioned
Every audit records exactly which version of our methodology produced it — down to the individual analysis prompts. If we re-run your audit, we can reproduce it with the versions that built it, not whatever is newest. When the methodology improves, the version number says so.
Changes are tested against reference sites
Before any change to the engine ships, it re-runs against a fixture suite of reference sites with recorded baseline scores. If a change moves any baseline by more than a few points, it's blocked until a human explains why. Consistency isn't a promise; it's a gate in our release process.
A consultant reviews before delivery
Paid audits are reviewed by our team before they reach you. The pipeline is rigorous, but judgement about your business — what to prioritise, what to say to your board, what to hand your developer first — is a human's job.
Why we built it this way.
Repeatable
Same site, same score, every time. If your score moves, your website moved — not our mood, and not a model's.
Comparable
Because every site is measured on the same locked rubric, your score can be honestly compared — against competitors, and against your own site last quarter.
Accountable
Every finding traces to recorded evidence, and every audit records the methodology version that produced it. Ask us why a number is what it is, and we can show you.
House rules
Three things we refuse to do.
No invented numbers
If a claim isn't backed by collected evidence or our own pipeline data, it doesn't go in a report — or on this website.
No fear-mongering
Findings are stated plainly, with effort and impact estimates, so you can decide calmly. We sell clarity, not panic.
No intrusive scanning
We observe what any visitor or search engine could. We don't attempt to breach, stress, or probe beyond the public surface of your site.
See the instruments pointed at your own site.
The free scan runs the same pipeline on your homepage — scorecard in your inbox in about fifteen minutes.