We measured the market, then built its opposite.
Across three rounds we timed 278 Melbourne business websites on a phone. The median took 8.7 seconds to show something and weighed 2.4 MB; 5 of 269 made Google's 2.5-second bar, and 116 of 269 took longer than ten. That is not a design problem. It is a specification, written by the market, for the site nobody in it is building.
So we built it. Below are 8 trade and clinic sites, whole and working — one for every category in the study. None of the businesses is real; they exist so you can open the thing itself instead of reading a list of promises about it.
01 · The same instrument, both times
8.7 seconds and 2.4 MB, against 1.6 seconds and 24 KB.
Comparisons are only worth something when both sides are measured the same way. Google Lighthouse 12, mobile profile, throttled connection, no extra-headers — the same instrument used on all 278 sites in the studies, taking the median of five interleaved live runs per site (8 Nov fault-only + cadence-decision checkpoint, closing the 1-7 Nov empty-backlog run: no code touched — site-audit 0 errors/3 known warnings, AHPRA scan clean on all four clinics (urgent-bar lines state same-day/this-week availability by symptom only, no testimonial, no superiority claim, no treatment pressure), no horizontal overflow at 390px on any of the eight (scrollWidth=clientWidth=innerWidth=390 via a mobile-emulated headless Chrome, audit header on every request). TBT 0 and CLS 0 held on all 40 readings; weight 24KB on six of eight, Fernridge read 25KB on 2 of 5 rounds — noise, median still 24. The family's known bimodal swing continued: Tallowood and Fernridge read 99 this session, the other six 100; LCP spread 0.9-2.0s round to round with no code touching any site, so published the honest medians rather than chase session noise with cuts. Published honestly; this measuredAt is the actual time of this run). No hand-picked run, no desktop profile, no "up to" — the figures below are the worst of the 8, not the best. Measured 8 November 2026 at 4:59 am Melbourne time.
02 · Open them
Whole sites, not screenshots.
Tap the number, read the printed prices, open the questions, send the form. Every part works. Every one carries a banner saying it is a demo, because a page that pretends a business exists is a lie and we do not publish those. The four clinics are written to AHPRA's advertising rules — no testimonials, no superiority claims, nothing that presses a patient toward treatment — because a registered practitioner cannot use copy that is not.
03 · Why they are like this
Every decision on the page was made by the data first.
A studio designs a site, then measures it, then apologises. We measured 278 of your competitors before a single one of these pages existed, and every choice on them answers something the sample got wrong — the 8.7-second wait, the buried phone number, the price nobody prints, the hours no assistant can read. The speed is not a feature bolted on at the end. It is the shape the evidence forced.
That is the whole argument for using us, and it is not one an agency can borrow. It needs the 278-site sample, the instrument that took it, and an engine trained on the answer by Auden, the AI we built and trained here — none of which arrives with a design brief. Nothing we send a client leaves here without a person reading it first.
The study these figures come from →