file 03 · residential and commercial solar and battery · australia · 2026
this client is shown without their name or their logo, at their request. the figures are theirs and are unchanged. solarprojects is a stand-in name.
a four-office australian solar and battery installer, trading since 2014, doing residential and commercial rooftop pv plus storage, with the commercial side, 80 kw to 200 kw rooftops and embedded networks, as the priority book of business. the existing site was a 2019 build that measured badly on every axis.
what was wrong
- a 146 kb render-blocking stylesheet on every page.
- the homepage html was 240 kb raw, 40.3 kb gzipped, because a decorative field was drawn as 5,489 svg circle elements in the markup.
- lighthouse performance 74. the largest element took 5.0 seconds to appear.
- total blocking time 350 ms.
what was decided
the decorative field moved out of the html.
5,489 circle elements became 14. the homepage html went from 240 kb to 67.7 kb raw and 40.3 kb to 14.4 kb gzipped, and total blocking time went from 350 ms to between 90 and 190 ms.
the offices map is drawn, not embedded.
one inlined svg path from public-domain natural earth geodata: 0.97 kb raw, 0.47 kb gzipped, zero extra requests, no api key and no account. a real embed needs a cloud account with billing attached and is one of the heaviest things that can go on a page.
rejected: a live interactive map. if it is ever wanted it goes on the contact page only, behind a click-to-load placeholder so it stays out of the first paint, and the gate gets re-measured.
no verifiable fact about the business was invented anywhere.
no licence number, no saving or payback figure, no certification, no review count, no date and no client quote was written that could not be sourced. where a fact was missing the design works around it, which is why one section is a row of names and system sizes rather than figure cards with empty slots.
the numbers
every figure carries three lines: what it measures, the file it was read from, and what it does not prove.
146 kb to 4.1 kb
the render-blocking stylesheet on every page
m REPORT.md
does not prove: the stylesheet alone, not the whole page.
127 kb to 41 kb
the shared css bundle across the site
m REPORT.md
does not prove: same change, measured at the bundle level.
74 to 90, and lcp 5.0 s to 2.3 s
lighthouse performance and largest contentful paint on the first measured build
t REPORT.md
does not prove: lighthouse is simulated lab data on one machine. it is not what a real visitor experienced.
5,489 to 14
svg circle elements in the homepage html
m REPORT.md
does not prove: element count, not rendered appearance. the field looks the same.
240 kb to 67.7 kb raw, 40.3 kb to 14.4 kb gzipped
homepage html weight
m REPORT.md
does not prove: html only.
350 ms to 90-190 ms
total blocking time
m REPORT.md
does not prove: a range because it was measured across several runs.
0.97 kb raw, 0.47 kb gzipped
the drawn offices map, from public-domain geodata, as one inlined svg path
m REPORT.md
does not prove: it is a drawn map. it does not pan, zoom or search.
99 / 100 / 100, cls 0, tbt 0 ms
lighthouse mobile on the built output: performance, accessibility, best practices
t REPORT.md
does not prove: simulated throttling, served from disk. a real network and a real cdn will differ.
680 to 800 ms, cls 0 on all five
largest contentful paint across five pages at 390x844 under applied 4x cpu throttle and slow 4g
m REPORT.md
does not prove: applied throttling on one device profile. the two lab models disagree by more than 2x on this project, which is why both are published.
0 of 1,060
broken internal links across 44 files
m REPORT.md
does not prove: internal links only. external links rot on their own schedule.
22.0 s desktop, 21.9 s on a 390px phone
a real person completing the enquiry form, keyboard only, at 300 ms per character with a 1.2 s pause to read each label
m REPORT.md
does not prove: one person, one form, one run each. it is a timed walkthrough, not a usability study with a sample.
the form was timed by a person typing it, keyboard only, at about forty words a minute with a pause to read each label, because a form that takes under a minute is a form somebody finishes. it came out at 22.0 seconds.
what they own now
- the repository, the built site and the deploy workflow
- a written report naming every placeholder, every designed-around absence and every switch to flip at launch
- a stated boundary: no fact about the business was invented, and the report says where each one came from
what this does not prove
this section is here on every file. it is the reason the rest of the page is worth reading.
blocked out: the rebuild increased enquiries
the site is a preview awaiting approval. there is no live traffic, so there is no enquiry figure, and this run stays painted out until there is one.
the client is not named here and their logo does not appear, at their request. that is why this page carries no brand marks and no screenshots of the identity.
the site was built as a preview for approval. it is not indexed and is not connected to a paid service.
the photography in the preview is free-licence stock and the client’s own images, listed for replacement in the report. none of it is presented here as their work.