Every project here opens with a number we measured, not an adjective
Fast, accessible and well made are things every studio writes about itself. We run the tool, write down the result and the date it was taken, and leave the parts that are not perfect yet on the page.
Full disclosure, before you read another line
The three projects below were built for companies inside our own group, not for outside clients who bought from us on the open market. We are the creative arm of the Dogware group, and these sites belong to our sister companies.
We say it here rather than in a footnote, because one question from a prospect would surface it anyway. The credibility does not come from the logo at the top of the site. It comes from the numbers being real, reproducible, and something you are welcome to run against us yourself. The external client book is opening now.
01 · dogware.ai · performance
The group site, on a WordPress theme we wrote ourselves
The brief: a home for the group, fast enough that it could carry the group's own argument about speed.
Measured 25 July 2026 · Lighthouse 12.8.2, desktop preset, six live runs against https://dogware.ai/ · the figures above are the median, range in brackets: performance 99 (98–99), LCP 0.8s (0.78–0.93), CLS 0.002 (0–0.002), TBT zero on all six · same series: accessibility 100, SEO 100, best practices 100.
What we did
We wrote our own WordPress theme, Dogware Flex, and shipped the site on it with no plugins at all. No page builder, no animation library, no stack of third party scripts. Images go out at measured sizes, fonts load without blocking the first paint, and the layout is reserved up front so nothing jumps after load. That is why total blocking time came back as zero rather than as a small number.
The tradeoff, honestly
These numbers come off a stack we control end to end: our theme, zero plugins, our hosting. A client site that has to carry an external booking system, an advertising pixel and a live chat widget will not reach 99, and nobody can promise you otherwise. We say that at quote stage rather than after delivery, and we agree a target number with you that is achievable with the tools you actually need.
One more thing about the number itself. A single Lighthouse run moves a point up and a point down, so we ran it six times and published the median and the range rather than the prettiest run. If you run it yourself and get 98 instead of 99, that is inside the range written here, not a contradiction of it. A studio that publishes its best run as if it were its average has already started selling you something.
dogware.ai, the home page as measured on 25 July 2026.
dogware.aiThe live site. Run Lighthouse on it yourself.
02 · ziko.co.il · accessibility
The compliance report surface of an Israeli legal-compliance product, component by component
The brief: the report the client reads is the product itself, and it had to look like something you could put in front of a regulator.
0WCAG 2.1 level A/AA violations on a live scan of the report, at both widths
15/16
rules run, and passed
1
item left for manual review
1920
checked on desktop
390
and on mobile
Measured 25 July 2026 · axe-core against the wcag2a + wcag2aa + wcag21a + wcag21aa tags, live run on a real report · 16 rules ran, 15 passed, one for manual review, zero violations, at both widths · on 24 July the same scan returned one violation.
What we did
We wrote a component by component elevation spec for all six parts of the report, then a nine section style guide showing every component in every state it can be in, including empty, loading and failed. The client's own developer implemented it in their code. The rule that held the whole report together: severity reads through font weight, an inline-start hairline and a glyph, never through colour. That way a finding stays a finding for someone who cannot separate red from green, and in the black and white printout that goes to a lawyer.
What this said on 24 July
Until 24 July a less flattering number stood here, deliberately: one violation left in the report, a call to action button still carrying the colour from before the redesign, white on it measuring 3.82:1 against a standard that asks for 4.5:1. We found it with the same scan we run for clients rather than by eye, we published it here while it was still broken, and we handed their developer the exact node and the rose our spec locks. He applied it. We rescanned on 25 July: white on that same button now measures 6.04:1, exactly the value in the spec, and zero violations at both widths.
What is still a real constraint is the one item the scan returns as needing manual review. No automated tool decides it, we check it by eye, and here it is counted separately rather than as a pass. We are keeping both versions of this paragraph for one reason: a studio that only publishes the scans that came back perfect is a studio that is not really scanning, and a number that improved is worth something only if you also saw the one before it.
From the elevation spec we wrote: the score ring and the finding cards, with component annotations.
ziko.co.ilThe report itself is a private per-client page, so we link the site rather than a report. The number was measured on a live one.
03 · native.dogware.ai · Hebrew and RTL
A service site built Hebrew first, rather than poured into a Latin template
The brief: a site for a localization company, where one mistake in text direction is evidence against the product it sells.
0direction defects across the three failure classes we test, at two screen widths
0
direction flips inside the flow
0
ragged start edges
0
horizontal overflow
19/19
accessibility rules passed
Measured 24 July 2026 · our measurement script audit_rtl.py at 1920 and at 390, plus axe-core on WCAG 2.1 A/AA: 19 rules, all passed, zero violations.
What we did
We built the whole surface in Hebrew from the start, using logical CSS properties only: start and end instead of left and right, so the layout mirrors correctly on its own and there is not a single per-side patch. Latin brand names, language codes and figures are isolated to their own direction inside the Hebrew sentence, and that is exactly where Hebrew sites break: a punctuation mark that jumps to the wrong side at the end of a line. We run this as a measurement rather than by eye, because the eye misses a few pixels of disagreement between neighbouring edges.
The constraint, honestly
The site shipped in Hebrew only, and its English surface is a second phase that is not live yet. So what is measured here is RTL correctness and the quality of the Hebrew, not full bilingualism. For bilingualism the exhibit is this site, which runs Hebrew and English side by side.
native.dogware.ai, the home page as measured on 24 July 2026.
An eight page brand book, an eleven file vector logo pack, a locked palette and a social kit, for the Dogware group. The full case study is open to read.
Tell us about the project and we will come back with a detailed quote within one business day. We agree the measurable target with you up front, and measure it with you at the end.
