Skip to main content
Allways

CASE STUDIES

Three projects. One of them is this site, and that one you can audit end to end today.

The three projects

Web Development

Allways Website

CLS 0.000Measure it yourself: the PageSpeed link is on the home page
Our own project
Whoever buys software cannot check what they bought until they have already paid for it. We had the same problem in reverse: asking you to take our word.

Next.js 16 and React 19, with a single reveal system that also works without JavaScript, an in-house graphics engine in the hero and a custom scrollbar.

The six checks below come from the project's own measurement files, and they are recalculated from those files every time the test suite runs.

Scroll down to the six proofs and repeat whichever you like. Two of them are measured with tools that are not ours.

0 / 5,284contrast failures
0 / 30accessibility errors
0without JavaScript
Next.js 16React 19TypeScriptFramer MotionGSAP
Real source from this repository, verbatim. Our codebase is written in Spanish.
AI Orchestration

Sisyphus Engine

Autonomous Orchestration
Reviewing every delivery by hand does not scale, and skipping the review is exactly where the expensive problems get in.

Multi-agent system that orchestrates quality, evolving memory and pre-deployment validation on top of Claude.

The three figures on the card are not typed in: they are counted against the repository on every test run, and the approval threshold is published in the document that sets it.

If the count stops matching what is on disk, the check fails.

5layers
10modules
85/100approval threshold
Claude CodeMCP ProtocolPythonTypeScriptChromaDB
Illustrative excerpt of the architecture. This system does not live in this site’s repository, so its source cannot be opened from here.
B2B SaaS

Command Center

Pipeline with no manual stepFrom capture to CRM, with nobody in front
A sales team burns its hours moving data between tools that do not talk to each other, and whatever falls through the cracks is never claimed by anyone.

End-to-end B2B sales pipeline: capture, scoring, enrichment and CRM sync. It runs on its own, and it raises a flag when something breaks.

This system does not live in this site's repository, so its three data points are qualitative on purpose: they describe the scope, not a quantity we could not count from here.

Capture → CRMend to end
No shiftsruns on its own
It flags itwhen something breaks
Next.jsSupabasePythonAutomation
Illustrative excerpt of the architecture. This system does not live in this site’s repository, so its source cannot be opened from here.

The one case you can audit yourself

This site is built with Next.js, React, TypeScript and Tailwind. The cases above use other pieces —each card publishes its own— and they do not live in this repository.

  • 0.000layout shift
    Nothing moves around while the page loads. No button sliding out from under your thumb.Measure it with PageSpeed Insights
  • 0 / 5,284contrast failures
    Five thousand two hundred and eighty-four text nodes measured across the thirty routes. None below the legible minimum.Measure it with WAVE
  • 0 / 30accessibility errors
    Automated axe violations across the thirty routes. The manual review is counted separately.Measure it with WAVE
  • 0without JavaScript
    With JavaScript turned off the site stays whole: the content reads and the links work.Turn JavaScript off in your browser and reload any page. It reads the same.
  • 390screens with no overflow
    Thirty routes by thirteen widths, from a 320 px phone to a 2560 monitor. On none of them do you have to drag the page sideways.Stretch and shrink the window on any page: no horizontal scrollbar appears.
  • 42b26bb4reproducible seal
    The simulation behind the graphics engine on the home page yields the same result, byte for byte, on any machine. Change a line that alters the physics and the seal changes.With the repository in hand: `npm run repro`. It has to print that same seal.

None of these six figures is typed in by hand. They come from the project's measurement files, and an automated check recalculates them from those files every time the test suite runs: if one stops being true, the check fails.

Want something like this for your company?

Want your project to be the next one?

Tell us what you are building. We reply in under 24h.

And if we run late, what happens is written down: what happens if we run late.