Atlas astrum a star atlas

Software testing that fixes mechanisms, not symptoms.

Exploratory QA, Playwright automation and local-first AI setup, from an independent tester who has been finding out why things break since 2018. Every claim comes with its evidence.

01 Services

Three things you can bring me, and one way I do all of them.

Short, well-scoped engagements for product teams who want to know what is actually wrong, and why, before anyone starts fixing.

01

Manual and exploratory QA

Charter-driven sessions across widths, themes and real data shapes. Bug reports you can reproduce, rated by what they cost, traced to the mechanism that produced them.

About exploratory QA
02

Test automation with Playwright

Browser workflow checks and layout checks, written in Playwright with TypeScript, that run on every change, in your CI, with a clear line between what a green run proves and what it does not.

About automation
03

Automation and AI setup, local-first

Workflows and AI assistance that speed up testing and reporting while your code and data stay on machines you control. Cloud AI only where it is safe to share.

About local-first AI
How

The honest-evidence way

Written, tested, deployed and independently checked are four different things. Every report keeps them apart, so you always know what a claim rests on.

Read the method

02 Evidence

What a release looks like before I let it near production.

These are the checks that ran before release 1.2.0 of Five Rivers, a library system I build and run, shipped on 14 October 2026. The same discipline is what I bring to your project.

848layout checks
73security cases, two databases
24browser workflow checks
51upgrade rehearsal checks
2real display bugs caught first

848 layout checks = 4 screen widths × light and dark × 2 data windows. The two bugs were a phone-width overflow and a dark-mode contrast failure, both fixed before release. Read the Five Rivers case study.

03 Method

Fix mechanisms, not symptoms.

A tester reports what a person experienced. An experience is an outcome. The defect lives in the mechanism that produced it.

So my reports do not stop at “the button does nothing on a phone”. They trace the behaviour back through the layers that built it, name the mechanism, and say what evidence that conclusion rests on. Fixing the mechanism closes the whole family of symptoms, not just the one someone happened to notice.

The same idea runs through how I ship my own software: proof against a copy of live data before anything live changes, phases that refuse to continue on a mismatch, rollbacks that have actually been exercised, and an independent review by a second AI model before release.

The whole method, in one page

04 Work

Two systems I built, written up with nothing private in them.

Both are my own products. Both are in use. Each case study says what is written, what is tested, what is deployed and what has been independently checked.

Case study · web application

Five Rivers

A case, copy and custody system for a network of libraries. Guided releases with a proof phase, three kinds of rollback, and a 30,300-copy dataset that has survived every upgrade intact.

Case study · testing workbench

LumenPort

A local-first browser testing workbench. A controlled browser with per-site permissions, crawls, record and replay, and a chat assistant answered by a local model, so client pages never reach a cloud AI.

05 Start

Three doors. Pick the one that fits where you are.

Everything here is delivered privately to me. No third-party form service, no scheduling widget, no tracking.

Just a question

Send a message

A few lines about what is on your mind. I reply myself.

Want to talk it through

Book an intro call

Offer two or three times that suit you. I confirm one by email.

Have a defined need

Request a proposal

Four short steps: need, timeline, budget range, and how to reach you. You get a written proposal back.