I build in public.

A QA engineer's defect-evidence tool: capture the screen, annotate what is wrong clearly enough that a developer needs no further explanation, and export both a composited image and the annotation data as structured JSON. Built twice from one specification — a native PySide6 desktop application and a browser build — so the same spec is proved against two independent implementations.

• 335 pytest tests
• 669 assertions
• 49 browser tests
• 122 assertions
• 3 workflows green
• 4 tagged releases

The Python suite splits into a 333-test offscreen lane and a 2-test visual lane, deliberately run as a separate CI job so a slow real-font-rendering check can never fail the main suite. That separation is documented in the workflow itself, with the reasoning.

Repository · Try it in your browser · More about Test Assist

A local-first library tracker with optional Firebase sync. The regression suite covers the hard part — offline-first behaviour, and what happens to local state when sync is switched on afterwards.

• 70 Playwright tests
• 114 assertions
• 11 describe blocks

Stated plainly: suite not yet wired into CI. The suite runs locally and passes; wiring it into GitHub Actions is open work, not finished work.

Repository

A realtime multiplayer PWA where game behaviour lives in editable prompt files rather than code, with structured fallback when a model call fails. Firestore sync, hardened security rules, and a phase decision record for every scope lock — including one written mid-implementation when a design proved wrong.

It is also the most useful thing here, for a reason that has nothing to do with test counts. The repository contains 229 Playwright spec files, 228 of which are generated scaffolds that call test.skip(). Rather than let that file count imply coverage, the repository says so itself:

Read this before drawing conclusions from the number of files in this directory. The count is large and most of it is scaffolding. That is deliberate, and stating it plainly is more useful than letting the file count imply coverage that does not exist.

• 4 implemented tests
• 13 assertions
• 228 declared scaffolds

That README tabulates every figure and carries the date it was last verified. If you are hiring someone to tell you the truth about your software's quality, this is the exhibit that matters — a tester who publishes the number that makes his own work look smaller.

Repository

Including this one. Every deploy runs a Playwright regression suite, an axe-core scan at WCAG 2.1 AA, Lighthouse budgets, and a claim-verification step that re-counts every number on the Work page against the live repositories and fails the build if any of them has drifted. Nothing ships red.

Combined across both Test Assist builds: 384 tests.

How the gates work →