My tests were green. The app was broken. Both were true.
I tore my response layer apart, one file of 6,533 lines and 90 handlers, split it into seven modules, and rebuilt how hck_GPT thinks. Ask it to optimize your PC and it now walks you through it step by step, then proves the result with a measured delta: "RAM 82% to 61%".
Scars are the story
Scar 1. The refactor silently switched my whole AI layer off. No crash. The only reason I know is a guard test I wrote the same day, for a blind spot I did not predict.
Scar 2. A cleanup pass converted debug prints to a gated logger. Result: an error on the first line of the navigation handler, and every sidebar page silently fell back to the dashboard. My 96 tests stayed green, because not one of them built the real window. A five-minute human click test caught what the whole suite could not. Now one test boots the actual window and clicks every page.
Scar 3. Bumping to v1.8.4 I audited every version string. Version now lives in one file, and a test fails on any hardcoded copy, anywhere.
Also in the build: the sidebar rebuild I promised on Monday, one merged FANS page, a cleaner Guide.
- Tests: 21 in June, 149 today.
- builder.py, from 6,533 lines to a 590-line facade.
- Engine-layer pyflakes from 221 issues to 0.
A hectic, intense week.
This is the project behind the post. PC Workman is a free, open-source Windows system monitor
with an offline AI assistant - everything described here is real, shipped code.
Download it or read the source.