build-log / experiments

I test before I say “fixed”.

I record what I built, what I actually ran and what those results still leave unanswered.

I freeze expected behavior first. REPRO must reproduce the original failure, then test a patch against the same assertions.

I keep implemented code distinct from completed runs. A passing target test covers only the cases listed.

01 / IMPLEMENTATION

What I built

How I separate expected behavior, the original failure, a candidate patch and the test verdict.

The first hard rule in my implementation was that a patch could not move the test that would judge it.

I already had a streaming pipeline. The useful next step was to make its decisions inspectable without building a second system to animate them.

02 / CONTROLLED RUN

A controlled fixture run checked the sandbox harness; no Groq request was made.

I ran three small functions in a real Vercel microVM to see whether the assertions, rather than the process exit code, would decide the result.

03 / SITE NOTES

I record how the site presents the story, source code and project evidence.

The screenshot showed a black rectangle around the homepage bug, so I traced its edge back to the source image and fixed the transparency there.

I changed two reading problems together: the background left too much unused space, and Source made readers scroll inside a code box.

I changed the site shell so REPRO's story, experiment and source code each have room to do their own job.