I built a repair comparison tool because passing tests kept wasting my time
I did not build Code Factory because I wanted another AI coding platform. I built it for myself because I was tired of the same cycle: a coding assistant made a change, the tests went green, I trusted the result, and the feature still failed when I actually used it.
That wasted time was worse than the original bug. I had to reconstruct what changed, whether the test could ever fail, and which “fix” was least risky. The frustration pushed me to bring the process under one proof-first workflow, and I am sharing it free so other builders can avoid some of that pain.
The latest release adds ProofSearch to Graph Ops. It compares several supplied repair candidates, rejects anything with failed evidence, surviving test mutations, changed receipts, or edits outside the approved slice, then explains why the smallest verified candidate won. It does not silently apply the winner; that decision stays with the person reviewing the work.
Solo builders get a readable answer before accepting an AI fix. Teams get a hash-bound decision record they can reproduce.
I also ran the Graph Ops UI through the open Prestige design skill. From my own use, it made a material difference to the visual hierarchy, readability, and polish. That is my observation, not a made-up conversion statistic, and I would genuinely welcome fresh feedback.
The savings panel is equally strict: without an exact paired baseline, time, token, and cost savings say “Not measured.”
Free release:
https://github.com/zrk222/code-factory/releases/tag/v0.32.0
What is the most expensive “green test, broken feature” incident you have had?
The distinction between green tests and a working feature is painfully familiar. Curious what kinds of failures ProofSearch catches that your existing tests consistently miss.