proofrun/loopvera

LoopVera is an early stage, in progress project that wants to fix a specific frustration in computer vision development: when a model's accuracy suddenly drops, developers usually end up saving debug images by hand, eyeballing folders full of screenshots to compare versions, and quickly losing track of what changed and why something seemed to work. LoopVera's goal is to build a tool that automatically records every intermediate step of a vision pipeline, lets you tweak a single setting and rerun it, then shows a clear side by side comparison between runs, with a human able to sign off on changes when the stakes are high. The README is very upfront that this repository currently contains only planning documents, a roadmap, and design writing, not working software yet. The commands and code examples shown, like installing it with pip or opening a local browser workbench, describe what the finished tool is meant to feel like, not something you can run today. The project is being built and released in public, in stages, with the current stage being a runnable governance shell planned as the next milestone. Underneath the vision debugging idea, the author frames a broader concept called a runtime evidence ledger, meant to track what changed, why, and who approved it, which could apply to other fields beyond computer vision, such as embedded systems or medical devices, in later phases. The project also positions itself as something AI coding agents could plug into, so that automated changes to a pipeline leave the same kind of traceable record a human change would. This is a single developer's solo, build in public effort, not a company or a finished product, and the README explicitly asks for feedback on the direction rather than code contributions at this stage. No license file is mentioned in the README shown here.