Judge should read clearly the moment you land on it
Judge now has two explicit lanes. Public users browse deterministic repo scorecards. Operators run a pinned static scan, then optionally attach an AI review as a second pass.
Public grade
Static, deterministic, and tied to one pinned commit. This is the public score users compare over time.
AI review
Optional second pass for richer context. It is visible separately so the public grade never feels like an arbitrary LLM verdict.
Operator import
There is now a real submit surface at `/judge/submit` so importing a snapshot no longer requires guessing the admin API contract.
One public score, one optional AI review, one pinned commit
Run static on a pinned commit
This creates the deterministic public grade and the historical scorecard.
Optionally run an AI review
Import it as a separate `agent` snapshot so it adds context without rewriting the public grade.
Publish the scorecard
Users see the grade, commit, scan date, findings summary, and whether an AI review exists.
Need to run and upload a repo yourself?
Use the Judge submit page. It gives you the run commands, the accepted report formats, and a real import form wired to the existing admin endpoint.
Public scorecards
pallets/itsdangerous
networkx/networkx
psf/black
Choose the right Judge path
Public repo suggestion: use the form on the left when you want a repo added to the queue.
Operator import: use Judge submit when you already have a pinned report and want to publish or update a scorecard yourself.
AI judge / AI review: yes, Judge can store it. Import it as the optional agent snapshot so it stays separate from the deterministic public score.