Every PR scored on what actually matters
Izri runs four independent quality signals on every push and aggregates them into a single izri/quality umbrella. One hard rule fails the delta; a chorus of weak soft signals fails it too.
Scope
Does the diff match what the PR said it was going to do? Did it touch paths it shouldn't?
Tests
Did the relevant tests run, and did they pass against the diff?
Hallucination
Do the tests that passed actually exercise the lines that changed? Or are they passing without touching the diff?
Visual
What visually changed on the routes the PR touched?
izri/quality — the umbrella
The four signals roll up into a single verdict with a hybrid worst-of-hard-rules + composite-soft-signals rule. One hard finding on any child fails the umbrella; absent hard findings, the soft scores are weighted and compared against a per-project threshold. Signals that didn't run for this delta have their weight redistributed — you're never penalized for not having wired a signal yet.
Aggregation model