Yann Martel's Life of Pi ends with a survivor offering two irreconcilable accounts of the same 227 days and asking his investigators not which is true but which they prefer [1]. This paper takes that scene as the best available model of a situation clinicians meet constantly and are poorly trained to recognise: the patient who has more than one account, both their own, mutually untranslatable, and who is watching to see which one the room will accept. Two decades of criticism on the novel have divided sharply over whether its refusal to adjudicate is a philosophical achievement or a sleight of hand, and I survey that division because its terms are exactly the terms of the clinical problem. I then argue that the interpretive tradition best equipped to handle a text that is undecidable in reference while remaining determinate in use is the midrashic one, and that the rabbinic mashal — as analysed by Stern, Boyarin, Fishbane, and Kermode — supplies the structural category that the narrative-medicine vocabulary of "the patient's story" lacks. The mashal supplies an application and then withholds its sufficiency, leaving a residue the application fails to collect. Martel performs precisely this: he hands the reader the key, through the investigators' own decoding, and the first story does not collapse. Four consequences for practice follow — that the addressee participates in producing the version, that accounts are maintained at cost, that anomalous detail signals the account's limit rather than its falsity, and that institutional documents systematically require the version that helps patients least. Two symmetrical clinical errors are named, four criteria of discernment proposed, and the limit case examined, where no account can be constructed at all. Throughout, I distinguish three questions clinical training conflates — the etiological, the theodicy, and the endurance question — and argue that the patient's account answers the third while clinicians grade it as an answer to the first.