Four pieces, read in order, tracing the same idea — Promption and Provocation as cognitive modes for studying clinical cases — from a first working demonstration, through progressively heavier ground-truth tooling, to a deliberate step back toward something small enough to use every week.
Introduces the dual-mode architecture — Promption (scaffolding messy narrative into structure) and Provocation (stress-testing the resulting hypothesis) — and walks a single DKA/pneumonia case through both modes end to end, surfacing an HbA1c discrepancy a single-pass answer would likely have missed.
Adds three structured layers on top of Part 1's cognitive modes — a chronic kidney disease knowledge graph, an acute pancreatitis institutional pathway, and appraised evidence — and examines what each one fixes, what new risk each one introduces, and why disagreement between layers is a signal worth surfacing rather than smoothing over.
Asks which layer deserves the most weight, for which case: a three-tier rarity framework crossed against an EBM task-complexity axis, a verification burden that rises with both, and an honest accounting of what remains a proof of concept versus what would need a matched case series to validate — plus a category of uncertainty, input curation, that no layer can ever reach.
The deliberate turn. Rather than adding a fourth ground-truth layer, this piece asks what the smallest usable version of the whole stack looks like — a ten-minute weekly loop, a one-line ledger instead of a formal study, and a simple gut-check for when it's actually worth reaching for the graph, the pathway, or the evidence layer described in Parts 2 and 3.