Writing is learned by revising your own thinking — not by receiving a better version of it. Everything below is the short version of why Prove is built the way it is, and the research it leans on.
Why a loop, not draft-and-submit
Nancy Sommers found that novice writers treat revision as proofreading — swapping words, fixing grammar — while experienced writers revise globally, re-seeing the whole argument and reshaping it. The gap is not vocabulary; it is the willingness to change one's mind on the page. Prove enforces that move by sequencing global concerns (claim, structure, evidence) before local ones (sentence, word).
John Hayes and Linda Flower modeled composition as recursive — planning, translating, and reviewing cycle into each other rather than running in a line. The loop is a deliberate scaffold for that recursion: each stage feeds the next, and you can re-enter at any point without losing the work you already did.
Why five stages
Map is problem representation — labeling what each paragraph is trying to do before judging whether it works. Diagnose is the knowledge-transforming move Carl Bereiter and Marlene Scardamalia identify as the single biggest difference between expert and novice writers: experts use writing to interrogate their own ideas, while novices use it to dump what they already know.
Plan, Revise, and Reflect map onto the cluster of strategies the Carnegie Writing Next meta-analysis identifies as having the largest evidence base: explicit strategy instruction, summarization, and the study of models. Diagnose comes before Plan on purpose — you cannot fix a draft you have not yet honestly described.
Why feedback, not generation
John Hattie and Helen Timperley frame effective feedback around three questions: Where am I going? How am I going? Where to next? Every coach turn in Prove is shaped by that spine. Paul Black and Dylan Wiliam's earlier synthesis showed formative assessment to be one of the highest-leverage interventions available in a classroom— provided the feedback is specific, timely, and addressed to the work rather than the writer.
Why the model refuses to write the sentence
Robert and Elizabeth Bjork's principle of desirable difficulties is the cleanest statement of the cost of help: the easier a tool makes the task, the less learning sticks. Effortful retrieval and generation produce durable knowledge; fluent reception produces the illusion of it.
Evan Risko and Sam Gilbert's work on cognitive offloading sharpens this for the AI era: when a tool reliably does a cognitive task for you, your own internal representation of that task degrades. A model that writes your paragraph trains you to need a model that writes your paragraph. So Prove will not. The coach reads, names, questions, and waits. The sentence is yours.
That is the loop. Five stages, one draft at a time, and a model that holds the line on not doing the writing for you.