Positioning: the frame every draft is written inside
Identity has two halves: Positioning (who you are for, what you replace, why you win) and Brand Kit (name, look, and how each brand speaks). Together they are the aura every article and email is written inside.
A positioning framework has sections: Overview, Constituents, Target Buckets, Value Pillars, Value Proposition, Messaging Ladder, Persona, Vision, Mission, Guardrails. Each section can be drafted with AI Assist from what is already filled in above it, grounded in your website and the messaging the account already sends. Items are marked evidenced, inferred or assumed; check the assumed ones first.
Missing fields and Fill
Once a section has anything in it, every field in it that is still empty is marked Missing: on the field, on its row, as a count on the tab, and on the section's box on the Overview. A section you have not started shows nothing; it is simply not started yet.
Fill appears next to AI Assist when a section has empty fields. It writes only those fields, fitted to the rows they sit in, and changes nothing that is already written. AI Assist redrafts the whole section; Fill completes it. After either, Undo Draft takes it back, and nothing is saved until you press Save.
AI Assist also reads its own draft back before you see it. Value Pillars, Value Proposition and Messaging Ladder get three reads in one pass: through a competitor's eyes, against the section's own rejection tests, and as a reader would, with the tells that a machine wrote it taken out. Only what fails is rewritten; what holds is kept word for word. Then any field it left empty is filled, and the draft is checked for form: the ladder's word counts, the proposition's "by:" and "resulting in", two to six words on a pillar header with two or more messages under it, the "X, not Y" shape of a voice trait, one sentence for the vision without the company's name in it. A field that fails goes back to the AI once with the check named ("108 words; 97 to 103 wanted"). Whatever still fails is listed after the draft so you can fix it by hand. Your own edits are never checked this way.
Rate and Fix
Rate reads the section you are on back against its own rejection tests and the sections above it, and names what fails: a blocker, a weak spot or a note, each with the line it quotes, why it fails and what to do. Nothing is changed by the read. Tick the findings you agree with and press Fix: only the fields those findings name are rewritten, each to answer its finding, and everything else in the section stays word for word. The dialog then shows each change as before and after, and Rate again reads the result back. A finding about the whole section redrafts the section with that finding as the objection. Undo Draft takes any of it back, and nothing is saved until you press Save. The Challenge button is gone; the rater's findings are the objection now.
Compare
Compare, next to Quality test on the Positioning list, drafts the sections you choose of one positioning twice: once with prompt pack A and once with pack B, each side building on its own earlier sections. A rater scores both blind, as Left and Right, on seven criteria (clarity, resonance, differentiation, economics, consistency, voice, specificity) and says which side reads as written by a person. You then press Keep Left or Keep Right before you are told which pack wrote which. Picks and scores are kept under the run, so a doctrine change can be judged before it is published. Writing rules for B runs side B under a draft of the global writing rules, so a change to the rules every feature reads can be judged here too; side A always uses the live rules. The summary also shows Tells counted for each side: phrases people almost never write, overused words and summing-up closes, counted in code. Nothing is saved to the positioning. Each section is three AI calls, charged to the account like any draft.
One framework is the Default. Drafts, the Advisor and the Composer use it unless you pick another. An account can have several, for example one per product line.
Positioning is compiled into the prompt at the moment of use, so a change here reaches the next draft with nothing to republish.