Ideate

---
name: ideate
description: Fan wide takes on ONE thing — an offer, a Mechanism, a big idea, a hero, an angle, a name — then refute, iterate, and return 3–5 survivors that stood criticism for the owner's pick. The shared idea process every option set runs. TRIGGER when one decision point wants a set of real options instead of a single guess. DO NOT TRIGGER to write a full piece (the write skills) or to grade a draft (the check skills).
---

# Ideate

One Process: turn one thing into a set of distinct options that survived criticism, so the owner picks against real, tested ideas. Picking beats correcting — and a survivor that stood an adversarial review beats a fresh guess. draft-offers, mechanism, and big-idea all run this Process on their one thing; it lives here once.

Owner law (verbatim): "I don't want ONE agent to just shit 4 bad ideas; I want a few ideas that went through iterations, stood criticism, and make sense... I should be PICKING and using my taste & copywriting eyes to help, not doing basic fucking analysis of ideas; we need processes that include adverserial reviews".

## 1. Fix the thing and the count

Name the one thing in play — offer, Mechanism, big idea, hero, section structure, angle, name, headline, layout. The owner sees 3–5 survivors; brainstorm far wider than that so the review has something to cut.

## 2. Brainstorm wide across assigned lens cells

Generate many more options than will survive. Each must differ at its foundation — a different angle, a different structure, a different core idea. Options that share a skeleton and differ only in wording are one option, not several. Every option draws on the piece's business facts and graded research records, never invention.

### Assign each option a lens cell from the roster, by procedure
Span the set across the Lens Roster below, assigned by procedure — walk the axes and deal cells out — never by the orchestrator's taste. An unassigned brainstorm collapses to one calm consultative voice restated; measured: ten baseline angles shared one personality and one cadence, while the same evidence under assigned cells produced six unmistakably different pitches. The roster is a living list — extend it through record-decision, never trim it silently.

Lens Roster:
    Voice (who talks): showman · forensic auditor · peer confessor · contrarian insider · drill sergeant · storyteller · professor · skeptical friend
    Register: formal · conversational · over-the-top
    Structure: problem-agitate-solve · AIDA · story arc · fact cascade · us-vs-them · before/after bridge · open-loop · celebration
    Drive: fear · anger · curiosity · identification · proof · relief/humor · status · hope
    Frame origin: problem-driven · enemy-driven · dream-driven · evidence-driven · contrarian
    Offer shape (offers only): guarantee-led · done-with-you · unbundled wedge · named-system flagship · pay-on-proof · anti-offer · bundle/stack · trial-first

### Dispatch cells fresh-context, sibling-blind, committed hard
One fresh-context generator per cell or small cell group, each dispatch carrying its cell, the subject, and the evidence paths — no sibling cell's output, no meta context. The dispatch orders full commitment to the cell: no hedging toward the middle. A generator that sees a sibling's take regresses to it.

### Anchor each option on a different named speaker
Every option names its anchor — the one speaker story or record it is built on — and no two options share one. Distinct lenses herd onto the same best quote otherwise; measured: six committed voices all leaned on the same two anchors until the quota forced the spread.

### Carry each option's ratings, filter nothing by threshold
Every option carries the ratings of the problem, buyer, and offer shape it rests on (see grading) and their existence ratings from the reality attack (see check-reality), so the owner picks with them visible. Nothing is removed for a low rating — a low-rated option rides to the pick alongside the rest, its rating and reasoning shown.

### Generate from reactions to the problem
An angle is an interesting way to talk about the problem, and the space to draw from is every possible reaction to the problem — humor, fear, fun, entertainment, jokes, weirdness, and everything in between. That space is infinite; span distinct reaction kinds across the set.
Never: 3–4 ultra-predictable mid angles that are just the sexiest way to talk about the features — a feature-derived angle fails before the set.

### Fan one thing, never the strategy's blocks
ideate fans ONE thing — offer detail, Mechanism, big idea, hero, angle, name. Inside a picked strategy, the problem, buyer, offer shape, and positioning are FIXED by that strategy and never vary here; varying THOSE — recombining whole directions — is the strategy layer, which lives in plan-copy, not this Process. The job is many distinct takes on the one thing, not many problems or audiences.

### Fail a collapsed set by rule before judging quality
A set whose options collapse to one take of the thing reworded FAILS BY RULE — before anyone judges its quality. Voices collapse too: options whose assigned voices read interchangeably — one clipped rhythm across four cells — are collapsed the same as ideas, and measured: a first pass merged four of six voices until per-voice corrections separated them. A failed set reruns wider, or the collapsed options rerun with per-cell corrections. Record the failure so it can be traced back into system fixes; a collapse that keeps recurring is a system fault, not a one-off rerun.

### Treat an owner-given angle as one option
An angle, offer, or idea the owner hands in is ONE option among the competing ones — brainstormed against, refuted, and iterated, sometimes changed fundamentally, like any other. It is never the settled direction and never skips the set. An owner input seeds the set; it does not pre-empt the owner's pick against it.

## 3. Refute in fresh adversarial context

Dispatch fresh-context adversarial reviewers to refute each option — a reviewer that did not generate the ideas and carries no attachment to them. An option that is trivially refuted dies here, before the owner ever sees it. This is the gate that stops four bad ideas reaching him.

### Assign refuter frames from the fixed list
The refuter frames are a FIXED list, assigned by procedure, never an orchestrator's pick: economics (buying power, already-paying, willingness to spend) and competition (competitor density, is this claim already worn). Argument fit is NOT an ideate frame — whether the option's argument fits the problem and buyer belongs to the buyer-reviewer at the check-strategy gate. Existence and sensibility are NOT ideate's frames either — whether the buyer, problem, and language exist outside the document, and whether a claim survives normal human behavior, belong to the reality-reviewer alone, ruled on the landscape before ideation runs and never re-litigated here. Assign the options across both frames so criticism arrives from fixed angles, not whatever the dispatcher happened to think of.

### Dispatch each refuter on a named input contract
The dispatch names exactly what the refuter reads and nothing else: the problem (`Problems.md`), the buyer hypotheses (`Buyers.md`), and the relevant research records under `research/`. A refuter loaded with more than its contract refutes from noise. Every refutation cites what it refutes against — the Problems.md row, the buyer hypothesis, or the research-record ID it stands on — so the refutation itself runs on facts, never a taste assertion.

## 4. Iterate the survivors against the criticism

Take each option that survived and sharpen it against the specific criticism it drew. A survivor reaches the owner having answered its strongest objection, not as a first draft.

### Re-sharpen each survivor into its assigned cell
Stacked constraints spread the ideas but dilute the intensity — a showman constrained three ways barely barks. The iterate pass pushes each survivor back to full commitment in its assigned voice and register; spread comes from the brainstorm's constraints, intensity comes from here.

## 5. Return 3–5 survivors as the set

Present the survivors side by side, each labeled by the distinct idea it plays and what criticism it answered, so the owner reacts to concrete, tested things and picks. Every survivor and every combination carries its grading per the grading skill (see grading), so the owner compares on numbers and evidence, never on an agent's disposition. Do not recommend one and bury the rest — the set is the deliverable, none pre-chosen.

## 6. File the set as the option file

### Write the set and the pick to options/
Write the surviving set to the project's `options/` folder as this thing's option file — one file per thing (offer, Mechanism, big idea each its own). When the owner picks, mark his pick in that same file. This option file is the home check-strategy, record-decision, cold-read, and plan-copy cite for the pick and its rationale.

Verification: options brainstormed wider than the returned set; refuted in a fresh adversarial context with refuters assigned across the fixed frame list (economics, competition) and trivially-refutable ones cut; the set spans distinct takes on the one thing with the strategy's problem, buyer, offer shape, and positioning held fixed (varying those is the strategy layer in plan-copy), and a set that collapses to one take reworded fails by rule, reruns wider, and records the failure; any owner-given angle rode as one competing option, not the settled direction; survivors iterated against the criticism; each survivor and combination graded per the grading skill with its ratings riding, none filtered by threshold; 3–5 distinct survivors returned for the owner's pick with none pre-chosen; the set written to the project's `options/` folder as this thing's option file, marked with the owner's pick once he chooses.