Scenario Forge

Name any workplace topic below and the Forge writes a playable branching scene. The catch, and the point: it has to follow my four scenario rules, printed on the right — consequence-based feedback, no obvious right answer, wrong options drawn from real instincts. On its own, a generator writes generic scenes quickly. The rules are where six years of scenario work went. That part does not come with the subscription.

Input · Any topic, any audience

Output · A playable scene, my rules

Point · The designer designs the system

SCENARIO_FORGE v1 {{ statusLabel }}

The performance problem

{{ busyLabel }}

{{ sceneKicker }}

The design rules it must obey

This is the actual instruction set the Forge runs on — four testable rules, distilled from six years of writing scenarios.

— a specific moment, not a topic. One scene, one decision, real names.
— no obvious right answer. Every option must be something a smart person would actually do.
— feedback names consequences, never rules. "Incorrect" is banned.
— the risky option is the kind one. The design bet of every scenario on this site.

// In production this sits inside a review workflow: the Forge drafts, the designer edits, the SME verifies. Speed changes; accountability doesn't. And because the rules regenerate a scene in seconds, content stops being precious — when the policy changes, you re-forge, you don't re-project-plan.

WHERE THE LINE IS — WHAT I TRUST THE MODEL WITH, TODAY (I RE-DRAW THIS AS MODELS IMPROVE)

TRUST IT WITH

Drafting scenes inside explicit rules — it's fast and tireless, and drafts are cheap to reject

Playing a counterpart that improvises — better than any script I could branch by hand

Scoring transcripts against a rubric of observable behaviors — with the evidence quoted back

Explaining a wrong answer in context, when the pedagogy of the explanation is specified

STILL HUMAN

Choosing which moment to build the scene around — the analysis that makes it worth practicing

Verifying facts with the SME — the model is confident about things that are wrong

The feedback voice — drafts drift toward "Great try!" the second you stop editing

Shipping. If it goes in front of a learner, it has my name on it

Live generation isn't available in this copy of the site, so the Forge is showing a pre-built example run. The design rules and the pipeline are the point either way.

Designer's note

Why I made these choices

The rules are the deliverable

A design system used to be a style guide; now it also has to work as an instruction set a model can follow. Writing those four rules — testing them against dozens of generated scenes, tightening the ones the model wriggled around — was the design work here. Editing what comes out still is.

Fast drafts, same bar

A generated draft arrives in seconds, and most read like it. Nothing about the bar has changed: the scene still has to be built around the right moment, the wrong answers still have to come from real instincts, the feedback still has to be worth reading. The Forge is fast because that thinking happened ahead of time, in the rules.

Accountability stays human

A generated scene ships only after a designer edits it and an SME verifies it. AI shortens the draft from days to seconds; it does not shorten the review. Someone's name is on the work before a learner ever sees it, and I want it to be mine.