Quest Ladder
TeachFORGE.ai
Quest Ladder / AI build
AI build: one pass from setup to learning
Complete the setup, press once, and get a full draft you then read and correct.
AI supplies the heat. You do the forging.
How it works
Full AI build asks you to complete Setup — the details and the shared settings — and then generates the whole type in one pass: the changing values, the content, and the learning blocks.
The order is deliberate. Setup is the brief: the question format, the description, the subject, the age group and the marks are what the model is writing against, and a thin setup produces a thin type.
Spend the time on the description. It is the single biggest determinant of what comes back, and a paragraph that says what the question should test and what a wrong answer usually means is worth more than any amount of re-running.
Binding shared settings
The panel shows a Binding shared settings snapshot: the scoring, timing and answer settings the generation must respect. They are not suggestions to the model, they are constraints, which is why they are settled first.
Your saved exit routing is also protected. If you have already designed one, the panel says it was kept and the AI's routing was not applied — with a link to edit either.
Running it
Generate prompt compiles the prompt and shows it, with its size in tokens. You can read and edit it before anything runs. Then choose how to run it:
| Mode | What it is |
|---|---|
| Copy prompt | Free on every plan. Run it in any chat model and paste the JSON reply back. |
| My GPU | One click to the endpoint you configured under Profile → My AI Endpoints. |
| My API key | One click using your own provider key, billed by them. |
The panel states the thing worth knowing about all three: the same compiled prompt and the same task defaults are used, and only the model and the billing differ. If one mode produces better results than another, that is the model, not the platform.
A run survives navigation — you can move around while it works — but closing the tab ends it, and your own GPU needs the tab kept open.
Exemplars
An exemplar is a finished type you pin as a worked example, so the model copies the structure without copying the subject matter. It is the most effective single thing you can do to raise the quality of a generated type.
Pick one you are genuinely happy with, from the same format and roughly the same age group. The tenth type in a family generated against the first is usually close to publishable.
Keep and improve
Open a type that already has content and this step changes: it offers Keep and improve alongside Rebuild from scratch. Read which is selected before running, because the second discards what is there.
Afterwards, Review AI draft walks what came back, and quality advisories flag the parts the platform is least confident in. Then read the whole thing in teacher review — it is the only view that puts the hint and the solution side by side, which is where generated types most often contradict themselves.