00Guides
A prototype is an experiment. It is worth building only if it can change a decision, and that decides what you build.
A prototype teaches you something when it is a working application deployed on day one, with real authentication, data and model integration, instrumented against a decision you have agreed in advance. Anything that does not inform that decision is out of scope.
Name the riskiest assumption the prototype exists to test. The build scope is then everything that assumption needs to be exercised by a real user, and nothing else.
Most prototype sprawl comes from building the parts that are comfortable rather than the part that is uncertain. Decide the question first and the scope follows from it.
Throwaway code produces a throwaway result. If the prototype works, you want to keep building on it, so it should be typed, deployed and using real data from the start.
Production code is what makes the result something you can act on, and what carries the work forward if you continue. Because the prototype is a real deployment built search-led, it is already the start of the product and can begin being found while you decide its future.
Model output is non-deterministic. A credible prototype builds the guardrails, logs inputs and outputs, and shows the failure modes rather than demoing only the happy path.
What you are testing is how the feature behaves in real hands, including when the model is wrong. That behaviour is the result.
Define the events that answer your question and put them in before launch. Without instrumentation a prototype produces opinions and anecdotes rather than a result.
Decide, before users touch it, what result means continue and what means stop. Criteria set after the fact get rationalised to fit what you hoped for.
What stops a prototype from producing a usable answer.
The service and the answers that go with this guide.
The scoped engagement this guide describes.
Built so the product can be found once it works.
What a prototype costs and where the judgement is human.
Tell us the decision the prototype needs to inform. You get a fixed scope before anything starts.