TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · Judgment · AGE 17—19 · POSITIONAL · METHOD
Define done before generating
Good enough is a stopping threshold written before the first output arrives.
When to use
Before any run longer than one shot: a render batch, an options study, a long text draft, an agent task. Before you open the tool, not after the first result.
The method
Write four lines on paper or in a note you can see. One: what the output must do for the project. Two: what must not move, the fixed things. Three: what would make you reject an output on sight. Four: how many you will make before you stop and judge. Then run. Hold every output against the four lines and nothing else. Mark pass or fail. Only then look at what you like.
Watch for this
Writing the four lines after seeing the first output. The first output then becomes the definition of done, and everything after is judged by its resemblance to it.
The Lab's note
Defining done first is the Lab's recommended method, not the only one. RUBRIC FIRST, on this strand, is the criteria side of the same habit: must do, must hold, must avoid. This card adds the fourth line, the count — how many before you stop and judge — because "good enough" is a stopping threshold, what engineers call satisficing, and the threshold is the part the criteria alone do not set.
Try it
Tomorrow's generation task: write the four lines first. After the run, count how many outputs you would have accepted without them. That number is the cost of not writing them.
Prove it
Show one run where the four lines rejected an output you liked, and say what the lines caught that your eye missed.
How it works
Elsewhere on this map you give criteria to the model, inside the brief, so that it can test its own work. This card gives the criteria to you, and fixes the timing. The two are different moves. A criterion written into the prompt asks the model to check; it does not make the check happen — a written instruction asks, the system's own controls enforce, and the control here is on your side of the screen. A model reading your criteria still returns what is plausible. A person reading them before the flood still has a standard when the flood arrives. "Good enough" is a real engineering term, not a shrug. It means the point at which further generation buys images but not decisions. Defining it first is what makes stopping possible later.
What this idea builds on
What this idea opens up
Sources
Open this idea on the map · The complete map · Logika · RBDS AI Lab, India · revised every edition.