TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · Judgment · AGE 18—21 · POSITIONAL · HELD
Knowing when to stop
Iteration without a stopping rule is gambling with better manners.
The dilemma
Cycle six gave you the image that answers the brief. Cycle seven might give you a slightly better one. So might eight. It is cheap to find out. When do you stop?
The choices
Stop when you run out of time or patience. Or set the stopping rule before you begin and keep to it: stop when changing the brief no longer changes the output, or stop when the output answers the question you actually asked. Either condition ends the run.
The consequence
Without a rule, iteration quietly turns back into the lottery it was meant to replace. You are no longer testing a brief. You are hoping for a better draw. With a rule, the run ends on a decision rather than on exhaustion, and the last cycle is the last useful cycle, which is something you can say out loud to a juror.
The case
A student is iterating a façade study for a housing block in Hubballi. Cycle five answered the question: how deep must the balcony be to shade the west wall at 4 p.m. in May? Cycles six to fourteen were "just seeing." By cycle fourteen the balcony depth had drifted back to a number that no longer shaded anything, because each cycle was chasing the previous image rather than the question. The tutor asked where the question had been answered. Cycle five. Nine cycles had bought images and lost a decision.
Our position
Iteration without a stopping rule is gambling with better manners. Stop when changing the brief no longer changes the output, or when the output answers the question you actually asked. Going past that point buys images, not decisions.
Why we hold it
Every cycle costs attention, and attention is the scarce resource in a studio, not images. The Lab's iteration log exists to make decisions visible. A run that continues past its last decision fills the log with noise and teaches the student that more is the same as better. It is not, and the jury can tell.
The strongest objection
Discovery does not keep to rules. Some of the best moves in a project arrive in the cycle after the one that "answered the question," because the answer revealed a better question. A stopping rule set in advance can close the run at exactly the moment it was about to become interesting. Designers who stop when told to may be disciplined and dull.
What would make us revise it
The rule allows a new run with a new question. It forbids drifting on the old one. We would revise the card if logs showed that the cycles after the stopping point were reliably where the good ideas came from, which would mean drift is where design happens. In the Lab's workshops so far, drift is where balconies stop shading.
Try it
Set the stopping point before an iteration run: the question to answer, or the convergence you will accept. Stop when you reach it, even if you are curious. Then, for a finished set, explain why the last useful cycle was the last one.
Take it to crit
Ask the student why they stopped where they did. Could they have stopped three cycles earlier? Ask them to name the cycle where the question was answered, and what the cycles after it were for.
How it works
Two signals tell you a run is done. The first is convergence: you change the brief and the output barely moves, which means the brief has stopped carrying new information and you are now sampling noise. The second is that the question is answered: the image now lets you make the decision you started the run to make. Either signal is enough. Both are easier to see if you wrote the question down before cycle one, which is why the stopping rule and the rubric live on the same page of the log.
What this idea builds on
What this idea opens up
- Nothing yet names this as a foundation.
Sources
- disciplined-iteration-log
- judgment_register
- The AI Design Director Protocol
Open this idea on the map · The complete map · Logika · RBDS AI Lab, India · revised every edition.