I had a goal and no data. Now what?

Kay and Jo have a target for a room but no measurements, so they write the target down as an objective and let Konjugate search. The strip is below, and the steps to repeat it are under it.

Transcript

Four panels. Kay, a round teal character, and Jo, a dry blue one, are joined by a dotted line. Step 1, a goal. A card reads: the room near 60 for 40 seconds, no measurements. Kay says: I want the room near 60. I have no data. Jo says: Then write the goal down. Step 2, optimise an objective. A card reads: minimise the integral of the squared difference between the room and 60, by changing the ambient gain. Kay asks: No CSV at all? Jo says: A number to make small. You pick what may change. Step 3, the search. A chart compares the room before and after tuning over 40 seconds. Kay asks: Did it help? Jo says: Closer to 60, on average. Step 4, the result. A card reads: the objective went from 11480 to 9824, 14 percent lower, with the ambient gain going from 0 to -0.52. Kay says: But it ends above 60. Jo says: One constant cannot do it all.

Image for sharing

What just happened

A hot plate at 90 warms a room that starts at 20. The goal was a room that stays near 60 for 40 seconds, and there were no measurements to fit, so the goal was written as an objective: the total over the run of the squared distance from 60. The one thing the search was allowed to change was the room's ambient gain, anywhere from -1 to 1.

Konjugate ran the 40-second simulation 17 times with different gains and kept the best. It chose a gain of -0.52, which lowered the objective from 11480 to 9824, 14% lower. The room still ends above 60, at 71.51 instead of 80.53. A single constant cannot lift the room quickly at the start and also stop it overshooting later, so the search balances the two.

Repeat it in Konjugate

1. Build the hot plate and room air model from "Getting started", with the cooling rate at 0.05 and an Ambient gain source term on Room air. Mark the gain Tunable with bounds of -1 and 1.
2. Open Parameter tuning and choose Optimise an objective.

3. Write the objective with the ∫ over the run button and the Room air · Temperature chip, as the integral of the squared difference from 60, and set each simulation to run for 40 seconds.

4. Press Run fit and watch the best score fall.

5. Review the result, then apply the value.

What this does and does not say

The model is invented, and the figures were read from the run described above on 2026-11-02. An objective gets exactly what it asks for and nothing else, so the search found the best single constant for this score, not a room that holds 60. Writing a better objective, or adding a limit that keeps a number within a bound, is how to ask for more.

Go deeper

• "Tuning a parameter against measured data" covers both goals, the optimiser choices and the review panel.