The False Pattern

Trviksha spent three more days with the patient records. She had established that living near water and being over fifty were both associated with sickness. Now she was looking for more.

The Clinic Pattern

She examined a factor she had initially overlooked: prior clinic visits. How often had each patient visited Grothvik's clinic in the year before the rainy season?

The result was striking. Among the patients who got sick, eighty-four percent had visited the clinic at least once in the prior year. Among the healthy patients, only thirty percent had.

Trviksha double-checked. The pattern held. Prior clinic visits were a stronger predictor of rainy-season sickness than any other single factor — stronger than water proximity, stronger than age, stronger than diet.

Trviksha: Prior clinic visits are the strongest signal in the data. Patients who visited your clinic last year were nearly three times more likely to get sick this rainy season.

Grothvik: That is interesting. And wrong.

Trviksha: The numbers are right. I checked them twice.

Grothvik: The numbers are correct. The conclusion you are about to draw is wrong.

Trviksha: I have not drawn a conclusion yet.

Grothvik: You were about to say that visiting the clinic somehow causes patients to get sick. Or that the clinic itself is a risk factor.

The Arrow

Trviksha paused. That was, in fact, exactly what the data seemed to suggest. Patients who went to the clinic got sicker. Reduce clinic visits, reduce sickness?

Grothvik: Think about who visits a clinic. Healthy people do not walk in for the pleasure of my company. Sick people visit — people who are already unwell, already fragile, already vulnerable. They visit the clinic because they are in poor health. Then, when the rainy season arrives and the next illness sweeps through, the same fragile people get sick again. The clinic visit did not cause the sickness. The underlying frailty caused both the clinic visit and the sickness.

Trviksha: The data shows that these two things appear together. It does not show which one caused the other.

Grothvik: Correct. And in this case, neither caused the other directly. A third factor — the patient's general health — caused both. The clinic visit is a symptom of fragility, not a cause of future sickness.

Trviksha stared at the records. The pattern was real. Clinic visits and rainy-season sickness genuinely did co-occur. But the story the pattern told — "clinics cause sickness" — was completely false. The co-occurrence had a different explanation entirely.

Trviksha: How do I know which patterns in the data are real causes and which are... this?

Grothvik: From data alone? Often you cannot. I know that clinic visits do not cause sickness because I understand how sickness works. The data does not understand anything. It shows what happened alongside what. Nothing more.

The Distinction

Trviksha: Then what good is the data?

Grothvik: It is good for prediction, even when it does not explain causation.

This stopped Trviksha short.

Grothvik: If I know that patients who visited the clinic last year are more likely to get sick this rainy season, I can prepare for them — regardless of whether the clinic visit caused the sickness. I do not need to know why. I need to know who. Prediction does not require understanding. It requires patterns that hold.

Trviksha: But the clinic visit pattern holds for the wrong reason.

Grothvik: It holds for a reason you do not fully understand. But it holds. If a patient visited my clinic three times last year, she is more likely to fall ill this season, whether or not the clinic visits themselves are the cause. The prediction is still useful.

Trviksha sat with this for a long time. The distinction — between explaining why something happened and predicting that it will happen — was new to her. Every instruction tablet she had written in GlagalCloud's history was explanatory. "If the herd count drops by more than ten, send a pterodactyl" was a rule built on understanding what a count drop meant. Grothvik was asking for something different: a system that identified who was at risk, based on patterns in data, without necessarily understanding why those patterns existed.

Trviksha: So I should use the clinic visit pattern for prediction, but I should not tell anyone that clinics cause sickness.

Grothvik: Correct. Use the pattern. Do not mistake the pattern for an explanation.

What Next

Trviksha returned to the full set of associations: water proximity, age, diet, prior illness, clinic visits, household size, occupation. Each factor, alone, showed some degree of association with rainy-season sickness. Some associations were causal (living near stagnant water probably did contribute to disease). Some were not (clinic visits were a marker, not a cause). Some she could not tell.

But Grothvik did not need Trviksha to sort out causation. She needed a system that could take a new patient's data — age, location, diet, all seven factors — and produce a prediction: high risk or low risk. If the prediction was accurate enough, it was useful regardless of whether Trviksha understood the causal mechanism behind every factor.

Trviksha: I need to build something that takes all seven factors for a patient and combines them into a single prediction. Not by rules I write — by patterns the data reveals.

Glagalbagal: You are going to teach the pebbles to predict?

Trviksha: I am going to build a system that finds the patterns for me. I know they are in the data. I just cannot extract them by hand.

Glagalbagal: In my day, pebbles counted sheep.

Trviksha: In your day, pebbles did a lot of things. They are about to do one more.