A week later, a school principal used one of the new explanation summaries during a meeting with a student named Tomas.
The summary stated that declining grades, absences, and repeated transfers had contributed to his elevated risk score.
The principal interpreted the list as a diagnosis of why Tomas was struggling.
Tomas disagreed.
His family had moved twice because their landlord had ended the leases, and his absences followed a period when he travelled across the city to care for a younger brother.
His grades had fallen after the transfers, not before them.
The explanation accurately described predictive associations but implied a causal story the system had never established.
Priya called the incident an example of narrative overreach.
Humans are unusually skilled at converting lists of correlated factors into coherent stories.
Coherence can feel like understanding even when the causal structure is unknown.
Elias reviewed the wording and realized that phrases such as 'your score is high because of' invited stronger interpretations than the evidence supported.
The team replaced causal-sounding language with more precise descriptions such as 'these factors were associated with the model's prediction.'
That wording felt less natural and less satisfying.
It was also harder to misuse.
Tomas asked another question the team had underestimated: what could he do to change the score?
The model had not been designed as a personal recommendation engine.
Changing one visible feature might not lower risk if other correlated features remained, and artificially optimizing the score could distract from the student's actual needs.
Priya warned against turning the model into a game in which students learned to perform data that looked safe.
The goal was support, not score management.
The incident led the team to separate explanation from recourse.
Explanation concerns how a decision was produced.
Recourse concerns what meaningful actions an affected person can take in response.
A system can be explainable yet offer no realistic recourse.
The team added human case review whenever a student or family challenged a score.
They also prohibited staff from presenting model features as personal diagnoses unless independent evidence supported that interpretation.
The distinction between explanation and recourse reframed transparency from passive disclosure into actionable intelligibility.
Information becomes ethically meaningful only when affected people can use it to correct error, supply context, or contest an inference that carries consequences.
Tomas's case became a reminder that technically accurate information can still mislead when its pragmatic meaning is misunderstood.