Interventions with a measured effect
Record what the school did for a student at risk and see, right beside it, what the dropout-risk score did between the reading taken at the start and the one taken when the window closed. When either side has no number, the screen says "not measurable" and explains why instead of showing zero. Part of the Education vertical, optional and off until an admin turns it on.
The Education screen gains the Interventions and measured effect block, right below the dropout-risk block. It exists for a simple reason: calling the family, booking tutoring and reviewing financial aid turn into routine nobody can evaluate when there are no two readings of the same number beside them. An intervention with no measured effect becomes a ritual.
How it works: two snapshots of the risk
When you record an intervention, the system immediately freezes that student's current dropout-risk reading: that is the before. Alongside it you declare the measurement window in days, which is the promise to look again. When the window closes, the system takes the second reading on its own: that is the after. The Observed change column shows the difference between the two.
Neither snapshot is a loose number: each one points at the risk-history row that backs it, with the model version, the weights and the inputs of that day. That is what lets you answer, months later, where each end of the comparison came from.
The change is observed, never causal
What the screen shows is what the number did in that period, not what the intervention caused. Between one reading and the other there were classes, assessments, bills, holidays and life. That is why the panel never says "tutoring cut the risk by 15 points": it says the risk was observed lower, and it keeps the caveat in sight the whole time. Use it as a signal to look closer, not as proof.
Not measurable never becomes zero
If the reading at the start or the one at the end has no number, there is nothing to compare. The risk score has no number when none of the six dimensions had data, and in that case the screen shows "not measurable" with the reason spelled out: there was no reading at the start, the reading had no dimension at all, no new reading has been taken yet. Zero never appears, and "no effect" never appears: we could not measure and we measured and nothing changed are different statements, and treating them as the same would be fabricating a result.
The rule also holds in the database: a change can only exist when the comparison was actually observed. There is no way to store a disguised zero, not even through a technical path.
Different models do not compare
If the risk model version changed between the two readings, the screen shows "invalid comparison" and reports no change at all. Two different rulers do not subtract, and the result would look like a measured number without being one.
The by-type summary, which is what coordination really asks
The by-type table answers "does tutoring help?" in the only way you can defend in a meeting: how many times it was done, how many were comparable, what the average observed change was among those, and how many were left out. The not-measurable ones and the ones with a changed model are counted separately and never enter the average. A type with nothing comparable shows "no average", because an average of zero items is not zero.
Running the intervention
Execution reuses the playbooks the system already has: running a playbook on the intervention materialises its steps as tasks with due dates. Because the Education vertical is a separate module, the tasks are created on the record you pick as the anchor, usually the student's or the family's contact. With no anchor the intervention is still recordable and you move it along by hand; what does not exist is make-believe execution.