The rule a teacher can explain
Explore this deployment in the PAN Lab ↗
A large urban school district operationalized a transparent ninth-grade indicator - course credits earned plus no more than one core-course failure - from consortium research showing it predicts high-school graduation with about 85 percent accuracy, and wired it to school-level attention rather than to an opaque score. District graduation rates subsequently rose to record highs. The indicator is a rule anyone can read: a teacher can explain to a student exactly why they are off-track and exactly what would change it, so the contest-and-correction loop that opaque early-warning deployments sever is open by construction.[†]
What happened
A large urban school district turned a research finding into district practice. The finding, from a university consortium's study of the district's own students, was that a simple ninth-grade condition - earning enough course credits and failing no more than one core course - predicts high-school graduation with about 85 percent accuracy. Not a model, not a score: a rule, computable from a transcript, readable by anyone.
The district operationalized the rule as its on-track indicator and wired it to school-level attention. Teachers and counselors see which ninth graders are off-track while the year can still be changed, and the response is theirs: a conversation, a schedule change, a credit-recovery plan. In the years after the district committed to the practice, its graduation rate rose to record highs.
What makes this case the education domain's counter-case is everything the rule is not. It cannot be opaque: a teacher can tell a student exactly why the flag is on - you are short credits, you have two core failures - and exactly what would turn it off. That means the person the flag is about can contest it, correct it, and act on it, which is precisely the loop the domain's opaque deployments sever. It has no learned bias to audit, because it learned nothing: it reads course outcomes recorded for purposes far older than itself. And it points at something changeable. The consortium's central insight was that freshman-year course performance - a condition schools can act on - predicts graduation better than the fixed characteristics students arrive with. The rule's power is that it names a lever, not a label.
The documented limits deserve equal weight. The accuracy figure and the graduation rise are associational: no randomized trial assigned schools to use the indicator, and a district that adopts it is usually doing other things too. And the mechanism runs through the intervention, not the flag - an off-track list nobody staffs is a list, exactly as the domain's failed deployments show. The district's practice resourced the attention, and that resourcing, not the arithmetic, is where the benefit lives.
The honest reading is that the best-documented early-warning success in this domain used the least technology: a transparent rule, pointed at a changeable condition, wired to resourced human attention - with its accuracy published, its logic contestable by a fifteen-year-old, and its results measured at district scale.
The sociotechnical reading
This case is the education domain's transparency portrait, and the map reads it against the domain's opaque anchor deliberately: the same task - flag ninth graders at risk - approached with a readable rule instead of an ensemble (several models combined into one score) model, and ending in record graduation rates instead of a quiet retirement. The instruction it carries is that opacity was never the price of prediction here. A rule at 85 percent documented accuracy, explainable in one sentence, outperformed in practice a model whose own agency could not say what it had learned - because the deployment around the rule could do things the deployment around the model could not.
The load-bearing property is contestability by construction. When the flag is a readable condition, the person it is about can check it, dispute it, and change it - the correction loop the domain's ML deployments sever is open by default, and no disclosure policy or appeal process has to be built, because the explanation IS the flag. The map draws that as checks that run at baseline rather than latent ones a player must open: the indicator's accuracy is published and re-studied by an external consortium, and the staff who act on the flag can verify it by reading the same transcript it reads.
The second lesson is where the benefit lives. The rule names a changeable condition - freshman course performance - and the district resourced the response to it. An off-track list nobody staffs is exactly as useless as an opaque score nobody trusts; the domain's failed deployments prove both halves. The map keeps the readiness lesson attached: the flag's value is entirely downstream of the attention it triggers, so the governable object is the intervention's resourcing, and the indicator is only the pointer.
The Lab network models only the deploying district: its rule, its transcript record, its school staff, and the consortium's standing evaluation over it. No student outcome is computed on any diagram. The accuracy figure and the graduation rise live in this case file as the associational, district-scale evidence they are - never as a computed result, and never as proof the indicator alone caused the rise.
The concepts used in this reading are defined in the Field Guide; the governance responses live in the Practice Library. The model organization for this case can be stress-tested in the PAN Lab.