Skip to content

Latest commit

 

History

History
28 lines (23 loc) · 2.1 KB

File metadata and controls

28 lines (23 loc) · 2.1 KB

Metric definitions

The dashboard separates participation, dosage, outcome completeness, outcomes, and implementation conditions. These measures answer different questions and should not be interpreted interchangeably.

Metric Definition Grain Target Caveat
Eligible students Distinct students in the synthetic enrollment dimension Student Denominator is fixed at the generated cohort
60-day activation rate Share of students with any activity in September or October 2025 Student 75% Activation does not imply sustained use
Dosage target rate Share of student-months with at least 60 minutes of use Student-month 60% A month below 60 minutes is not necessarily a program failure
Follow-up completion rate Share of students with a valid follow-up assessment Student 80% Missing outcomes may be systematically different
Average score change Mean follow-up minus baseline score among students with both assessments Student Positive Descriptive; no untreated counterfactual is present
Support SLA rate Share of support tickets with an initial response within 24 hours Ticket 80% Ticket complexity is not risk-adjusted
Training rate Staff trained divided by staff expected School 80% Does not measure training quality
Fidelity score Synthetic implementation rubric scored from 0–100 School 70 Intended for prioritization, not personnel evaluation
Data freshness Days since the school data feed last refreshed School ≤14 days Stale data can distort every downstream metric
Risk score Count of six thresholds missed: activation, dosage, follow-up, support SLA, fidelity, freshness School <3 Equal weights make the logic transparent but are not empirically optimized

Risk bands

  • Critical: four or more thresholds missed
  • High: three thresholds missed
  • Watch: two thresholds missed
  • Stable: zero or one threshold missed

The risk score is an operational triage rule. It is not a predictive model and should be combined with local context before resources are reassigned.