Logistic regression
3D view & settings
Response surface across intake severity and first-month service days; other profile fields stay fixed.
L2 penalty · C = 0.1 · LBFGS · max 2,000 iterations
Applied research · interactive prototype
Treatment-gap
simulator.
Choose an invented intake profile and mark attendance during the first 30 days. The models estimate the chance of reaching a 60-day service gap during days 30–119.
Intake example
Invented profile. No client records are used.
Attendance calendar
Select service days in the first 30 days. Day 1 is intake day.
Model output
Estimated gap probability
Reaching a 60-day gap during the 90 days after day 30.
KNN
3D view & settings
Nearest synthetic training examples projected into three principal components, which retain 56.8% of training variation. Matching uses all 32 features.
k = 15 · uniform votes · Euclidean distance · standardized features
Neural model GRU
3D view & settings
Path of the first three latent GRU units across the 30-day attendance sequence; color moves from early to late.
Daily projection = 16 · hidden state = 32 · dropout = 0.1 · fusion layer = 16 · 20 training epochs
Loading synthetic model bundle…
The three models were trained on invented histories only. The 3D views help inspect model behavior; they are not explanations of real-world outcomes. The simulator runs in your browser and does not send or store scenario selections. Model differences are exploratory and are not evidence of real-world performance.