| name | pra-learning-curves |
| description | Fit sigmoidal (S-curve) learning curves and forecast future values with the PRA package. Use when a user has historical time/completion (or unit/cost) data showing slow start, rapid acceleration, and plateau, and wants a fitted model plus predictions with confidence intervals.
|
When to use
The user has historical x (time or unit number) and y (completion, cost,
etc.) data and wants to model the learning curve and project it forward.
MCP tool
fit_and_predict_sigmoidal_tool(x_json, y_json, model_type, predict_x_json, conf_level) fits the model with nonlinear least squares
(Levenberg–Marquardt) and returns fitted coefficients plus predictions.
x_json: JSON array of x values, e.g. [1,2,3,4,5,6,7,8,9,10].y_json: JSON array of y values, e.g. [5,15,40,60,70,75,80,85,90,95].model_type: "pearl", "gompertz", or "logistic".predict_x_json: x values to forecast at; omit to predict at the original x.conf_level: e.g. 0.95 for 95% prediction intervals; omit for none.
Choosing a model
- Pearl: symmetric S-curve; growth and saturation balanced.
- Gompertz: asymmetric — slow initial growth, rapid mid-phase acceleration.
- Logistic: growth from a known initial value.
Fit all three and compare residuals to select the best fit; use confidence
intervals to gauge prediction uncertainty (bands widen as extrapolation extends
beyond the observed range).
References
- ICEAA Learning Curve Analysis Guide.
- DAU Learning Curve Workshop materials.
