Ideas

- [x] AUC ROC inspired by:
- https://kennis-research.shinyapps.io/ROC-Curves/
- http://www.bioestadistica.uma.es/analisis/roc1/
- http://mfviz.com/binary-predictions/
- http://www.navan.name/roc/
- [ ] Probability / Venn / conditional probability
- [ ] [Simpsons paradox](https://github.com/peterdalle/undervisning/blob/master/simpsons-paradox.r)
- [ ] Linear regression / OLS
- https://easystats.github.io/performance/
- https://easystats.github.io/performance/#comprehensive-visualization-of-model-checks
- https://setosa.io/ev/ordinary-least-squares-regression/
- Visual direction: points + fitted line + residuals / squared errors; let the interaction make the least-squares objective visible rather than adding many controls.
- [ ] Gradient descent
- Visual direction: loss surface / contour + current parameter point + step/play.
- Keep main controls minimal, ideally learning rate + step/play.
- [ ] Calibration
- Reliability curve + calibration summary; useful bridge between general ML and credit-risk apps.
- [ ] Neural-network playground (later)
- Only after more foundational pieces; avoid building it merely to mirror TensorFlow Playground.

Catalog direction from the 2026-09 audit:
- Current coverage is intentionally strong in ML / credit risk / explainability.
- Prefer filling foundational gaps (OLS, optimization, probability/calibration) before expanding into trendy topics such as transformers/LLMs.
- Keep each idea aligned with the Visual Data Lab principle: a small number of interactions should reveal the core concept.

Reference collection to revisit:
- http://www.rossmanchance.com/applets/index.html?s=03