Explore the calculation.
Start with a working model. Change inputs, scan a parameter grid and keep an interactive document with the evidence inside.
Choose a model
Six bounded models. Each has documented assumptions, exact inputs and numerical exports.
Classification thresholds
Import labeled scores. Inspect the confusion matrix, ROC and precision–recall; sweep the decision threshold.
Gradient descent
Follow exact updates through a quadratic valley. Scan step size and curvature, then inspect stability.
Loan repayment
Compare extra payments and repayment timelines. Sample interest-rate assumptions and inspect every month.
RC charging
Explore an ideal RC step response. Sweep resistance and capacitance, including equilibrium.
SIR dynamics
Compare intervention assumptions in a teaching SIR model. Inspect the fixed equations and conserved population.
A positive test
Change prevalence, sensitivity and specificity. Count true and false positives, including undefined cases.
A complete study, in one file.
- Inspect the model and its assumptions.
- Explore inputs or import labeled scores.
- Study one or two parameters and local sensitivity.
- Export the interactive file, data and source fingerprints.
node bin/counterplot.js init classification --out experiment.json node bin/counterplot.js build experiment.json --out experiment.html node bin/counterplot.js sweep experiment.json --config study.json --out study-results.json
Works with your coding agent through the included agent skill. The numerical engine calculates results from a fixed model; titles and explanations are editable.
Models are bounded teaching and diagnostic tools. Dataset metrics describe the supplied observations; loan, circuit and SIR examples retain their stated simplifications. No model API, telemetry or external runtime requests.