Changelog

1.0.0 October 24, 2026 · source

Added

• A rotatable 3D surface of the log-likelihood, with the optimizer's path and the ℓ(μ) and ℓ(σ²) slices drawn on it. Drag the handle under the point to move μ and σ² together, or turn to a side view to see the profile likelihood edge-on.
• A 3D view of the three tests: the likelihood surface at the chosen n, with the null hypothesis μ = μ₀ drawn across it and the geometry of the likelihood-ratio, score, and Wald tests marked on it (after Buse, 1982).
• 95% profile-likelihood and Wald confidence intervals below the inference figure.
• A step-size slider for gradient ascent. A large step overshoots the peak, and the run stops with a note.
• Before you press play, each algorithm shows where its next step would land.
• Both figures can be expanded to fill the browser window.

Changed

• Complete rewrite in Svelte (previously React), with a new visual design.
• New defaults: the sample was shifted by −0.38 so that μ̂ = 104 exactly (σ̂² is unchanged), and the null hypothesis starts at μ₀ = 96, where the three tests disagree at the .05 level. The μ₀ slider moves in whole numbers. Test statistics at a given μ₀ therefore differ from earlier versions.
• The default gradient-ascent step size was halved (γ = 0.05) so the first steps no longer overshoot.
• The second optimizer is now labeled Fisher scoring, which is what it always computed. Its path is unchanged.
• Moving the point by hand now resets the optimizer, so the next run starts from the new point.
• p-values are shown in APA style (p < .001 rather than 0.00).
• The explanations have been revised.

Removed

• The contour plot and the separate ℓ(μ) and ℓ(σ²) plots. Their content is now drawn on the 3D surface. Without WebGL, the contour plot is shown instead of the surface, and the inference figure is replaced by a message; the test results and intervals are still shown.

Fixed

• The likelihood-ratio and score annotations now use the variance re-estimated under the null, as the tests do, rather than σ̂².
• The score test's explanation said the score keeps growing as μ₀ moves away from μ̂. With the variance re-estimated under the null, it shrinks again far from μ̂, while the test statistic keeps growing.
• The density curve on the sample plot is now accurate in the tails when σ² is small.