Changelog

1.1.0

Multivariable X support: all three estimators now accept X of shape (n, k); func(x, params) receives X[i] directly
Zero-error feature handling: set x_err[:, j] = 0 for controlled/exact variables — MC won't perturb that column, Deming pins latent values to observed
Tutorial 4: multivariable regression (heat output model P = a·I² + b·T + c) with mixed-error inputs
Performance guide: analytical vs Monte Carlo runtime comparison with figures
More dramatic tutorial examples: sensor calibration (9:1 noise ratio, 5.4× confidence band difference), radioactive decay (σ_t=15 s, +32% bias + 3× underestimated uncertainty), method comparison (σ_x=25 μg/mL, −19% attenuation bias)

Type hints throughout the codebase (from __future__ import annotations, full mypy coverage)
Ruff linting (E, W, F, I, B rules) added to pre-commit hook alongside mypy
Replaced black/flake8 with ruff in dev dependencies
Benchmark suite: 13 physical scenarios, 200 parameter samples each, bias/RMSE/coverage statistics
Method comparison page in docs with figures and benchmark tables
Tutorials: sensor calibration, radioactive decay, Deming method comparison
MkDocs Material documentation with API reference via mkdocstrings

Redesigned API: three sklearn-like estimators (WeightedRegressor, XYWeightedRegressor, DemingRegressor)
Both analytical and Monte Carlo solvers for all estimators
Welford online covariance tracking for MC solver
Full covariance matrix output