Places where the book computes by hand, or only in R, what process_improve already provides

[kgdunn]

Collected while executing every Python case against the library (#274). None of these are errors: every number listed here reproduces. They are places where the book and its companion package have drifted apart in coverage rather than in results, so a reader following the text does not meet the function that exists for the job. Each is an editorial call, hence one issue rather than a pull request.

Solutions given only in R. least-squares-modelling/least-squares-exercises.rst is the largest case. All of these were checked against process_improve and reproduce the quoted figures exactly:

• The prediction intervals at 430, 480 and 520 degrees F (line ~78) are described as "calculating this term manually, or using predict(model, newdata=, int='p') in R". OLS.prediction_interval returns a Bunch with predicted, lower and upper and gives all three intervals to the printed digits. There is no Python version of this solution today.
• The 1/T model (lines ~118-146), including both standard errors.
• The RMSEP, influence plot and point-removal exercise (lines ~186-198): OLS with leverage_ and influence_ reproduces the slope, the residual standard error, both confidence intervals and both RMSEP values.
• summary-of-steps-to-build-and-investigate-a-linear-model.rst gives steps 2 to 8 only in R. print(OLS().fit(X, y)) renders the summary(lm(...)) layout the steps describe, and conf_intervals_, residuals_, fitted_values_ and prediction_interval cover the rest.

Computed by hand where the library has it.

• multiple-linear-regression.rst recomputes adjusted R-squared from R-squared (line ~315); adj_r2_ holds it, and agrees to three decimals.
• The same chapter forms residuals as y - predict(...); residuals_ holds them, NaN-padded to the original row count so they align with the raw frame.
• outliers-...rst fits the same model twice, once with OLS for leverage and Cook's distance and once with statsmodels for the studentized residuals. OLSInfluence supplies all three; alternatively a studentized_residuals_ attribute on OLS would remove the second fit, which is a library request rather than a book one.
• univariate-review and process-monitoring do their interval and chart arithmetic with scipy throughout. univariate.t_value, univariate.confidence_interval, monitoring.ControlChart and monitoring.calculate_cpk cover much of it. This is the largest surface and the one where the case for changing is weakest: the chapters are teaching the arithmetic, so showing it explicitly may be the point.

Style, not coverage. Several least-squares blocks pass .values into OLS.fit, which discards the column names the class is built to display; passing the frame makes print(model) show formula: VapourPressure ~ InvTemp3 and label the coefficient row, which is closer to the R output the prose compares against.

Now that every case runs in CI, any of these can be converted with the numbers checked as part of the change.