[antonsoo]

A negative Manly parameter discards valid inverse values:

from gstools.normalizer import Manly

normalizer = Manly(lmbda=-0.1)
print(normalizer.denormalize(normalizer.normalize([0., 1., 2.])))

This returns [nan nan nan] instead of [0. 1. 2.]. The inverse requires 1 + lmbda*y > 0, so the upper bound for negative lmbda is -1/lmbda. Correct its sign without changing the transformation or fitting formulas.

Regressions cover negative parameters, negative/zero/positive observations, NaNs and invalid inverse inputs. The full suite passes (214 tests and three subtests; two existing skips). Source/benchmark Ruff checks and wheel/sdist builds also pass.

Separate native kriging runs over the bundled DWD temperature observations restore all 576 formerly NaN held-out predictions, agreeing with independently assembled covariance solves and scalar inverses within 1.23e-12. Nine zero/positive-parameter controls remain identical, as do all variance and forward-transformation outputs.