* 2026-10-19
- Version 0.2.1.
- Select the confidence set using cumulative MCS p-values, retaining equality
  at alpha; progress messages now describe the actual confidence set.
- Fix the moving-block start range so the final observation can be resampled.
  Document k=0 (iid) and reject full-sample or invalid block lengths.
- Repair multi-column forecast inputs in all loss functions. Validate input
  domains and calculate logistic/logarithmic losses more stably.
- Handle zero and constant nonzero loss contrasts explicitly; reject accidental
  zero bootstrap variance for nonconstant contrasts. Remove blanket NA-to-1
  replacement and include bootstrap ties in upper-tail p-values.
- Select automatic k using significant coefficients in pairwise-difference
  AR models; skip constant differences and validate min.k against sample size.
- Support numeric data frames and single-model sets; reject invalid controls,
  empty inputs and ambiguous model names.
- Compute bootstrap model means once, stream resampling indices, and avoid
  B-by-m-by-m arrays. TR bootstrap maxima are accumulated backwards along
  the elimination path so each pair is evaluated once per pass.
- Update help, terminology, references, package metadata and add base-R
  regression tests, including independent statistical reference calculations.

* 2026-03-19
- New version 0.2.0
- Bug fixes and improved speed
- Parallel computation is not supported anymore