BreedReliabR 0.1.2
(2026-10-31)
Operational-usability
release
- Add
br_check_inputs() as a non-destructive preflight
audit for candidate coefficient means, full joint covariance,
environmental-basis compatibility, optional TPE/training bases, and
fixed-k decision settings. The helper reports structural problems but
never repairs covariance matrices or scientific inputs.
- Add breeder-facing guidance that separates raw breeding data,
upstream genomic or GxE model fitting, and the BreedReliabR
decision-reliability layer.
- Add a beginner vignette showing how ordinary multi-environment data
become candidate coefficient means and a full joint covariance in a
transparent reaction-norm example.
- Add a genomic GxE vignette with phenotyped training genotypes and
separate unphenotyped candidates, deriving candidate
environmental-response coefficients and their full joint PEV before the
BreedReliabR hand-off.
- Add a model-integration guide covering GBLUP, RR-BLUP, random
regression, Bayesian workflows, factor-analytic GxE, discrete
environment-state predictions, and machine-learning uncertainty while
distinguishing validated, high-feasibility, conceptual, and insufficient
hand-off routes.
- Clarify candidate-major covariance labels, expected retained
fraction, Omega as relative environmental-to-prediction uncertainty, and
the geometric-only scope of
br_support().
- Preserve the seven-function statistical research core unchanged.
Version 0.1.2 adds usability, documentation, and preflight diagnostics
rather than a new uncertainty method or calibration claim.
BreedReliabR 0.1.1
(2026-10-16)
- Add Santosh Patil as a package author for substantial R programming
and coding contribution. Ashutosh Sawarkar remains the sole
maintainer.
- Update the maintainer email and add author contact and ORCID
metadata.
- Metadata-only release; no statistical, API, or package behaviour
changes.
BreedReliabR 0.1.0
First research release
- Freeze the seven-function research API for externally supplied
prediction means and full joint prediction-error covariance.
- Provide M0–M3 comparison, expected selected-set reliability,
candidate-level advancement probabilities, and the relative uncertainty
diagnostic Omega(H).
- Add geometric target-environment support diagnostics and an H-curve
workflow.
- Clarify assumptions, fixed-k interpretation, Monte Carlo error, RNG
behavior, candidate-major covariance ordering, and the limits of
geometric support.
- Retain the simulated maize data and add a reproducible
methods/empirical validation vignette. No G2F raw data are
included.
- External evidence: 2024 validation and forward year-held-out 2022
replication found poor absolute calibration, M2 worst, a small M3 Brier
advantage over M1, and overpredicted selected-set reliability. The
corrected L3a 2022 baseline is canonical; the original denominator error
and correction are retained in the separate audit archive.
Known limitations: probabilities are conditional on the supplied
upstream predictive distribution and target-environment specification.
Empirical calibration is poor in the two validation settings; software
release readiness does not imply calibrated probabilities or guaranteed
genetic gain.
BreedReliabR 0.0.3
- Added br_compare() and br_curve() as wrappers around the unchanged
H2 core.
- Added range and standardized nearest-environment support
diagnostics.
- Added compact S3 print, summary and plot methods for the new result
classes.
- Added entirely simulated maize data, a reproducible generation
script, and an executable end-to-end vignette with explicit fit-import
and calibration limits.
- Added executable examples to all principal public functions and
regression tests.
- Clarified expected-fraction set reliability, Monte Carlo limitations
and fixed TPE assumptions. No core mathematics or H2 tests were
changed.