NEWS


kernR 0.7.0.9000

kernR 0.7.0

New features

kernR 0.6.0

New features

kernR 0.5.0

Fidelity-provenance awareness when consuming PESTO manifests

kernR 0.4.0

New features

kernR 0.3.1

Testing

Minor improvements and fixes

kernR 0.3.0

Correctness

Features

Documentation and packaging

kernR 0.2.0

First publish of the local development line to AAGI-AUS. Lands the 0.0.0.9001 → 0.0.0.9015 cycle on top of the existing AAGI v0.1.x lineage; subsequent 0.0.0.x history below is retained as the per-feature changelog. The version-number jump (origin v0.1.1 → v0.2.0) signals the substantial new public surface arriving in this release.

Public surface (31 exports)

Cross-package contracts

Documentation

R CMD check posture

R CMD check --as-cran on the AAGI / CI Linux environment is expected to report 0 errors / 0 warnings / a small number of environmental NOTEs (new-submission boilerplate, HTML Tidy version on macOS). Two Apple-clang-21 toolchain WARNINGs surface only on the maintainer's local machine (R's own R_ext/Boolean.h and a personal ~/.R/Makevars); both are absent on CRAN's build farm and on standard GitHub Actions runners.

Historical development log

All historical entries below are retained for full traceability (0.0.0.9001 sole-authorship consolidation → 0.0.0.9015 third downscaling method).

kernR 0.0.0.9015

Third downscaling method: aggregate_downscale()

Closes the orchestra-completion gap surfaced after the 2026-05-16 deferral closeout: a downscaling method for the no-paired-training- data regime, where only the aggregate observation and a known aggregator are available. Companion to the proxymix Tier-2 stub from_aggregate_likelihood(), hosted on the kernR (consumption) side so proxymix's CRAN pre-submission stays untouched.

This is the third kernR downscaling method, structurally distinct from kernel_downscale() (CME, paired training data) and dist_regression() (bag-of-points, distribution-level regression). Each occupies a different cell of the (training-data-shape × aggregator-knowledge) matrix.

kernR 0.0.0.9014

Deferral closeout (post 2026-05-16 critical review)

Closes the three kernR-side tickets named as deferred in DRAINSTORMING/reports/orchestra_critical_review_response_2026-05-16.md: density-ratio fit/predict refactor, backend diagnostics, and properly null-calibrated total-order significance test.

Density-ratio fit/predict refactor (closes P0 #2)

Backend diagnostics for proxymix (closes P1 #3)

Conditional-permutation total-order significance (closes P0 #1 remaining work)

Test counts

Test suites at L99-coverage: density-ratio fit/predict + diagnostics

kernR 0.0.0.9013

Pick-Freeze p-value retraction (post critical-review, 2026-05-16)

The 0.0.0.9012 total_order_p_value = TRUE mode was found by critical review to be not null-calibrated. The pair-bootstrap samples the empirical joint distribution, not a null-of-no-effect, so under pure-noise Y every parameter was assigned a tiny p-value (p ≈ 1/(1 + B)) by mechanical bootstrap geometry rather than real signal. The orchestra smoke output active min-p = inert min-p = 0.010 was the visible failure.

Changes:

A separate feedback_total_order_calibration.md memory entry has been written to ensure the lesson — "bootstrap-around-empirical is not a null" — persists across sessions.

kernR 0.0.0.9012

Pick-Freeze bootstrap p-values for total-order HSIC sensitivity

kernR 0.0.0.9011

Cross-package contract formalisation

kernR 0.0.0.9010

proxymix density-ratio backend (UQ ag-stack roadmap §C1)

Contract symmetry: mmd_ppc() for the PESTO 0.3.0 manifest

kernR 0.0.0.9009

Fixes to §B2 (dr_date_scenario()) shipped earlier in 0.0.0.9007

kernR 0.0.0.9008

New features — kernel-based downscaling

Two complementary downscaling primitives, both rooted in the Park-Muandet-Fukumizu-Sejdinovic / Szabó-Sriperumbudur-Póczos-Gretton family of RKHS regression methods.

kernR 0.0.0.9007

New features (UQ ag-stack roadmap §B2)

Notes

kernR 0.0.0.9006

New features

Notes

kernR 0.0.0.9005

New features

Release-gate investigation (no package change)

kernR 0.0.0.9004

New features

Release-gate cleanup

kernR 0.0.0.9003

New features

kernR 0.0.0.9002

New features

kernR 0.0.0.9001

Authorship & administrative

kernR 0.0.0.9000