fix: rc7 review findings across the 4.0.0 RC series

[ehrlinger]

What

These are fixes for the seven findings from a rollup review of v4.0.0-rc1..main (239 commits, about 2,300 added lines in R/), on one branch so rc7 includes them all.

#	Where	Fix
1	gg_vimp.rfsrc()	Multi-class nvar kept the least important variables. The importance matrix is in predictor order and was trimmed before it was sorted: gg_vimp(rf, nvar = 2) on iris returned Sepal.Length and Sepal.Width. It now ranks by the overall all column first. The randomForest method already did this. This bug is also on CRAN (3.5.x).
2	.varpro_rank_of()	The digit-suffix fallback for expanded factors let a real column x1 borrow x10's rank. It now applies only to names that aren't themselves columns of the feature matrix.
3	gg_auct.rhf()	method is recorded (attr(, "method")) and named on the y axis ("Cumulative/dynamic AUC(t)" / "Incident/dynamic AUC(t)"). A method that contradicts a supplied auct_fit now warns instead of being dropped.
4	gg_ale_rfsrc() categorical	Each level-to-level step now averages over the observations at both adjacent levels (Apley & Zhu; ALEPlot), not just the upper level.
5	gg_ale_rfsrc() efficiency	Continuous, categorical and interaction ALE now make one predict() call per variable or surface. A default 25 × 25 interaction surface used to make up to 2,500. Continuous and interaction values are identical to main, checked with identical() on the airquality fixture.
6	plot.gg_partial_varpro()	The default type draws the two level curves only; causal must be asked for. It's a contrast that starts at zero, and on a regression it squeezed the level curves into a thin band. The varpro vignette now asks for all three explicitly. This changes the default on CRAN.
7	DESCRIPTION	Date updated for the next RC (it was 2026-08-31).

NEWS has new bullets for 1 and 6, the two changes CRAN users will see. For 3 and 4 the existing v4 bullets are amended, since both features are new in v4. Item 2 is internal.

Baselines

Five were regenerated, and I checked each diff:

• four gg-partial-varpro-* baselines lose only the causal series (fewer shapes, no causal legend text);
• gg-auct-chf changes only its y-axis label.

No other baseline changed, and main hadn't moved since branching.

Verification

• document(), lint_package(): 0 lints
• NOT_CRAN=true VDIFFR_RUN_TESTS=true devtools::test(): FAIL 0 | SKIP 6 | PASS 2212. The skips are pre-existing and deliberate, and no snapshot files were deleted.
• New tests, each seen failing before its fix:
  • rfsrc multi-class nvar keeps the top 2;
  • x1 doesn't borrow x10's rank;
  • gg_auct warns on a method mismatch and records the method;
  • categorical ALE steps equal the two-level means;
  • ALE makes one predict() call.
• R CMD check --as-cran on a clean export of this branch: running, and I'll post the result here.

🤖 Generated with Claude Code

[ehrlinger]

R CMD check --as-cran on a clean git archive export of ca708931, with the PDF manual: Status: OK (0 errors, 0 warnings, 0 notes). Timings: examples 14s, examples with --run-donttest 35s, tests 21s, vignette rebuild 47s. That is down from 60s on main; the explainability vignette runs gg_ale_rfsrc() live, so the batching may account for some of it, but one run on each side is not a controlled measurement.

[Copilot]

Copilot review overview

🟡 Changes recommended

ALE aggregation can replace occupied-group NA predictions with zero, and the new AUC method attribute remains undocumented.

Review effort: Balanced
Findings: 1 · 1

Open (2)

• Preserve NA means for occupied groups · New
• Document the new method return attribute · New

What changed in this PR

Addresses seven v4.0.0 release-candidate review findings across importance ranking, ALE computation, AUC metadata, and partial-varpro plotting.

Changes:

• Corrects VIMP and categorical ALE calculations while reducing ALE prediction calls.
• Records and labels AUC methods and changes partial-varpro plot defaults.
• Updates tests, snapshots, documentation, NEWS, and release date.

File	Description
DESCRIPTION	Updates release date.
NEWS.md	Documents user-visible fixes.
R/​gg_ale_rfsrc.R	Corrects and batches ALE calculations.
R/​gg_auct.R	Records AUC method and warns on conflicts.
R/​gg_vimp.R	Ranks multiclass VIMP before trimming.
R/​plot.gg_auct.R	Adds method-specific axis labels.
R/​plot.gg_partial_varpro.R	Excludes causal curves by default.
R/​utils.R	Tightens expanded-factor rank matching.
man/​plot.gg_partial_varpro.Rd	Updates generated plot documentation.
tests/​testthat/​test_gg_ale_rfsrc.R	Tests ALE correctness and batching.
tests/​testthat/​test_gg_auct.R	Tests AUC method handling.
tests/​testthat/​test_gg_vimp.R	Tests multiclass top-variable selection.
tests/​testthat/​test_plot_gg_auct.R	Updates axis-label expectation.
tests/​testthat/​test_varpro_importance_order.R	Tests digit-suffix rank isolation.
tests/​testthat/​_snaps/​snapshots/​gg-auct-chf.svg	Updates AUC label baseline.
tests/​testthat/​_snaps/​snapshots/​gg-partial-varpro-both.svg	Updates two-panel default baseline.
tests/​testthat/​_snaps/​snapshots/​gg-partial-varpro-categorical.svg	Updates categorical baseline.
tests/​testthat/​_snaps/​snapshots/​gg-partial-varpro-continuous.svg	Updates continuous baseline.
tests/​testthat/​_snaps/​snapshots/​gg-partial-varpro-mortality.svg	Updates mortality baseline.
vignettes/​varpro.qmd	Explicitly requests all three estimators.

Files not reviewed (1)

• man/plot.gg_partial_varpro.Rd: Generated file

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[codecov]

Codecov Report

❌ Patch coverage is 98.07692% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 90.88%. Comparing base (c042902) to head (e1f44b0).

Files with missing lines	Patch %	Lines
R/plot.gg_auct.R	80.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #302      +/-   ##
==========================================
+ Coverage   90.85%   90.88%   +0.02%     
==========================================
  Files          60       60              
  Lines        5875     5882       +7     
==========================================
+ Hits         5338     5346       +8     
+ Misses        537      536       -1     

Files with missing lines	Coverage Δ
R/gg_ale_rfsrc.R	96.17% <100.00%> (+0.82%)	⬆️
R/gg_auct.R	95.00% <100.00%> (+1.06%)	⬆️
R/gg_vimp.R	88.50% <100.00%> (+0.05%)	⬆️
R/plot.gg_partial_varpro.R	92.64% <100.00%> (+0.03%)	⬆️
R/utils.R	98.27% <100.00%> (ø)
R/plot.gg_auct.R	93.54% <80.00%> (-2.75%)	⬇️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.