[MaartenMarsman]

prior_sensitivity_check() loses the vary argument. The sweep holds the standardized diagonal rate eta fixed and moves the raw rate with the slab scale. A GGM or mixed fit whose diagonal prior was given as a raw rate stops with an error; the man page gains a section with the refit loop for that case.

Background: with the raw rate held fixed, the graph-dependent normalizing constant of the hierarchical prior changes with the scale and the importance weights did not carry it; the refits also ran with Z-ratio constants for the original rate. With eta held fixed the standardization separates the scale from that constant, which cancels in the importance ratios; a GGM and a mixed hierarchical fit reweighted to 2x reproduce direct refits to within 0.01 in inclusion probability (dev/audit/checks/2026-10-22-hierarchical-reweight-check.txt).

Verification

• test-prior-sensitivity.R and test-anchor-curve.R pass (two blocks on the removed argument replaced by one).
• R CMD check --as-cran, run before the Date update: 0 errors, 0 warnings, 1 note (Date field over a month old). The Date was then set to 2026-10-22; the check was not re-run.
• Develop vs branch demo: dev/review-2026-08/repro/run-rawrate.sh. Develop runs the fixed-rate sweep on a raw-rate fit; the branch errors. The default eta fit gives identical log Bayes factors on both sides.