Retain harm bound in to_integer() for AHR designs

[yihui]

Problem

to_integer.gs_design() rebuilds an AHR group sequential design with gs_power_ahr() at the rounded integer event counts, but the reconstruction did not forward the harm bound arguments (harm, hpar, test_harm). As a result, the harm boundary was silently dropped whenever an AHR design with harm bounds was converted to integer events:

x <- gs_design_ahr(
  analysis_time = c(18, 30),
  upper = gs_spending_bound,
  upar = list(sf = gsDesign::sfLDOF, total_spend = 0.025, param = NULL),
  lower = gs_spending_bound,
  lpar = list(sf = gsDesign::sfHSD, total_spend = 0.1, param = -2),
  harm = gs_spending_bound,
  hpar = list(sf = gsDesign::sfHSD, total_spend = 0.05, param = -2),
  test_harm = TRUE
)
unique(x$bound$bound)              #> "upper" "harm" "lower"
unique(to_integer(x)$bound$bound)  #> "upper" "lower"   <- harm lost

This surfaced while wiring harm bounds through the gsDesign2 Shiny app (keaven/gsDesign2Shiny#76), where integer sample sizes are on by default.

Fix

Forward harm/hpar/test_harm from x$input into the gs_power_ahr() call in the AHR/WLR branch. These fields are always present on the input and default to gs_b/-Inf/FALSE for designs without a harm bound, so non-harm designs are unaffected. gs_power_wlr() does not accept harm arguments, so the change is guarded to the AHR case (is_ahr).

After the fix, to_integer() retains the harm bound and the cumulative harm crossing probability under H0 at the final analysis matches the input (0.05 in the example above).

Tests

Added an independent test asserting the harm bound survives to_integer() and that the final cumulative H0 harm crossing probability equals the requested total spend. Existing to_integer tests still pass.

🤖 Generated with Claude Code

[jdblischak]

xref: follow-up to #640