Bound issue in Power.ssrCP

[tianxiaowen]

• https://github.com/keaven/gsDesign/blob/master/R/ssrCP.R#L547 , should the upper bound be "b <- x$x$upper$bound[1]" instead of 'x$x$upper$bound[2]'?

• https://github.com/keaven/gsDesign/blob/master/R/ssrCP.R#L607, there might be a missing 'x$' i.e., "changepoint2 <- x$x$upper$bound[1]"

[nanxstats]

Let's test how AI thinks:

GPT-5-Codex thinking high

• R/ssrCP.R:547 - In the "no SSR needed" branch, grid integrates over the interim statistic z1, so the bounds must cover the continuation region [x$x$lower$bound[1], x$x$upper$bound[1]]. Using x$x$upper$bound[2] shrinks that interval whenever the stage-2 boundary differs (typically smaller) and drops probability mass for z1 between the true interim upper bound and the final boundary, which biases the computed power/expected sample size. Changing b to x$x$upper$bound[1] fixes the integration limits.
• R/ssrCP.R:607 - The fallback assignment changepoint2 <- x$upper$bound[1] omits the leading x$ (i.e., it tries to take $upper from the ssrCP object instead of its embedded gsDesign object). Whenever cpmax <= x$cpadj[2] this path is taken and x$upper is NULL, so attempting $bound[1] raises $ operator is invalid for atomic vectors/NULL. The statement should read changepoint2 <- x$x$upper$bound[1].

Claude Sonnet 4.5

Issue 1 (Line 547): Valid issue

The bound should be x$x$upper$bound[1] instead of x$x$upper$bound[2].

Reasoning: This section integrates over interim Z_1 values in the continuation region (between a and b). The expected sample size calculation on lines 544-545 correctly uses x$x$upper$bound[1] (the interim upper bound). The grid bounds should match this - integrating between the interim lower and upper bounds. Using bound[2] (final analysis bound) is inconsistent and incorrect.

Issue 2 (Line 607): Valid issue

Missing x$ prefix - should be changepoint2 <- x$x$upper$bound[1].

Reasoning: The ssrCP object structure (lines 426-437) shows that x$x contains the gsDesign object, not x$upper. Throughout the function, bounds are consistently accessed via x$x$upper$bound and x$x$lower$bound (see lines 524, 526, 529, 535, 544, 570, 591, 595, 598). Line 607 is missing the intermediate x$ reference.

Conclusion

Both issues are legitimate bugs. The corrected lines should be:

• Line 547: b <- x$x$upper$bound[1]
• Line 607: changepoint2 <- x$x$upper$bound[1]

These bugs would cause incorrect power/expected sample size calculations in specific scenarios where the respective code paths are executed.