NEWS

adaptDiag 0.1.1 (2026-06-28)

binom_sample_size(): fixed the normal approximation used to initialise the sample size search. The previous formula used only $p_1$ in the variance term; the correct Chow et al. (2017) formula uses $z_\alpha\sqrt{p_0(1-p_0)} + z_\beta\sqrt{p_1(1-p_1)}$. The exact discrete search result is unchanged, but the approximation now matches the cited reference.
multi_trial(): improved documentation for prior_sens, prior_spec, and prior_prev to explain that c(0.1, 0.1) is a bimodal near-Jeffreys prior that places most mass near 0 and 1, and that c(1, 1) (uniform) or an informative prior from pilot data may be more appropriate.
multi_trial(): added a seed argument for reproducible simulations (#5). Reproducibility is implemented via doRNG::%dorng%, which correctly handles RNG stream splitting across all parallel backends.
multi_trial() now uses a single foreach + doRNG::%dorng% dispatch path for all platforms, replacing the previous platform-specific branches (pbmcapply::pbmclapply on Unix, foreach::%dopar% on Windows). This fixes a latent Windows bug and simplifies the implementation. pbmcapply has been removed as a dependency and doRNG added.
Fixed an error in multi_trial() that occurred at early interim looks with low prevalence, when zero observations fell into a cell of the 2×2 contingency table (#5).