AccSamplingDesign 0.1.3 (development build)

Saddlepoint-MLE plan search now starts from the delivered Delta-MLE sample size and acceptance constant. If that initializer fails validation, the previous Normal initializer is used automatically.
The initializer affects search efficiency only. All final probabilities, feasibility decisions, adjacent-size checks, and reported risks continue to use the saddlepoint MLE OC.
Search diagnostics now report the initializer, fallback reason, evaluated sample sizes, and distribution constructions. The public method names and defaults are unchanged; delta_mle remains the default.
Validation found equivalent delivered plans in all 43 commonly supported configurations. Timing improved overall but remained configuration-dependent.

AccSamplingDesign 0.1.2 (development build)

Added opt-in saddlepoint_mom, displayed as Saddlepoint MOM OC with an experimental/incomplete-validation label. The estimator and rule are unchanged.
Connected integer plan search, manual plans, acceptance probabilities, summaries, OC data, and plots. Successful calculations require both integration refinement and population-CGF quadrature checks. Unsupported calculations return explicit errors, including ordinary cases that failed development tests.
This integration does not certify MOM accuracy or risk control. No tolerance was relaxed, and no MLE/Delta substitute was introduced. delta_mle remains the default; existing method formulas and reporting conventions are preserved.

Added opt-in method = "saddlepoint_mle" for unknown-precision Beta variables plans. It numerically integrates a normalized joint saddlepoint approximation to the same MLE inspection rule. delta_mle remains the default.
New saddlepoint plans use integer sample sizes, verify their own numerical analytical risk constraints, and expose integration diagnostics. They do not provide a guarantee of finite-sample risk control or global size minimality.
Numerical integration uses no simulation, has an initial setup cost, and caches a distribution for repeated acceptance-constant calculations. Unsupported or nonconvergent calculations raise errors; no Delta fallback is used.

Delta-MLE construction and evaluation reject invalid sample sizes below two, nonfinite parameters, and negative acceptability constants. MLE refinement now respects the size bound and rejects analytically infeasible results.
The Delta formulas and existing Delta/Govindaraju-Kissling reporting conventions are unchanged. The MOM implementation is unchanged.

Known-theta Beta variables plans now meet both risk constraints at the delivered (rounded-up) sample size. Previously, the soft-penalty constrained search could return a plan that already violated PR <= alpha or CR <= beta, and rounding the sample size up could not repair it. The search is now refined by solving the two binding risk equations (Nelder-Mead in log coordinates), the sample size is rounded up from the binding solution, and both risks are verified at the delivered size. Known-theta sample sizes are now the minimal feasible integers, e.g. the reported moisture case (PRQ = 0.005, CRQ = 0.01, USL = 0.05, theta = 500) changes from n = 112 (violating both risks) to n = 117. If the refinement does not converge, the constrained-search solution is delivered after verification, with a warning.
For known-theta Beta plans, the reported PR and CR, the OC curve and the plots now describe the delivered integer plan, and n equals sample_size. Unknown-theta plans keep the previous reporting semantics.
accProb() no longer fails for a Beta plan with acceptability constant k = 0 (floating-point cancellation made the closed-form discriminant marginally negative).

Added analytical Delta–MLE and Delta–MoM methods for Beta variables acceptance sampling when the precision parameter, theta, is unknown.
Delta–MLE is now the default unknown-theta method. The sample-size adjustment used in version 0.0.8 remains available as "gk_adjustment".

Added the citation for the AccSamplingDesign article published in The R Journal, volume 18, issue 1, pages 368–381 (doi:10.32614/RJ-2026-007).