Inputs, reproducibility and scope

Input dictionary

Input	Meaning	Contract
prob	Supplied success probabilities in sequence order	Non-empty numeric vector, finite values in [0,1]
x	Observed binary sequence	Zeros and ones in the same order as prob
tail	Inclusive one-sided probability	"lower" or "upper", not a two-sided test

The four exported functions are described in the exact-distribution guide. No external dataset is needed: the small example uses supplied probabilities.

A deterministic reproducibility check

This four-position example is a numerical identity check, not a simulation study.

library(BernoulliRuns)
prob <- c(.15, .40, .75, .60)
dist <- runs_distribution(prob)
stopifnot(abs(sum(dist$probability) - 1) < 1e-12)
stopifnot(abs(sum(dist$runs * dist$probability) - runs_mean(prob)) < 1e-12)
count_runs(c(0, 1, 1, 0))

## [1] 3

Record the package version with packageVersion("BernoulliRuns") and the R environment with sessionInfo(). Exact refers to the model’s run-count distribution, subject to floating-point arithmetic, not to guaranteed test calibration after estimating probabilities from these same observations.

Scientific and publication boundary

The trials must be mutually independent and their probabilities treated as fixed. Longest runs, dependent sequences, estimated-probability calibration and two-sided p-values are outside the public API.

Use citation("BernoulliRuns") to cite the software. The underlying finite-state approach is credited to Fu and Koutras (1994), DOI: 10.1080/01621459.1994.10476841. That reference is not a new paper by the package author.