To audit single-case randomization test software, reproduce the design and one small hand-worked example before trusting the output. Lock the assignment set, selection probabilities, test statistic, direction, tail, tie rule, missing-data rule, exact or Monte Carlo mode, draw count, seed, package version, and input order. Compare results field by field. Differences should be explained through settings or implementation, never averaged together.

Verify the input sequence

Confirm case, tier, condition, time, outcome, phase labels, missing states, and sort order. A reordering error can alter schedules and statistics.

Verify the design generator

Compare the software's permissible assignments with the protocol list. Check minimum phases, blocks, adjacency limits, and probabilities.

Verify the statistic implementation

Inspect formula, windows, direction, weighting, ties, precision, and case aggregation against a hand-worked example.

Verify inference settings

Check tail, exact or sampled mode, draw count, replacement, seed, finite-sample correction, and denominator.

Preserve versions and outputs

Archive code, package dependencies, system information, configuration, logs, assignment table, and reviewer sign-off for reproducibility.

Build Kira's analysis configuration diff

Start with a versioned record for Kira's reconciling software results. For the audit single-case randomization test software question, record the protocol date, design, randomization unit, complete permissible assignment set, assignment probabilities, actual draw, test statistic, direction, tail, tie rule, missing-data rule, analysis mode, software, and reviewer. Keep Kira's raw time series in chronological order and link every derived value to its source. The packet should let an independent analyst reproduce both the observed statistic and its reference distribution.

Work the example for Kira

Kira's locked dataset produces p=0.08 in one configuration and p=0.04 in another. The audit finds that the first uses a two-sided inclusive tail and the second uses a one-sided strict tail with a smaller assignment set. The example ledger does not supply either configuration's assignment count or extremeness numerator, so the two decimals cannot be independently recomputed from this page. The team restores the prospectively specified inclusive one-sided scheme and documents its reproducible fraction rather than choosing the smaller output. Preserve every intermediate assignment count, statistic, fraction, and rounding step. The example is fictional and teaches the stated method; it does not create a universal alpha, minimum design, treatment recommendation, or effect category for Kira's outpatient communication research clinic.

Audit Kira's randomization evidence

Kira's configuration diff lists design type, phase limits, schedule count, schedule probabilities, input order, statistic code, one-sided direction, inclusive ties, precision, missing-point handling, exact enumeration, software and package versions, and output checksum. A four-schedule hand example matches the selected implementation. The audit also checks protocol amendments, assignment deviations, duplicate or missing observations, phase-label changes, output precision, code and package versions, and whether the reported test matches the actual randomization. Unresolved discrepancies remain on hold with an owner and due date.

Prevent the main inference error in Kira's review

Matching file names can hide different analysis states. Defaults also change between versions. Kira's pipeline stores machine-readable settings with each result and refuses release when a setting is absent or conflicts with the dated protocol. A randomization p value depends on the assignment mechanism and predeclared statistic. It does not turn nonrandom treatment changes into randomized experiments, quantify the effect's size, establish clinical importance, or estimate a population effect.

Integrate design evidence for Kira

A methods reviewer resolves implementation ambiguity before clinical or research interpretation. If published software behavior is unclear, the report states the uncertainty, preserves raw data and code, and avoids a definitive p-value claim until the discrepancy is understood. The WWC Version 5.0 handbook is a research-review standard rather than a universal care protocol. The responsible team still evaluates the design's prediction, verification, replication, phase patterns, attrition, assignment fidelity, measurement quality, and fit to the question.

Protect participation and access for Kira

Research design never outranks Kira's welfare. Keep augmentative and alternative communication, interpreters, mobility, food, water, bathroom use, prescribed care, health support, rest, relationships, and emergency help available. Use accessible communication for consent and assent processes, withdrawal, discomfort, and feedback. If the selected schedule becomes unwanted, unsafe, medically inappropriate, or infeasible, qualified people act for safety and document the design consequence.

Use randomized single-case sources within scope for Kira

For Kira's randomization-software audit, the BACB ethics hub and CASP public summary provide professional context, and the BCBA Test Content Outline identifies examination content on measurement and single-case design. A randomized SCED methods paper explains prospective assignment sets, randomization tests, and attainable p-value limits. An effect-measure paper describes a priori statistic choice and the boundary between tentative causal inference and effect magnitude. Research on rapidly alternating designs, changing-criterion designs, and randomized case series shows that assignment schemes and test behavior are design-specific. ASHA supports continuous AAC access.

Rehearse Kira's analysis before outcomes exist

Before the outpatient communication research clinic begins, Kira's team runs the analysis configuration diff with a small fictional sequence. The dry run confirms that the assignment generator returns exactly the permitted schedules, the selected schedule can be traced to the random draw, and the randomization-software audit statistic matches a hand calculation. Reviewers deliberately include a tie, a missing observation, and an assignment at each boundary so default behavior becomes visible. They save the test inputs, expected outputs, code version, and correction log. This rehearsal checks the machinery for reconciling software results; it never uses Kira's future outcomes or changes the predeclared analysis after data collection.

Close Kira's methods review

Review the analysis configuration diff with Kira, the responsible clinician, and a statistician or methodologist experienced in randomization inference. Preserve protocol, assignments, raw observations, graph, code, outputs, deviations, inference limits, client input, decisions, and future design changes. Keep the page draft and noindex until the named external reviews confirm that the research description, accessible participation, and clinical boundary are accurate.

Related resources

Sources