Supported analysesLimits and conditions

Supported analyses:limits and conditions.

Each workflow has fixed inputs, assumptions, and claims. Requests outside them are refused or paused, not approximated.

Across all workflows

Pre-specified choices: The test, comparison direction, and threshold are fixed in advance. Methods are not switched after diagnostics or significance are seen.
Bounded inputs: Each workflow has defined sample sizes and value ranges—for example, 2 to 200 per group or cell in several workflows. Inputs outside the range are refused.
Not covered: Repeated measures beyond two time points, multi-factor designs and interactions, covariate adjustment, general mixed models, single-cell data, and a general catalog of non-parametric tests.

Workflow by workflow

Two group means: Two independent groups, one continuous outcome, one pre-specified comparison. Welch’s t-test relies on an approximation; no exact finite-sample error rate is claimed.
Before and after: Complete pairs and exactly two time points. Incomplete pairs are not dropped or imputed silently; paired rank tests are not supported. A before–after change is not presented as a causal effect.
Three or more groups: One factor and independent groups under the adopted one-way model (normal errors with equal variances). Individual intervals are not simultaneous intervals, and a significant overall test does not mean every pair differs.
Ranks: The result states a rank-based hypothesis; it is not converted into a mean or median difference or a causal effect. Only tie patterns covered by the adopted calculation are accepted; all-identical values or ties created by rounding are refused or paused. Paired and multi-group rank tests are not supported.
Several measurements: Same two groups and an outcome set fixed in advance. Choosing outcomes after seeing results is not covered. The correction controls false positives across the set; it does not say how likely any single outcome is to be true.
Bulk RNA-seq: Gene-level unnormalized integer counts, two independent groups, one comparison, a single batch. Not covered: alignment or quantification from FASTQ, estimated or arbitrary-format counts, paired or time-course designs, more than two groups, custom preprocessing or outlier removal, multi-batch adjustment, single-cell, pathway analysis, causal claims, and sample-removal stability. False-discovery control applies to the gene set, not to each gene.
Equipment failure: One outcome-independent failure type, analyzed with Student’s t-test. Inference concerns the population mean difference; missing values are not recovered.
Three fixed batches: Exactly three fixed batches, two groups, each day weighted equally. Real cage independence and pre-computed slopes are not verified, no model is chosen by significance, and results do not generalize to future batches.
Replication: One original summary and one replication with individual data. An approximate reference interval may be shown; it does not include figure-reading error or correct for the original study’s selection or publication. Replication success, failure, or equivalence is not judged; there is no p-value and no meta-analysis. A known shared control stops the independent-model interval.
Two opportunities to continue or stop: Two fixed looks at 2.5% each. Each look reports a 97.5% interval; together they cover the effect at least 95% of the time across the whole procedure. This does not hold given that the study stopped early, the effect estimate at stopping may be biased, and per-look intervals are not shown as overall 95% intervals.