| Readers | Every Parquet column of both development bundles as DuckDB reads it, every Zarr array as zarr-python reads it, every TIFF's structure and pixels as tifffile reads them | every value |
| Bundle checks | The run's recorded values; thirteen bundles each broken one way | 22 values reproduced; exactly each one's findings |
| The Viewer | Every channel drawn at several levels, read back from the page | sample for sample |
| Nucleus expansion | The Lung bundle's own 5 µm expansion | the same 11,898 cells; 98.1% of pixels and 98.9% of transcripts in the same cells |
| Refinement | A simulated tissue's true cells | 84.7% of transcripts in their true cell (the instrument's 82.2%) |
| Clustering | scanpy, leidenalg, umap-learn and scikit-learn on both bundles and the simulated tissue | components within 10⁻¹³; the same neighbors and leading markers; the true types found (adjusted Rand index 0.95) |
| Labels from markers | scanpy's gene scores | every cell's score within 3 × 10⁻⁷; 97% of simulated cells their true type |
| Labels from a reference | SingleR on both bundles and the simulated tissue | correlations within 3 × 10⁻⁸; the same label on 99.97% to 100% of cells |
| Alignment | 10x's alignments of a published section's H&E and immunofluorescence; simulated images with known deformation | within 0.61 and 0.68 µm of 10x's without them; within 0.2 µm of the truth |
| Regions | QuPath 0.7.0; annotations on a simulated H&E | open there as written; within 0.18 µm of the truth |
| Cell graphs | SciPy and Squidpy | every edge |
| Enrichment and co-occurrence | Squidpy 1.8.3 on the same graphs | every count; every z within its permutation error |
| Distances and Ripley's K | SciPy's k-d tree; astropy | every median and mean; K without a correction to 10⁻⁹ |
| Regions measured | Shapely | every area to 2 × 10⁻⁷; every count |
| Niches | scikit-learn's k-means; the simulated tissue's domains | inertia within 0.2%; the kinds of domain found (adjusted Rand index 0.92) |
| Moran's I | Squidpy 1.8.3; an independent computation of both variances | every gene's I to 10⁻¹⁵; variances to 10⁻¹² |
| Design matrices | patsy | column for column |
| Differential abundance | statsmodels 0.15.0 (and R) on 19 cases | every coefficient, interval and p to 10⁻⁸ |
| Differential expression | limma 3.68.5 and edgeR 4.10.5 on nine pseudobulk tables | the same genes kept; every fold change, interval, t and p within 10⁻⁹ |
| Spatial statistics compared | limma 3.68.5 and statsmodels 0.15.0 on ten cases | within 10⁻¹³ |
| Error rates | 2,000 simulated null cohorts and 500 with known effects | each family within half a point of 5%; the effect gene found in 98%, its interval covering the truth in 95.0% |
| Figures | The suite's own readers of SVG and PDF; its own composite of the stains' samples | every mark as drawn; every pixel, and the Viewer's within one level |
| Certificates | Results made again by situcraft verify on x86-64 and arm64; copies tampered with | every file bit for bit; each edit found |
| Scale | A million generated cells, each analysis in a process of its own; a cohort of 2 million cells | within budget: Moran's I in 88 s and 0.94 GB on four cores; the cohort in 4.2 minutes and 0.65 GB |
