What "normal" actually looks like

The most common question about a new dataset is whether its numbers are good, and the usual answer is a threshold copied from a tutorial. Below is a different answer: the same metrics measured on public reference datasets released by the platform vendors, each produced by a script you can download and re-run. Compare your own numbers against these rather than against a rule of thumb.

Assay	Scale	Typical per-cell / per-spot values
Single-cell RNA (droplet) 10x PBMC 3k	2,700 cells	817 genes	2,197 UMI	2.03% mito	how it was run
CITE-seq (RNA + protein) 10x 5k PBMC TotalSeq-B	5,247 cells	32 antibodies	2,757 ADT counts	0.25% isotype background	how it was run
Single-cell ATAC 10x 5k PBMC scATAC	4,585 cells	14,256 fragments	FRiP 0.801	0.554 TSS-proximal	how it was run
Multiome (RNA + ATAC) 10x PBMC 3k Multiome	2,711 cells	1,791 genes	3,790 UMI	14,479 ATAC fragments	how it was run
Spatial (Visium) 10x human lymph node	4,025 spots	5,999 genes/spot	20,239 UMI/spot	0.99% mito	how it was run

How to use this

A number far below these is a question, not a verdict. Tissue, chemistry and cell type all move them. Nuclei give fewer genes than cells; solid tissue gives more mitochondrial signal than blood.
Mitochondrial fraction is the most misused of the five. 2.03% here is blood. Applying a 5% ceiling to heart, liver, muscle or kidney deletes the cells you are studying — see how to set and defend thresholds.
FRiP and TSS fraction have no RNA equivalent. They measure signal against background in ATAC, and a library can have plenty of fragments while failing both — see the ATAC example.
Isotype background is the calibration of a whole antibody panel. At 0.25% a positive stain means something; at several percent, weak markers cannot be interpreted at all.
Spot values are not cell values. A Visium spot covers several cells, so 5,999 genes per spot is ordinary while the same number per cell would be exceptional.

Check your own numbers

The no-upload self-check takes the five numbers from your Cell Ranger summary and compares them tissue by tissue — nothing leaves your browser. If the data can be shared, the free report computes the distributions rather than the medians, which is where the real answer lives.

Every value on this page is read from the metrics file the corresponding run produced; re-running the published script reproduces it. The scripts are at /code.