A second sample id is not a second patient: three of the interesting pairs in GSE252378 are one patient twice, and reading them as independent doubles the apparent n

**The concrete case.** GSE252378 contains three lineage-switching pairs, each of which is **one patient sampled twice**, stated in GEO's own `!Sample_description` field: | pair | fusion | states | what a careless read gives | |---|---|---|---| | `p279` / `p318` | KMT2A-AFF1 | infant B-ALL, then the AML it became | 2 patients instead of 1 | | `148752` / `152985` | KMT2A-ENL | infant B-ALL, then AML, **both under a menin inhibitor** | 2 patients instead of 1 | | `R3595` / `R3652` | KMT2A-EPS15 | infant B-ALL, lineage-switching | 2 patients instead of 1 | The paper says so too: it profiled "36 broadly representative MLL-rearranged leukemia samples, **including three samples that underwent a lymphoid-to-myeloid lineage-switching event**" - three switching events, six sample ids [E2]. **This corpus already contains the mistake in a mild form.** The existing entry recording this deposit's menin arm describes `148752` and `152985` as "two primary infant lineage-switching patient samples", which is literally true and reads as two patients. The dispatch brief for this beat stated that the drug-treated samples "belong to two different patients" and that the AFF1 arm contains "three PRIMARY patient samples". **Both are one patient short.** The correct counts are: menin under drug, **n = 1 patient**; AFF1 in primaries, **n = 2 patients in 3 states**. **The general rule.** Before counting samples as independent, read `!Sample_description` and `!Sample_source_name_ch1` for the words *paired*, *matched*, *relapse*, *diagnosis*, *pre-*, *post-*, *timepoint*, *switch*. GEO puts the pairing in prose, not in a structured field, so no parser finds it and no metadata table shows it. The consequence is not cosmetic: an n of 2 read as 4 halves every standard error and turns a paired design into an unpaired one.