@@ -0,0 +1,13 @@
+(v1.12.6)=
+### 1.12.6 {small}`2026-11-05`
+
+#### Bug fixes
+
+- Fix {func}`~scanpy.tl.ingest` truncating to the hardcoded default `settings.N_PCS` instead of the number of PCs that `pp.neighbors` actually used, whenever a reference dataset's `neighbors` params recorded neither `use_rep` nor `n_pcs` {smaller}`Siddhardha Nanda` ({pr}`4401`)
+- Return zero residuals for zero-count cells and genes in {func}`scanpy.experimental.pp.normalize_pearson_residuals`, avoiding undefined values in subsequent analysis {user}`HamzaAhmed-bio` ({pr}`4404`)
+- Fix {func}`~scanpy.pp.highly_variable_genes` selecting more than `n_top_genes` genes with duplicate variable names when using `seurat_v3` or `seurat_v3_paper`. ({pr}`4409`)
+- Fix {func}`~scanpy.pp.highly_variable_genes` with `flavor="cell_ranger"` and `n_top_genes` selecting no genes when the top genes have infinite normalized dispersion {smaller}`L Jorgensen` ({pr}`4419`)
+- {func}`~scanpy.pp.highly_variable_genes` with `batch_key` (`flavor="seurat"` or `"cell_ranger"`) no longer flags more than `n_top_genes` genes or merges the statistics of genes that share a name {smaller}`R García Revett` ({pr}`4439`)
+- Fix {func}`~scanpy.pp.highly_variable_genes` (`flavor='seurat_v3'`), {func}`~scanpy.pp.calculate_qc_metrics`, and {func}`~scanpy.tl.rank_genes_groups` loading entire row-chunked {class}`dask.array.Array`s into memory when checking for count data or counting nonzero cells per gene {smaller}`Y Su` ({pr}`4446`)
+- Fix {func}`~scanpy.pp.calculate_qc_metrics` for sparse matrices of small integer dtypes, which gave negative `pct_counts_in_top_N_genes` (e.g. `int16`) or raised (e.g. `uint16`) {smaller}`P Paladugu` ({pr}`4461`)
+- Fix {func}`~scanpy.pp.neighbors` and {func}`~scanpy.tl.ingest` with `umap-learn` 0.6 {smaller}`P Angerer` ({pr}`4480`)