[Reynold4k]

Threadfin provides reproducible clone-reclustering controls and a GC manuscript review based on saved, auditable real-data outputs. The three presets (cohesive, continuous, discrete) expose UMAP/graph parameters with explicit overrides and saved requested/effective settings. Existing defaults remain available. cluster_on="embedding" adds the historical notebook's Scanpy graph on clone UMAP; the default distance graph retains independent display controls.

Figure 2B shows newly computed GSE246382 clone embedding, coloured by unsupervised Leiden clone_cluster with a captured-size legend. It averages saved receptor-excluded cell UMAP within 49 frozen same-mouse sequence-defined families (at least three cells, 260 cells), embeds Euclidean distance-matrix rows with UMAP, then applies Scanpy neighbours and Leiden. Historical geometry/graph settings are fixed before gate/marker inspection. The primary map has three clusters with identical membership at seeds 123 and 7; all three presets and both seeds are retained. Post hoc measured gates and RNA annotate GC selection-associated, LZ/output-enriched and mixed states. These are descriptive model-antigen GC groups, without measured future fate or extension to non-GC samples.

Supplementary 8 preserves the historical Top2a output separately and supplies current GSE246382 cell gates, clone-averaged Myc and marker summaries by unsupervised cluster. Official GSE180920 day-7/day-14 counts, metadata and explicit notebook concatenation are verified, but the saved Top2a image's exact input chain remains unresolved. The original report map is available separately as an archived reference.

The GC manuscript also retains conditional Plasmodium controls, donor-level binding provenance, GC SHM interpretation and the RNA-centroid baseline's NP reporter readout. Author-template DOCX/RIS export contains 12 current images and 25 numbered references; template styles, media hashes and reference records are checked. Word pagination awaits visual review. The draft still awaits the existing RBD BiGCN recovery job and dependent finalizer for Figure 6/S7 and complete native comparison.

Validation: clean installed wheels on Linux Python 3.10/3.11 each pass 159 tests; Python 3.12 passes 158 with one existing expected failure in legacy repeatability. Dependency checks, tutorial and simulated quick start pass on all three. Seven installed modules match source/wheel bytes. Tests verify exact agreement with the notebook Scanpy/Leiden graph and AnnData serialization. The new GSE246382 run (Slurm 32344499) and compatibility matrix (32344702) completed successfully. Figure 2, Supplementary 8 and the standalone clone embedding were inspected, and current source image bytes were verified in the exported DOCX.

Review clone embedding, Figure 2, Supplementary 8, preset comparison, API instructions, manuscript and compatibility evidence.