05 Integration — the spec the human revised most

The integration spec started at 158 lines and grew to 488 lines — a 3× expansion. It was restructured twice (once shrinking after a split, then growing again as new methodology was incorporated). This volatility reflects the hardest scientific decision in the pipeline: how to integrate 12 datasets from different platforms without erasing biologically meaningful variation. Every revision here was a human judgment call.

Feb 26 — Initial
158 lines. Placeholder integration strategy, single workflow assumed.
Mar 9 — +185 lines
Restructured to include coarse annotation anchors for scANVI. Two-stage approach introduced.
Mar 10 — −127 lines
Split into 3 specs. Integration content extracted; clustering and annotation moved to dedicated modules.
Mar 20 — Multiple commits
Three integration workflows (CCA, scANVI, STACAS) specified. API updated to Seurat v5 IntegrateLayers.
Apr 9 — +197 lines
Configurable CCA parameters and parameter sweep plan added after first integration results reviewed.
Apr 13 — +63 lines
NP integration quality experiment spec added. Pipeline now supports per-compartment integration strategies.