Spatial
Each one takes the idea a recent paper is actually arguing and turns it into a control. Not an illustration of the science, a piece of interface that does a job which is hard to do any other way, and which happens to be the job the paper cares about.
All five are original rather than ports, which is the documented exception to the rule in AGENTS.md section 2. Every file header says so. Everything drawn here is generated: no reconstruction, no measurement, and no value taken from any dataset. They are interface primitives that borrow a way of thinking, not visualisations of results.
One
Job: a value that only means something in its place
Scrub the stage and the form folds. The expression band is defined in the form's own coordinates rather than the canvas, so it stays on the same tissue and travels with it. A heatmap on a fixed image can get away with being a texture. The moment the form moves, the field has to be attached to the thing, and the difference becomes obvious in one drag.
After 3D spatial transcriptome atlases of early primate embryogenesis, Nature Cell Biology, 2026, where a programme is only meaningful in the piece of geometry it occupies and at the stage it occupies it.
Two
Job: compare layers without losing registration
Drag the divider. The geometry channel runs the full width and the molecular channel is clipped to one side, both drawn from one set of coordinates. Because the shapes never move as the divider passes, you can see they are the same tissue. A crossfade could not tell you that: fading looks identical whether or not two layers are registered, which hides the one property worth checking.
After light microscopy connectomic reconstruction of brain tissue, Nature, 2025, where the appeal is a wiring diagram that carries molecular identity instead of trading one for the other.
Three
Job: jump six orders of magnitude and stay oriented
Pick a cell by what it does and the view travels to what it is made of. The ladder is the component worth keeping: any interface that jumps between very different magnifications leaves people lost, because the destination gives no clue how far it sits from the origin. A rung per decade and a marker that travels rather than teleports costs almost nothing and answers the question the jump creates.
After correlative voltage imaging and cryo-electron tomography, Nature Communications, 2025, which links an electrical phenotype to the nanometre scale architecture of that same class of cell.
Four
Job: show what a summary throws away
Press collapse. Every point keeps its identity through the morph, so the eye follows individual projections into the grid and watches them pile into a single cell. Two pictures side by side would make the same claim and prove none of it. The morph is the argument, and it is the reason this is a component rather than a pair of charts.
After whole brain reconstruction of single neurons, Nature Methods, 2025, where the questions are about the trajectories that an adjacency matrix does not keep.
Five
Job: prove a set of layers is one object
Close the stack up, then switch registration off. Every section is unchanged and the structure running through the depth disappears, which is a stronger statement than any caption about alignment. When a set of layers really is one object, the control that can take the alignment away is what demonstrates it.
After 3D reconstruction of serial spatial transcriptomic sections, 2026, where the third dimension changes the inference rather than the presentation.