Sensor fusion over a fixed warehouse state
Readers, scanners and cameras report small changes to a large, fixed state: where an asset is, who is in a zone, what an item count is. Each event moves a few cells. Delta Routing updates only the derived quantities those cells feed.
What was measured
Six event mixes, each run for 1,500 ticks against a 40,000 cell state with 40% fill. Every tick, a batch of events changes a sparse set of cells. The baseline recomputes the whole derived result from the full state each tick.
| Event mix | Cells changed per tick | Speed against full recompute |
|---|---|---|
| RFID tag reads | 14 | 9.9 times faster |
| Barcode station scans | 21 | 6.8 times faster |
| Asset location updates | 28 | 5.7 times faster |
| Mixed warehouse | 43 | 4.0 times faster |
| Camera, light activity | 88 | 1.7 times faster |
| Camera, heavy activity | 241 | Full recompute is faster |
Every path agrees with the full recompute to a relative error near 1e-16.
Conditions
- Host: Apple M1 Max, arm64.
- Compiler: not recorded in the results file.
- Events: synthetic, built from the six warehouse event types and normalized into one update stream. Not a deployment trace.
- Baseline: full CSR recompute of the same derived quantities.
- Status: an archived kernel result, not a product qualification.
Where this applies
Use Delta Routing when most events touch a few tiles, such as tag reads, scans and location updates. When each event changes many cells, as in dense per-frame camera tracking, the full recompute is the better choice.