Publish full-brain CUDA odor gain and recovery experiment

[freeman-1984-coder]

Full-brain sensory input can produce an early side response that is obscured by cumulative counts and continuing downstream activity. Add a prespecified experiment comparing four contact strengths and both stimulus orders, retaining every FlyWire neuron and source edge.

All eight A16 runs completed. The lowest strength produced no downstream output; higher strengths retained new downstream spikes during recovery. The highest-strength right-first run had an early ipsilateral response followed by contralateral bias. These one-seed data do not establish a reliable navigation preset, and no default changed.

Publish the original report, hardware/test/source evidence, interpretation and an interactive comparison page. Explain the Or42b/DM1 input assumptions on the voxel page and distinguish historical dimensionless results from synaptic mV results. Update the roadmap to distinguish hardware-verified main from the still CPU-only tagged release.

Validation: the pre-run commit 433c77f passed five actual CUDA gate cases and all 217 hardware-host tests with zero skips. After retrieval, 69 Python file hashes matched that commit; group counts and metrics were independently recomputed. A subsequent CPU mirror-world test also passed, ruling out a fixed left bias in antenna/body geometry. Browser checks covered missing-data failure, all eight table entries and condition/population switching; raw-report SHA256 is e72b30c19b76efb3a7f92227519a65ee68629690f216b35dfd6fdbe7170c011f. The rented test server was destroyed after evidence retrieval.