Connect FlyWire MN9 activity to a physical proboscis response

[bitdeep]

Sugar stimulation now propagates through the full FlyWire 630 graph and produces a physical flybody rostrum response from the two MN9 motor neurons. Neural and native MuJoCo steps share a 0.1 ms clock. An explicitly documented firing-rate adapter controls the existing rostrum servo; all other actuators remain disabled.

The observatory adds a measured brain-to-body workflow, spike/drive/joint traces, a head-focused camera, baseline and blocked-link comparisons, and pause/cancel controls. Trials start after natural physical rest and retain the existing CPU/RAM budgets. This is a direct-input motor experiment with an engineered adapter; environment sensing, neural walking and neural flight remain future work.

Validation:

• Full-graph causal comparison: MN9 fires 30/17 spikes with sugar input; the joint moves 40.03°. Blocking the motor link preserves those spikes with zero force and movement. No-input baseline is also zero.
• The first motor force follows the first MN9 spike at 25.5 ms simulated time. Both clocks advance 500 ms, excluded actuators produce zero force, and no physics warnings occur.
• Checks cover native wake, pre-spike immobility, concurrent-trial rejection, pause/resume including gravity controls, cancellation, delta telemetry and clock-mismatch failure.
• All seven existing physical/transport checks pass. JavaScript and Python syntax checks pass, and the motor report matches the current source hashes.
• Browser checks exercise all modes, head focus, shared-clock pause/resume, cancellation and the retained antennal assay. Desktop/mobile layouts have no horizontal overflow; all 295 captured requests return HTTP 200 with no JavaScript exceptions. A software graphics-context reset during resizing recovers.

README, product metadata and the browser/neural/motor guides describe the implemented behavior and its limits in English.