Pick the tissue
A segment of the Allen atlas reference series is chosen and registered to plate 100960084. The segment ID is written down before anything else happens.
MouseMindMouseMind replays recorded neural activity from a real, mapped mouse brain against a rule that was written and sealed before the run began. When the activity crossed the line, the system froze the evidence and executed the launch it had been trained to perform. No human step was required — the mouse brain triggered the token on its own.
The anatomy comes from the Allen Mouse Brain Atlas — the reference map of the mouse brain used across neuroscience. Beside this text is the actual Nissl-stained coronal section the experiment was registered to: plate 100960084, cut through the thalamus. Cell bodies show up dark; you can read off cortex, hippocampus, corpus callosum and thalamic nuclei by eye.
The 3D viewer above is built from the same atlas geometry: paired isocortex hemispheres, olfactory bulbs, cerebellum and brainstem, with the seven recording sites laid over the cortical surface. Hover a region to name it, drag to rotate.
This kind of mapping is an active scientific frontier: Google Research is working with the Allen Institute and others to map a full mouse brain at synapse resolution — MouseMind sits downstream of that effort, turning mapped neural activity into a deterministic, verifiable trigger.
Cellular counts and data compilation: BioNumbers ↗

The replayed brain drives a browsing agent: recorded activity is mapped to motor outputs, and every page it opens, clicks and reads is printed below alongside the live channel telemetry.
The whole point is that no step can be reordered or quietly edited after the fact. Each stage writes something the next stage cannot change.
A segment of the Allen atlas reference series is chosen and registered to plate 100960084. The segment ID is written down before anything else happens.
The trigger is written once and locked: the normalized score must exceed 0.82 for three consecutive readings. A digest of the rule text is stored, so any later edit is visible.
Recorded activity replays at a fixed 700 ms interval from a fixed seed. Every reading is stored with its index and timestamp, so the run can be replayed exactly.
The moment three consecutive readings clear the line, streaming stops and the exact rows responsible are frozen — no cherry-picking a nicer window afterwards.
Seed, segment, sealed rule and the triggering readings are serialized and hashed with SHA-256. Change any digit and the hash on the proof page stops matching.
Once the evidence is hashed, the trained agent submits the mint transaction itself. The rule was sealed beforehand, the signature wallet was fixed, and the mouse brain launched the token autonomously.
if score > 0.82 for 3 consecutive reads then launch_trigger = true
sealed 2026-03-14 02:11:40 UTC · digest 9f2c41b7ae30d6c8
fired 2026-03-14 02:17:59.200 UTC
40e1d1645e7514432e9c2d50aa13e0b1c44969bbb20f75bc9343dd6d810c8952
SHA-256 over seed, segment, rule and the three readings — recomputable on the proof page.
Seven cortical channels are sampled every 700 ms. Each window is band-passed, rectified and averaged across channels, then normalized against the quiet-baseline range recorded at the start of the session. The result is a single score between 0 and 1 — the only value the rule is allowed to look at.
Because the replay is seeded, anyone re-running the same segment with the same seed gets exactly the same score at exactly the same index. That is what makes the trigger reproducible instead of a story.
0x6a7cbf6c63f68ba69978991f19a925314ea3171d
The mint transaction was prepared on the server and signed by the wallet MouseMind was trained to use. The launch was the final step of the rule it was given — the token went on chain because the mouse brain crossed the threshold.