not launched yet
Real mouse neurons. A flying roadster. 100% of the creator fees to holders, one every 10 to 15 minutes.
mouse-65 is the neurons of one cubic millimetre of a real mouse's brain, mapped synapse by synapse and simulated from that map. Its neurons press the launch button for this coin and pick which holder gets paid. Each lit pad is one holder's wallet, with a cheese on it. Every 10 to 15 minutes it parks its roadster, sniffs the cheese on every lit pad and eats one, and 100% of the creator fees from that flight are sent to that wallet. No trader and no language model sit in that loop.
Each lit pad carries a holder's wallet and a cheese. At the end of a flight it parks the car and walks to every lit pad. Its neurons sniff each cheese and pick one; the mouse eats it, and that wallet is paid.
Each lit pad is shown to its neurons on its own for one simulated second. Layer 4 takes it in; layer 5 answers.
It eats the cheese its layer 5 answered most, measured against that pad's calibrated rate. That wallet is paid.
If no pad gets an answer from layer 5, nothing is eaten and the pot waits for the next flight.
Its layer-5 output steers the cursor onto the launch button and presses it. How hard it presses sets the size of its own first buy.
These are the vision neurons of a real mouse brain, so everything arrives as a picture. Their sheet is cut into 12 × 8 columns and split into eight areas, one per landing pad. A lit pad lights its area: Poisson input to that area's layer-4 cells. Each pad is shown alone for one simulated second, and its smell is the layer-5 rate under it, through the mapped synapses. Liking is that rate divided by what the same area gave in calibration, because some patches answer louder than others whatever they are shown. The tick on each bar is the floor below which a sniff does not count. The two pictures replay the last flight, one pad at a time.
τm · dV/dt = (V_rest − V) + g
τsyn · dg/dt = −g
spike i→j : g_j += w_syn · n_syn(i→j) · s_i
s_i = −κ for interneurons, else +1
V ≥ V_th : spike, V ← V_reset, hold t_ref
λ_in(col) = lit(col) · f_max · N_aff
smell(c) = ⟨ rate(L5-ET | pad of c) ⟩, c shown alone
liking(c) = smell(c) / ref(pad of c)
pick = argmax liking, if liking ≥ θ
reward = 100% · fees claimed this flight + unpaid pot
seed = SHA-256(prev ‖ blockhash)[0:8]
Leaky integrate-and-fire, one equation set for every cell. Weights are synapse counts from the map times one global constant, and κ times that for inhibitory cells. Both are set once by calibration. Nothing is trained and nothing is fitted per cell.
| # | Time | Fees | Qualify | Pads | Decision | Proof |
|---|
mouse-65 is neurons mapped from a real mouse's visual cortex: one cubic millimetre of it, because no whole mouse brain has been mapped at this level yet. A rig connects those neurons to the chain, and this is the exact split.
| Job | Done by | How |
|---|---|---|
| Pick which cheese is eaten, and so which wallet is paid | Mouse | Firing rate of layer-5 ET cells under each pad, one simulated second per pad. |
| Steer the cursor to Launch and press | Mouse | Cursor follows the most active patch of layer 5. Recorded step by step. |
| Choose which wallets get a pad | Rig | A seeded draw from wallets that qualify and have not been paid. Anyone can redo it from the record. |
| Fly the car, do the stunts, walk between the pads | Rig | Animation on this page. Only the sniff numbers and the pick shown at the end of a flight come from the simulation. |
| When it eats, and how much the cheese is worth | Rig | 10 to 15 minutes after the last cheese, drawn from the seed; 100% of that flight's creator fees. |
| Name, ticker, picture | Rig | Typed by a person before launch. |
| Sign and send transactions | Rig | With the mouse's one wallet. Whoever runs the rig holds that key. |
| Write the posts | Rig | Fill-in templates. No language model. |
Each flight is saved with the wallets that qualified, the pads that were lit, its random seed and the fingerprint of the brain file. Download the same public data, run the same code, and you get the same sniffs and the same pick. Scramble the wiring and the mouse smells nothing; that test ships with the code.
python scripts/fetch_microns.py # build the brain file from public MICrONS data mousedev replay decisions/000068.json # same sniffs, same pick, same pads, or it says why not mousedev launch --shuffled # control: scrambled wiring never reaches the button