Fly Poker
Two-player 13-card shedding game vs a full-connectome fly sim in the browser.
Neurons are simulated — leaky integrate-and-fire or better — with documented, tuned parameters.
Dispute this rating →The evidence
No evidence, no level.LIF constants and the per-step decay/reset in the CPU kernel: rest -52 mV, exp(-dt/tau_m) and exp(-dt/tau_syn) factors, refractory counter.
https://github.com/HappyAny/fly-poker/blob/8eb739b3860ad9a03121a97c7bec763cb9b3a74f/client/brain_cpu_block.rs#L1-L60Card tracking, candidate search and a trained external readout are application code on top of a fixed connectome; the graph is a prepared FlyWire v783 export.
https://github.com/HappyAny/fly-poker/blob/8eb739b3860ad9a03121a97c7bec763cb9b3a74f/dist/model/README.txt#L1-L26The README states the opponent is encoding plus search plus a trained external readout, not a fly that understands cards.
https://github.com/HappyAny/fly-poker/blob/8eb739b3860ad9a03121a97c7bec763cb9b3a74f/README.md#L76
The browser kernel is a real leaky integrate-and-fire simulation of a prepared FlyWire FAFB v783 graph (138,639 nodes, 15,091,983 connections, derived from Shiu and Spiller's published model with pinned upstream revision and SHA-256s): membrane voltage decays toward -52 mV with the exact per-step factors for tau_m = 20 ms and tau_syn = 5 ms at dt = 0.1 ms, spikes reset to rest and enter a refractory count. Documented neuron model, so L1. Not L2: nothing writes the weight array - the graph is packed integer weights unpacked read-only in both the WGSL and Rust/wasm kernels, and the opponent's strength comes from application-side card tracking, candidate search and an external scorer calibrated against neural responses. The project itself states that the fixed connectome participates in every score but has not learned card play.
Record
Nearby
All projects →MaleCNS simulation connected to ViZDoom through modeled visual inputs and a fixed button readout. Includes plasticity experiments and negative validation results; learned survival has not been demonstrated.
Research framework comparing a MaleCNS-constrained sparse recurrent controller with rewired graphs, conventional neural networks, and other controls on simple VizDoom tasks.
Ratings on thelearningfly.com are proposals, not verdicts. Every one of them is arguable in public.