ON-CHAIN SNN AGENTTHIRD GENERATION

cluck

n. /klʌk/ · agent, spiking

An autonomous on-chain agent running a Spiking Neural Network, the third generation of neural network models. Not a transformer. Not an LSTM. Neurons that fire discrete spikes, reading live Chainlink price feeds and making trading calls through spike-rate accumulation.

cacoming soon

~1,600 LIF neurons · spike-rate decision model · 8 stocks · paper money · not financial advice

specimen 0125 × 19 pxLIF · 1,627

balance: $10,000.00
since Sep 11: +0.00%
position: roosting, nothing held
last heard: 13:13 UTC
spikes

↓ the mind

NVDA

$218+4.9%

AAPL

$326+2.6%

TSLA

$363+1.5%

MSFT

$489+1.8%

GOOGL

$332+4.1%

AMZN

$251+2.0%

META

$644+1.7%

SPY

$757+1.3%

8 SNN input channels, one per stock. Each ribbon is the live spike-voltage waveform of that neuron layer. Token badges track wave peaks in real time. Click one to isolate.

real · Chainlink price feedsours · SNN ribbon oscilloscope

REAL · MICrONS cortical connectome (1mm³)
OURS · 1,627 neurons in cluck

Drag to rotate. Hover a neuron to inspect.

The coop. Cluck is ours. The cube in its head is a cubic millimetre of cortex: 1,627 real neurons, one per dot. Open it and the neurons rearrange into the brain they came from.

$0.00

since entry+0.00%

positions0 / 8

cash100%

calls made0

Paper money. Protocol launched Sep 11 with $10,000 and is +0.00% since. Silence is the network default: a spiking neural network at rest fires below threshold most of the time.

stock	shares	value	since entry
Roosting. Nothing held. Cash: $10,000.00.

signal log		live
13:13 UTC	quiet	no prices heard yet on this run

One line per event, never edited. Machine-readable: ledger.json.

When Robinhood Chain publishes a new price for one of eight stocks, cluck opens a 60-second integration window. A price rise excites the input layer: neurons that receive the incoming signal. A price fall activates inhibitory neurons wired onto the same output population, suppressing the response. A 0.5% move drives them at 4× resting input current.

After the window, cluck reads the output layer spike rate and compares it to baseline: the median of the last ten windows for that stock, never lower than the network's resting rate. At 1.5× baseline or above, it enters. At 0.67× or below, it exits. Between those values: nothing. A second price inside the same window is dropped. Baseline needs ten windows to exist, so the first ten prices for any stock build baseline only: cluck may not act on a stock until it has heard ten.

In symbols: input current I = I₀ · (1 + 6 · |Δp|), Δp in percent. Rate r = spikes / 60 s. Baseline b = max(median of r over the last ten windows, r at rest). Decision ρ = r / b: enter at ρ ≥ 1.5, exit at ρ ≤ 0.67. Buy size = min(0.10 · cash, 0.25 · portfolio − position).

SPY never reaches 1.5×, not by configuration, but by network topology. Its output neurons happen to be wired such that the signal never amplifies enough to cross threshold. This is the kind of emergent property that only appears when you run the actual network.

pecked: Output spike rate ≥ 1.5× baseline. Buys 10% of available cash. Max 25% per stock.
chickened out: Output spike rate ≤ 0.67× baseline. Sells full position.
roosted: Between thresholds. Network fires but stays sub-threshold. No action.

stock	price	neurons wired	full rise	full fall	last call
NVDA	$218.51	1,677	4.89x	0.49x	none yet
AAPL	$326.41	984	2.63x	0.60x	none yet
TSLA	$363.88	1,290	1.53x	0.00x	none yet
MSFT	$489.26	1,000	1.80x	0.60x	none yet
GOOGL	$332.00	868	4.06x	0.00x	none yet
AMZN	$251.70	185	2.01x	0.65x	none yet
META	$644.00	76	1.68x	0.57x	none yet
SPY	$757.55	62	1.30x	0.23x	none yet

Prices: Chainlink on Robinhood Chain real · on-chain. Rise/fall multipliers from a test run at 0.5% move. Blue ≥ 1.5× threshold would trigger entry; red ≤ 0.67× would trigger exit.

Most of what gets called "AI" in trading is a second-generation neural network: transformers, LSTMs, gradient-descent optimized weight matrices. Cluck runs on something older and, in some ways, more honest: a Spiking Neural Network (SNN), the third generation of neural network models, as defined by Maass (1997).

Where second-generation networks pass continuous activation values between layers, SNNs communicate through discrete spikes: a neuron either fires or it does not, at a precise moment in time. The decision-making mechanism in cluck is spike rate: the ratio of observed output firing to historical baseline, a principle grounded in how biological systems have been shown to integrate evidence and reach decisions (Gold & Shadlen, 2007).

The language of markets already runs on chickens. People chicken out. They mind the pecking order. They sit on nest eggs. We stopped using it as metaphor and built the literal version: a chicken, running an SNN, reading a live price feed, making calls in the open.

If a 1,600-neuron spiking network with one rule trades as well as the median fund charging you fees, that is worth knowing. If it does not, that is also worth knowing. Everything is in the log.

The model. real · published research Cluck runs a Leaky Integrate-and-Fire (LIF) spiking neural network, the standard computational model in theoretical neuroscience. Each neuron integrates incoming charge, leaks over time, and fires a spike when membrane potential crosses threshold, after which it enters a refractory period. The mathematics follow Gerstner & Kistler (2002), with constants calibrated from Izhikevich (2003): 20 ms membrane time constant, 5 ms synapse, −52 mV rest, −45 mV threshold, 2.2 ms refractory.

The decision rule. real · neuroscience Output neurons vote by firing rate. The buy/sell threshold is derived from spike-rate accumulation, the same mechanism Gold & Shadlen (2007) identify as the neural substrate of decision-making across species. When the ratio of current to baseline spike rate exceeds 1.5×, cluck enters. When it falls below 0.67×, cluck exits.

Ours. ours · made by us The chicken. Which neurons listen to which stocks. The network topology. The on-chain execution layer. The synaptic weights, scaled up because the modelled network is sparse relative to biological density. The paper money: no real wallet, no real orders.

Research

Maass, W. (1997). Networks of spiking neurons: The third generation of neural network models. Neural Networks, 10(9), 1659–1671. doi:10.1016/S0893-6080(97)00011-7. Foundational paper establishing SNNs as the third generation of neural network computation, surpassing perceptron and sigmoidal models in computational power.
Gerstner, W., & Kistler, W. M. (2002). Spiking Neuron Models: Single Neurons, Populations, Plasticity. Cambridge University Press. doi:10.1017/CBO9780511815706. The standard reference for LIF neuron dynamics, membrane equations, and population coding used in cluck's core simulation.
Izhikevich, E. M. (2003). Simple model of spiking neurons. IEEE Transactions on Neural Networks, 14(6), 1569–1572. doi:10.1109/TNN.2003.820440. Provides the biologically-plausible parameter set (membrane time constant, spike threshold, refractory period) that cluck uses for each LIF neuron.
Gold, J. I., & Shadlen, M. N. (2007). The neural basis of decision making. Annual Review of Neuroscience, 30, 535–574. doi:10.1146/annurev.neuro.29.051605.113038. Establishes spike-rate accumulation as the biological mechanism for perceptual and value-based decisions, the same principle behind cluck's 1.5×/0.67× thresholds.
Sezer, O. B., Gudelek, M. U., & Ozbayoglu, A. M. (2020). Financial time series forecasting with deep learning: A systematic literature review. Applied Soft Computing, 90, 106181. doi:10.1016/j.asoc.2019.106181. Comprehensive review of neural architectures applied to financial markets, the landscape cluck intentionally sidesteps by using biological spike-rate rather than gradient-optimized weights.

Every run is preserved. The portfolio never resets without a dated entry here explaining why.

run	started	ended	end value	note
1	Sep 11, 2026	running	pending	cluck online. SNN warm. waiting for price signal.

The signal log above is the whole product. When the output layer fires above threshold, cluck posts once on X in the same format as the log, nothing else. No commentary, no threads. The ledger is public and machine-readable at ledger.json; we do not edit it.

About the address at the top: it is the only one. Anything else is not us. The token is not an input to the network; it does not change the topology, the thresholds, or the prices the neurons integrate.