a biologically inspired approach to emergent architectural intelligence
This is a living experiment. An organism that lives inside a market simulation and learns to survive through necessity. It starts with nothing but hunger and the ability to buy. Everything else, it has to earn.
The organism is a street cat. It trades EUR/USD on 1-minute bars. It has a Q-table with 18 states, a hunger drive, and a threshold for action. No indicators, no features from financial theory, no reward function designed to maximize Sharpe ratios.
I call this Emergent Architectural Intelligence. The intelligence emerges from architecture, not from scale. The biology is the inspiration. The architecture is the innovation.
The cat perceives two things: momentum (how many of the last 5 bars closed up) and volatility (percent change over 5 bars). That gives 18 states.
teal = above breakeven. pink = below. the cat learned this, not us.
It has one drive: hunger. The hungrier the cat, the lower the threshold for action. A starving cat trades anything. A fed cat demands conviction.
Abilities unlock through survival:
Q-values are floored at zero. The cat can forget, but it can't fear. Not yet.
Feed the cat or starve it. Watch how hunger changes its body and its willingness to trade.
We ran the cat alongside a random baseline. Same market, same spread, same TP/SL, same rules. Over ~130,000 bars of EUR/USD M1:
The edge comes from selective inaction. The cat trades half as often as random.
None of the following were programmed. They fell out of the interaction between hunger, a Q-table, and 18 states.
When hungry and Q-values are low, the cat trades from any state with equal probability. Hunger erases preference. In behavioral science, stress narrows decision-making to pure reactivity. Nobody coded this.
A bug caused learning to hit the wrong state. The cat developed 0.79 conviction about its worst state (23% WR, -$37.50 over 240 trades) because winning trades mechanically ended there. Like an animal returning obsessively to where it found food once.
one line fix. completely different organism.
Intelligence has a metabolic cost. The fixed cat feels its losses. Doubt leads to fewer trades, fewer wins, less food, more hunger.
Two runs, identical code, identical data. Different Q-table hierarchies from random choices when all Q-values were zero. Both converged on the same general insight but the specific favorite differed. Early experience shapes adult preference.
The cat develops burst-and-rest cycles. The hunger-threshold feedback loop creates an oscillator. Nobody programmed periodicity.
Across every run, the cat converged on the same insight: low-to-mid momentum states produce better buy entries than high momentum. It figured out mean reversion from hunger and a Q-table.
The most important discovery was a mistake. The Q-update was supposed to hit the entry state. Instead, last_state was overwritten every bar while holding.
- cat.last_state = current_state # updated every bar
+ # freeze at entry only
+ cat.last_state = current_state # inside open_position blockOne line. Q-values dropped from 0.79 to 0.09. Trade distribution restructured completely.
In emergent systems, broken perception produces confidently wrong behavior that looks like learning. You need independent validation of belief vs. reality.
Follow Maslow's hierarchy upward.
Next: lift the Q-floor to allow negative values. That's fear. A hungry cat near a dog will still hesitate. Right now it can't.
Then self-perception. Then emotion. Then memory modeled on neuroscience. Then deep analytical thinking through tree structures.
And eventually: a house cat raised on treats alongside the street cat raised on hunger. Same architecture, different upbringing. Same genetics, different personality.
This is a street cat with 18 states and hunger. Imagine what happens when it can feel fear.