A trader quit his job at Citadel after his Polymarket bot turned $313 into serious six figures. Then the market regime changed — and the bot kept doing exactly what it was built to do. It lost $200,000 in four days.
He rebuilt the entire system overnight with Claude Fable 5 and is now up $300,000 since.
Here’s what actually happened, and why it matters for anyone building (or using) automated trading systems on prediction markets.
The Old Bot: A “Loaded Coin”
The original bot didn’t try to predict the future. It exploited a small, persistent statistical edge — something like 54/46 instead of 50/50.
It was essentially playing a loaded coin thousands of times. In a stable market environment, this works extremely well. Small edges compound beautifully when repeated at high frequency.
The problem: Markets are not static.
When the underlying regime shifted (liquidity, volatility, whale behavior, news flow, etc.), the edge disappeared. But the bot had no mechanism to notice or adapt. It kept pressing the same side of the now-unloaded coin.
Result: Rapid, mechanical drawdown.
The New Bot: Probability Shapes, Not Single Numbers
The rebuilt bot (powered by Claude Fable 5) operates on a fundamentally different level.
Instead of asking “What’s the probability that BTC closes above $X?”, it models the full distribution — the entire shape of possible outcomes and how that shape evolves in real time.
Key differences:
- It doesn’t just take the market’s single probability number.
- It builds its own richer probability distribution.
- It hunts for outcomes the crowd heavily discounts (the “far slopes” that most people round to near-zero).
- When its model sees a longshot trading at a massive discount to its calculated value, it buys.
One example in the story: A position the market priced at ~0.2% that his model valued at 1.6%. It bought and later one of those paid out 81x.
Why This Matters
| Old Bot | New Bot (Fable 5 powered) |
|---|---|
| Static edge (loaded coin) | Adaptive probability distribution |
| Single number probability | Full shape + dynamics |
| Fragile to regime changes | Can adapt when the map redraws |
| High-frequency small edges | Mix of micro-edges + occasional fat tails |
| No real “understanding” | Models how the market thinks |
The biggest lesson: A bot that can’t update its model of reality will eventually get destroyed when reality changes.
Even a well-designed, previously profitable system becomes dangerous without adaptation.
The Takeaway for Builders
- Edges decay — Especially in transparent, high-competition environments like Polymarket short-term crypto markets.
- Adaptation is the new alpha — Static rules eventually fail. Systems that can redraw their internal map win.
- AI as a reasoning layer — Using advanced models not just for code generation, but for live probabilistic reasoning and distribution modeling is becoming a real differentiator.
- Risk management still matters — Even the best model needs proper position sizing and kill switches when the regime shifts.
The story is a perfect illustration of why purely mechanical bots (no matter how clever the original edge) eventually hit a wall, while more adaptive, model-driven systems have a fighting chance.
Markets don’t stay the same. The bots (and humans) that survive are the ones that can notice when the table has changed — and update accordingly.
If you have more questions, please feel free to contact me at any time: https://t.me/FatherSon97
Top comments (0)