LeChatonFat is an automated directional system that trades Polymarket crypto Up/Down 15-minute markets (primarily BTC, with ETH and SOL support).
Public profile: polymarket.com/@lechatonfat
Proxy wallet: 0x9629804e001a60ebe8a289a334ccf2e5553aee15
It is designed for sub-hour horizons, taker-heavy execution, and incentive-aware economics.
Core Idea
Fifteen-minute crypto direction is close to a coin flip at the settlement horizon. The book usually sits near 50¢, so large structural mispricings are rare.
Instead of chasing big probability gaps, the system:
- Builds a low-latency fair P(Up) from Polymarket RTDS Chainlink TWAP
- Compares that fair value to the live CLOB price
- Buys the underpriced leg when the edge exceeds a threshold
- Scales in only on confirmation (pyramid with the move)
- Holds to resolution and collects rebates / volume rewards
The edge is deliberately small (~4 percentage points over breakeven entry). Everything else (pyramiding, rebates, capital recycling) exists to monetize that thin gap at scale.
Signal Layer
- Stream Chainlink TWAP (30s / 60s) via Polymarket RTDS
- Pin a window-open TWAP reference at each 15m epoch
- Map the move from reference → live TWAP (plus short momentum) into a fair P(Up)
- Trade when the book underprices that fair value
This aligns the signal with the same oracle family Polymarket uses for resolution.
Execution Style
| Aspect | Behavior |
|---|---|
| Markets | BTC / ETH / SOL 15m Up/Down |
| Side | One directional thesis per window |
| Entry | Taker FOK clips on the underpriced leg |
| Sizing | Scale-in only when price confirms the thesis |
| Hold | Almost pure hold-to-resolution |
| Capital | Continuous profit sweep keeps wallet light |
Sample behavior (18-day public activity window):
- Median clip ~$12, mean ~$22
- ~5.9 buys per window
- ~99.8% buys (very few discretionary sells)
- Avg stake on winning windows roughly 2× the stake on losing windows
Why Confirmation Pyramiding Matters
Early versions that simply added size on a timer suffered classic averaging-down behavior.
The current rule only adds when short-horizon price action moves in favor of the held side. This produces positive skew:
- Winners receive more capital
- Losers are starved of additional size
That asymmetry is a major contributor to the observed profit factor (~1.13 in the sample).
Economics & Incentives
The system is deliberately taker-heavy. Crossing the spread is the cost of acting on a fast-decaying signal.
Two offsets keep it viable:
- Entering the cheap leg (often below 50¢)
- Treating taker rebates + volume rewards as first-class inputs to edge calculation and sizing
In the sample period, incentives contributed roughly 37% of net PnL.
Hard Problems & Fixes
| Problem | Root Cause | Solution |
|---|---|---|
| Wrong-window orders | Clock / slug epoch races | Single slot clock + consistency checks |
| Stale fair value | Slow REST vs live book | Stream RTDS TWAP, recompute on every tick |
| Averaging down | Time-based DCA | Confirmation-gated pyramiding |
| Spread eating the edge | Backtest on mid prices | Use actionable side prices + net-edge model |
| Duplicate clips / nonce issues | Retry without idempotency | Intent IDs + leader election + reconciliation |
Performance Snapshot (18-day sample)
- ~784 distinct 15m windows (~56/day)
- Trading PnL ≈ +$5.4k
- Incentives ≈ +$2.0k
- Net ≈ +$7.3k
- Share win rate ~50% vs ~46% breakeven entry
- Profit factor ~1.13
These are reconstructed public-activity numbers, not audited lifetime results.
Takeaways for Builders
- On near-efficient short-horizon markets, a small calibrated edge + disciplined risk shape can be more durable than chasing large probability gaps.
- Aligning your fair-value signal with the actual settlement oracle (Chainlink TWAP via RTDS) is critical.
- Confirmation-based pyramiding is a powerful way to create positive skew without needing to predict which windows will win.
- Rebates and volume rewards must be modeled as part of the edge, not treated as free money after the fact.
- Clock discipline, signal freshness, and idempotent execution are non-negotiable on these markets.
The system is available both as a ready-to-run Windows build (WinnerLockBot.exe) and as open TypeScript source for developers who want to study or extend the architecture.
Top comments (0)