Repo: abrownfox001-twap60-prediction-trigger-system
Most people treat Polymarket’s BTC 5-minute Up/Down markets as pure noise. Books sit near 50¢. Outcomes look random.
I built a live system under @abrownfox001 that treats them as a short-horizon forecasting problem instead.
Wallet: 0x12b712029efeee7efc4066e98e4e1b6beb8b1b0b
Active since: Jan 12, 2026
This post is a second, more operational view of the same strategy: what the engine does every slot, why the mid-band matters, and where the real edge lives.
Different Game Than Convergence Farming
| Typical 94–99¢ farmer | abrownfox001 | |
|---|---|---|
| Zone | Near-certainty | 45–55¢ coin-flip band |
| Edge | Structural certainty | Directional micro-forecast |
| Volume style | Huge clips, tiny margin | Small clips, higher quality |
| Exit | Hold almost everything | Scratch weak, redeem strong |
Public resolved sample (150 markets near 50¢ entry):
- Held-side win rate: 60.7% (91W / 59L)
- ROI on capital at resolution: ≈ +24.5%
- Average held entry: ~0.50
That is not “always right.” It is a filtered book: only higher-conviction legs are allowed to reach settlement.
The Operating Loop (Every 5m Slot)
1. Stream latest Chainlink TWAP (settlement reference)
2. Lock slot-open TWAP as price-to-beat
3. Compute private P(up) vs that path
4. Enter only if edge is real AND ask is in 45–55¢
5. Re-score every 10–30 seconds
6. Scratch near flat if conviction dies
7. Otherwise hold → redeem winner at $1
Why TWAP, not CEX last price
Polymarket settles short crypto markets on Chainlink TWAP, not a single exchange tick.
A bot that forecasts Binance mid while the market pays out on a 60-second average is solving the wrong problem.
Why mid-band entries
At 99¢, fees and tiny edges dominate.
At ~50¢, the binding constraints become signal quality, latency, and fill discipline. That is the design target.
Why scratches matter as much as entries
In a large public window, ~930 markets show buy and sell at almost the same price (0.504 → 0.503). Those are not profit-takes. They are conviction culls.
Without that layer, a 55–58% raw signal still bleeds. With it, the settled book can look closer to ~61%.
Observed Footprint
From a ~5,000-trade public sample:
| Metric | Value |
|---|---|
| BTC share | ~99% |
| 5-minute share | ~96% |
| Fills in 45–55¢ | ~90.6% |
| Buy / Sell mix | 72.9% / 27.1% |
| Median clip | ~$58 |
| Median buy time | ~47s after open |
| Median sell time | ~226s after open |
Buys happen earlier to source fair liquidity.
Sells happen later as the slot matures and weak theses are cut.
Economics in One Formula
For entry near ( p = 0.50 ) and held-side win rate ( w ):
[
EV \approx 0.50 \times (2w - 1)
]
At ( w = 0.607 ):
- Naïve EV ≈ +10.7% of stake per resolved ticket
- Sample ROI can print higher because winners are often accumulated a bit under 0.50 and losers are removed before full loss
Fees are secondary in this band. Process is primary.
Architecture (Reference Implementation)
TypeScript / Node 18+ modules in the repo:
| File | Job |
|---|---|
twapFeed.ts |
Chainlink TWAP via RTDS |
signal.ts |
P(up) vs open reference |
engine.ts |
enter → hold → scratch → redeem |
cexFeed.ts |
auxiliary lead-lag |
markets.ts |
discover btc-updown-5m-*
|
clob.ts |
CLOB v2 execution |
Also shipped: standalone Windows binary (Abrownfox001.exe) with local setup prompts and a 2-day trial.
Important: the calibrated signal stays private. The repo documents the archetype and execution shell — not a drop-in copy of the live edge.
What This System Explicitly Avoids
- 94–99¢ convergence sweeps as core strategy
- Hold-everything-to-resolution without culling
- Multi-asset spray across SOL/ETH/politics
- Snapshot-close models that ignore TWAP settlement
- Oversized early entries that pay through thin books
Specialization is a feature. Breadth is optional later, not required for the core loop.
Main Failure Modes
- No real signal → 50¢ trading becomes coin-flipping minus costs
- Slow path → adverse selection on the wrong side of 50¢
- Regime shift → old thresholds stop mapping to new volatility
- Thin liquidity → “breakeven scratch” becomes real spread loss
- Wrong settlement target → model still aims at 30s/snapshot while markets pay 60s TWAP
After the August 2026 update, 5-minute crypto markets settle on a 60-second Chainlink TWAP. Any live system must stay locked to that series for both open reference and final path.
How to Think About Replication
You can rebuild the shell:
- Stream the official TWAP
- Pin open reference
- Score continuously
- Enter only in fair band
- Scratch decaying positions
- Redeem winners
You cannot download the alpha. Without a genuine short-horizon edge, the shell is just disciplined coin-flipping.
Bottom Line
abrownfox001 is built around one idea:
Forecast the official TWAP path, buy it only near fair value, and refuse to carry weak tickets to resolution.
That combination — TWAP alignment + mid-band execution + active culling — is the whole strategy.
Not financial advice. Short-horizon markets are high-risk. Paper or micro-size first.
Repo: https://github.com/abrownfox0/abrownfox001-twap60-prediction-trigger-system
If you have more questions, please feel free to contact me at any time: https://t.me/abrownfox001
My Polymarket Activity: https://polymarket.com/@abrownfox001?tab=activity
#Polymarket #BTC #TWAP #TradingBot #AlgoTrading #Chainlink #PredictionMarkets #QuantTrading
Top comments (0)