In the previous article we covered the high-level strategy behind @abrownfox001: a specialized directional system built for Polymarket’s BTC 5-minute Up/Down markets and explicitly aligned with Chainlink TWAP settlement.
This second article goes one layer deeper into the technical design — how the signal is constructed, how each 5-minute slot is managed, and how execution and risk controls are structured.
1. Design Goal Recap
The system is not trying to predict a single last-second price tick.
It is trying to answer a more precise question:
Given the path of the 30-second Chainlink TWAP so far, what is a calibrated probability that the final TWAP settlement will finish above the slot-open reference?
That framing drives every major design choice.
2. Signal Construction
Inputs
- Live Chainlink 30s TWAP via Polymarket RTDS
- Slot-open TWAP reference (pinned at the start of the 5-minute window)
- Short-horizon TWAP momentum / path features
- Time remaining in the slot
- Live CLOB prices for the Up and Down tokens
Core Transformation
At a simplified level:
twap_ref ← TWAP at slot open
twap_now ← current 30s TWAP
features ← path + momentum + time remaining
P(up) ← calibrated_model(twap_now, twap_ref, features)
The exact model is private, but the architecture is deliberate:
- The primary price series is the same TWAP family used for settlement
- The model is recalculated continuously as new TWAP updates arrive
- If the feed is stale, the system prefers to skip rather than trade on old data
This avoids the classic failure mode where a bot’s fair value lags the live market.
Edge Definition
The bot does not trade every time P(up) differs from 0.50.
It requires the model edge to clear a threshold after considering:
- Current book prices
- Expected fees
- Typical fill quality in the 45–55¢ band
Only then does it consider entering.
3. Slot Lifecycle (State Machine)
Each BTC 5-minute market is treated as an independent session with a clear lifecycle:
1. Slot detected / validated
2. Pin twap_ref at open
3. Stream live TWAP and recompute P(up)
4. Evaluate entry conditions
5. If edge is sufficient → begin accumulation in 45–55¢ zone
6. Continuously re-score conviction
7. Scratch if signal decays
8. Hold if conviction remains
9. Resolve and redeem
10. Log results and free capital
Why a Explicit Lifecycle Matters
Without a strict per-slot state machine, short-horizon bots commonly suffer from:
- Orders placed on the wrong epoch
- Signals bleeding across slot boundaries
- Orphaned positions
- Inconsistent logging
The engine treats each 5-minute window as a closed world with its own reference price, signal path, and risk budget.
4. Entry Style: Mid-Band Accumulation
Instead of chasing late momentum or trying to buy near-certain outcomes at 95¢+, the system prefers the fair mid-band (roughly 45–55¢).
Characteristics:
- Typical clip size in the $50–175 range
- Often multiple fills across the slot rather than one large order
- Focus on getting a reasonable average price while the model edge is still valid
This has two advantages:
- Fee sensitivity is lower than extreme-priced convergence trades
- More room for the probabilistic edge to express itself before settlement
The bot is not trying to win a huge payout on a long shot. It is trying to win a modest edge repeatedly on near-even markets.
5. Scratch Logic: The Real Risk Engine
One of the most important differentiators in this architecture is active culling.
Not every entry is held to resolution.
The system continuously re-evaluates whether the original thesis is still intact. If conviction decays, it prefers to scratch near breakeven rather than hope.
This produces a filtered set of positions:
- Weak trades are removed early at low cost
- Stronger trades are allowed to reach settlement
Public sample metrics that show ~60.7% win rates are measured on the held book, not on every temporary position. That distinction is critical. Without the scratch layer, the same raw signal would look much closer to a coin flip after costs.
6. Execution Constraints
Even a good signal dies under poor execution. The engine therefore treats microstructure as a first-class concern:
- Prefer actionable prices over mid-price fantasy fills
- Respect current CLOB rate limits and back off cleanly
- Avoid over-aggressive accumulation that damages the average entry
- Keep per-slot exposure bounded
- Reconcile open orders and positions so the state machine does not drift
Because the strategy is specialized on BTC 5-minute markets, the execution path can be tuned tightly for that one product instead of being generalized across many incompatible market types.
7. Capital and Redemption Flow
The preferred end state for a successful trade is simple:
- Hold to resolution
- Redeem winning shares
- Recycle capital into the next slots
Idle capital sitting in resolved but unredeemed positions is treated as operational drag. Clean redemption and capital recycling are part of the loop, not an afterthought.
8. Why Specialization Helps
Many Polymarket bots try to cover politics, sports, crypto, and long-dated events with one framework.
@abrownfox001 deliberately does the opposite:
- One primary market family (BTC 5-minute)
- One settlement definition (30s Chainlink TWAP)
- One execution style (mid-band taker accumulation + scratch)
- One risk philosophy (filter hard, hold selectively)
Specialization reduces the number of regimes the system must handle and makes it easier to keep the signal, execution, and settlement logic coherent.
9. Practical Lessons From This Architecture
- Align the signal with settlement. If the market resolves on TWAP, forecast TWAP.
- Treat each slot as its own world. Boundary errors are expensive.
- A scratch layer can be more valuable than a slightly better raw model.
- Mid-band entries change the fee and edge math compared with extreme-priced convergence.
- State management and feed freshness are not optional details — they are part of the alpha preservation system.
10. What This Article Does Not Claim
This is a description of an architecture and a public strategy profile. It is not a guarantee of future results.
Short-horizon prediction markets remain high-risk. Edges decay. Execution quality varies. A private calibrated model can stop working if market behavior changes.
The useful takeaway is structural: under TWAP settlement, a coherent directional engine built around the official averaging window is a more rational design than systems still optimized for last-tick games.
Summary
The @abrownfox001 engine is best understood as a closed loop:
TWAP path → calibrated probability → selective mid-band entry → continuous conviction checks → scratch or hold → redeem
Everything else in the stack exists to keep that loop fast, consistent, and aligned with how Polymarket actually settles BTC 5-minute markets today.
In the next article we can go further into live operations, failure modes, and how this specialized design compares with broader multi-market or copy-trading approaches.
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
Relevant Hashtags/Tags:
#Polymarket #BTC #TWAP #TradingBot #AlgoTrading #QuantTrading #PredictionMarkets #Chainlink
Top comments (0)