---

name: polymarket-trader description: Build, evaluate, and tune a Polymarket BTC 1h Up/Down trading strategy using Binance (resolution source) as the anchor. Use when: (1) designing a mispricing/edge model (fair probability vs market price), (2) adding regime filters (trend vs range), (3) debugging bad entries/exits from events.jsonl/state.json, (4) running quick offline analysis or parameter sweeps with the bundled scripts.

Polymarket Trader

Maintain a profitable BTC 1h Up/Down strategy by anchoring decisions to Binance BTCUSDT (the resolution source) and enforcing anti-churn/risk rules.

Workflow (use this order)

1. Confirm the market type

• This skill is optimized for bitcoin-up-or-down-* 1h markets (Binance 1H open vs close).

2. Compute the anchor signal (Binance)

• Fetch 1m closes + the 1h open for the relevant hour.
• Compute volatility (sigma) and time-to-expiry.
• Convert to fair probability for Up/Down.

3. Trade only when there is measurable edge

• Enter only if edge = fair_prob - market_price exceeds a threshold.
• Add a directional guardrail: do not bet against the sign of the move when |z| is non-trivial.

4. Exit using the right logic for the entry mode

• Model entries: exit on edge decay / model flip; hold to preclose when confidence is extreme.
• Mean-reversion entries: exit on reversion targets (not model-tp), with strict churn limits.

5. Validate with logs

• Every suspected “nonsense trade” must be explained via:
  • reason / entry_mode
  • Binance-derived fair probability + z
  • whether the correct exit block fired

Bundled scripts

All scripts are designed to be run from the OpenClaw workspace.

1) Fetch Binance klines

• {baseDir}/scripts/binance_klines.py
  • Pulls klines and prints JSON.

2) Dump/stabilization and regime metrics

• {baseDir}/scripts/binance_regime.py
  • Computes ret5/ret15/slope10 + simple “stabilized” boolean.

3) Explain fills (events.jsonl) with Binance context

• {baseDir}/scripts/explain_fills.py
  • Reads paperbot events.jsonl and prints a concise table for the last N fills:
    • side/outcome/px/reason
    • estimated fair_up + z
    • “against trend?” flag

References

• {baseDir}/references/strategy.md — the math model, parameters, and tuning checklist.