| name | orallexa-risk-sizer |
| description | FinPos-style position sizer for trading decisions. Takes an upstream Judge direction (long/short/no_op), a bankroll, a volatility regime, and Kelly parameters — returns a fund/skip verdict with a Kelly-cap-respecting, drawdown-adjusted position size. Never emits a direction. Ports from Orallexa's Python engine to a callable skill inside Claude Desktop / Cursor / OpenCode. Pairs with any upstream direction source: a fundamental analyst, an ML model, a discretionary trader, or Orallexa's own 5-voice debate.
|
| version | 1.0.0 |
| author | Alex Xiaoyu Ji |
| scope | trading:size |
| depends_on | ["orallexa engine/risk_sizer.py (upstream Python reference)","Byte-identical to Shadow trader-pack risk-sizer.js"] |
Orallexa Risk Sizer
Standalone position sizing for a trade thesis. Direction is an input, not an output.
When to invoke
The user's request combines all three of:
- A specific trade thesis with a stated direction — long, short, or no-op.
- A bankroll amount and volatility read on the underlying (low, medium, high).
- A calibrated win-rate estimate and average win/loss magnitudes (Kelly parameters).
If any of the three is missing, ask for it. Do not fabricate a Kelly p_win.
What it does
Applies a fractional Kelly formula (default 0.5 = half-Kelly) with:
- Max Kelly cap — default 25% of bankroll. Never lets a single position exceed this fraction regardless of what Kelly says.
- Volatility scalar — low = 1.0, medium = 0.7, high = 0.4. Scales the Kelly notional down as volatility rises.
- Drawdown adjustment — if the account is currently down, shrinks position size linearly toward the max tolerable drawdown.
If Kelly returns non-positive (bad thesis: p_win too low relative to R:R), verdict is skip and position_usd is null. This is the "no_op" case explicit in the sizer contract.
The named invariant
The sizer never emits a direction. Direction is the caller's decision. The sizer only decides fund/skip and position_usd. This is what makes it composable — you can pair it with any direction source (fundamental analyst, ML model, human trader, Orallexa's 5-voice debate).
Contract test coverage in tests/test_risk_sizer_contract.py (7 pinned invariants).
Input schema
{
"direction": "long" | "short" | "no_op",
"directional_confidence": 0.72,
"bankroll_usd": 10000,
"volatility_regime": "low" | "medium" | "high",
"kelly_p_win": 0.55,
"kelly_avg_win_pct": 0.04,
"kelly_avg_loss_pct": 0.02,
"current_drawdown_pct": 0.0,
"max_kelly_cap": 0.25
}
Output schema
{
"voice": "Risk Sizer",
"verdict": "fund" | "skip",
"position_usd": 350.00,
"kelly_notional": 500.00,
"volatility_scalar": 0.7,
"rationale": "Kelly=500 (p_win=0.55, R=2:1)...",
"metrics": { "direction": "long", ... }
}
Approval boundary
The Judge that supplies direction is upstream and out of scope for this skill. If the direction is no_op, the sizer returns skip immediately without running Kelly. If Kelly returns negative for a stated long/short direction, the sizer returns skip and cites the specific p_win / R:R that failed break-even in the rationale.
Non-goals
- Not a directional signal. Never tells you whether to buy or sell.
- Not a portfolio-level VaR or concentration limit. Position-level only.
- Not tied to a specific broker or execution venue. Returns a dollar amount; you decide how to fill.
- Not a full trading system. Pairs with an upstream Judge (5-voice debate, LLM analyst, or a human).
Reference
- Orallexa
engine/risk_sizer.py — the Python source of truth.
- FinPos: A Dual-Agent Framework for Financial Decision-Making. arXiv:2510.27251.
- Byte-identical port at
shadow-mentor/lib/personas/trader-pack/risk-sizer.js in the Shadow banking-compliance repo.
