OI-Contrarian desk
Fade perp open-interest extremes — tradeable only where an asset has earned the edge; on NEAR it is context only.
Agent token 8. This desk is the path by which a validated trading edge — fading crowded perp open-interest — could reach the PM. It only fires when both a registry-validated edge and a live signal align. On NEAR the registry is empty, so today it supplies context only (see the caveats below).
What it reads
firm/organization.py · tool_oi_contrarian():
- Validated-edge registry —
data/validated_edges.json, produced by a cost-aware OOS backtest (scripts/59throughfirm/edge_lab.py, orscripts/39for the OI backtest). Holdsp_win,payoff_b,sample_n,oos_roi_pctper asset that survived. - Live OI-contrarian signal — perp OI 24h-change quintiles: top quintile → SHORT, bottom quintile → LONG (fade the crowd).
The disciplined gate
| Perp OI 24h change | Crowd | Fade action |
|---|---|---|
| Top quintile (OI spiking) | crowd piling in | SHORT |
| Bottom quintile (OI flushing) | crowd capitulating | LONG |
| Middle | no extreme | no trade |
Output (shape)
{
"asset": "NEAR",
"edge_validated": false,
"actionable": false,
"live_signal": null,
"caveat": "No validated edge for this asset — context only; the firm abstains unless an edge is earned."
}Honest caveats
The method is only worth acting on when an asset has earned it. On NEAR it has not: Bitget publishes open interest only as a live snapshot, so the OI column is empty in the history and the hypothesis cannot be validated (Finding #6). Crucially, the registry is allowed to be empty — and it is. This desk is context only for NEAR, and the firm abstains. If an asset's edge ever clears the gate, it is sized with fractional-Kelly and hard risk caps, never bet as a sure thing, and it stays earned on every re-test.