Luma
Signal aggregation and directional bias from exchange derivatives data.
What Luma Does
Luma finds moments where independent signals across exchanges agree. When open interest, funding rates, volume, and cross-exchange flow all lean the same direction, that's conviction — not noise.
Luma runs on every candle close (default trigger: 5m, faster than Vanta) and publishes a directional bias score that Riven uses in the consensus debate.
What Luma Reads
Luma pulls derivatives data from the exchanges PerpDesk integrates with. It does not scrape social media, news, or on-chain wallets. The inputs are:
- Open interest — trends and cross-exchange divergence
- Funding rates — outliers and cross-venue spread
- Volume — conviction, concentration, and delta
- Cross-exchange flow — where positioning is building vs. unwinding
- Candles (20 per timeframe) for price-signal correlation
How the Bias Score Works
Four components, each scored ±25, sum to a composite in the range -100 to +100:
| Component | What it measures |
|---|---|
| OI | Cross-exchange open interest trends |
| Funding | Rate extremes and cross-venue spread |
| Volume | Conviction and concentration |
| Flow | Positioning shifts across venues |
The composite produces a bias label (bullish / bearish / neutral) with strength and confidence. A score of +62 means "moderately bullish, multiple sources agreeing." A score near zero means "no consensus across exchanges."
What Luma Publishes
A structured read containing the bias label, the four component breakdowns, and the composite score. Riven reads this as one of three inputs to the consensus debate.
Configuration
| Setting | Options | Default |
|---|---|---|
| Sensitivity | Calm / Alert / Hyper | Alert |
| Trigger timeframe | 1m / 5m / 15m / 1h | 5m |
| Signal filter | All / Strong only | All |
Under the Hood
Type: candle-triggered, LLM-powered analysis agent.
Trigger: candle close on configured timeframe (default 5m).
Candle limit: 20 per timeframe.
Signal sources: OI, funding, and volume pulled from all connected exchanges.
Output: publishes a structured analysis per symbol via the event bus. Persisted to the database and hot-cached.
Upstream dependencies: none. Luma is first-layer.