Pool scoring methodology
Every pool in every index is ranked by a single, transparent composite score. The score drives both deployment (highest-scored pools) and exit (lowest-scored pools), and it runs continuously.
The formula
score = 0.30\,(yield) + 0.30\,(7d momentum) + 0.20\,(low-IL) + 0.20\,(TVL)
| Weight | Factor | What it captures |
|---|---|---|
| 30% | Yield | Sustained APY (30-day basis), not a transient spike. |
| 30% | 7-day momentum | Is yield/flow strengthening or decaying? Direction matters as much as level. |
| 20% | Low impermanent loss | Pairs and ranges that preserve capital are rewarded. |
| 20% | TVL | Depth and resilience — thin pools are penalized. |
Each factor is normalized across the candidate set so the composite is comparable pool-to-pool and index-to-index.
Selection rules
- ≥ 40 pools per index — the score selects breadth, not a single winner.
- ≤ 2.5% per position — sizing is capped regardless of score; a great pool cannot become a concentration risk.
- Sustained 30-day APY — outliers and one-off spikes are filtered before scoring.
- Anomaly flags — APY collapse, TVL outflow, exploit signals, governance events, depeg risk and IL spikes override the raw score and force review.
Data inputs
Scoring runs on live on-chain data:
- DefiLlama — yields, TVL and momentum across ~13,000 pools on 50+ chains (no API key required).
- Price feeds — token prices by contract for honest USD valuation and spam filtering.
- Protocol-native reads — health factors, pool depth, fee tiers.
Top-10 / Bottom-10
For each index the console surfaces:
- Top-10 highest-score — the deployment index. New capital is routed here.
- Top-10 lowest-score — the exit/rebalance signal. Withdrawals and de-risking are taken from here first.
This asymmetry is deliberate: always enter strength, always exit weakness — it preserves the alpha of the high-yield set.
Human-in-the-loop
The AI Operator computes scores and raises flags continuously. A human operator validates the flags daily, confirms or adjusts the scoring fields in the console, and authorizes rebalances when the 2.5% cap is breached. The methodology is fixed; discretion is limited to confirming the machine's read — not overriding the rules.