Risk, priced into the score.
ArbiFlow doesn't bolt a separate risk grade onto a yield number. Resilience is built into the 100-point composite — 60 of the points reward depth, trust, stable assets and a holding forecast; the other 40 reward yield.
Where every point comes from.
Base + reward APY from DeFiLlama on a log curve that saturates near 30%, so a 4% pool and a 6% pool are close, not 50% apart.
Pool depth, log-scaled to $100M. Deeper liquidity is harder to move and cheaper to exit, so it scores higher.
Full marks for the curated blue-chip protocols; honorable-mention pools from the wider universe earn partial credit.
Stablecoin pools score on the asset; pools with no impermanent-loss exposure score on IL. A stable, single-sided pool earns both.
DeFiLlama's Stable/Up probability that the pool's yield holds rather than decaying — straight from their prediction feed.
What this model deliberately does not do.
The honest version: the score is a clear, deterministic composite — not a risk oracle. These are the things it does not claim to capture.
- Audit / security scoring
We curate by protocol reputation (trust factor); we don't grade individual audits or attestations.
- Asset volatility (σ)
No per-asset standard deviation or depeg history. Stability is a binary stablecoin / IL signal, not a variance model.
- Gas & impermanent-loss simulation
The score is not net of gas, and IL is a yes/no flag — not a position-level simulation against price ranges.
- Withdrawal latency
Cooldowns, queues, and exit delays aren't modeled. TVL is the only liquidity proxy.
- Point-in-time data
APY and TVL are the latest DeFiLlama reading (refreshed ~5 min), not a 30-day average — fresh spikes are visible, not smoothed.
- Other chains
Scoring is Arbitrum-only. Balances are read on Base and Optimism so idle capital can be bridged in, but pools are ranked on Arbitrum.