Risk Scoring Guide

EarnForge computes a composite 0-10 risk score for every vault. Higher score means safer. The score is deterministic, purely data-driven, and based on 7 dimensions with fixed weights.

Score Labels

Score Range	Label	Color
>= 8.0	low risk	Green
6.0 - 7.9	medium risk	Yellow
< 6.0	high risk	Red

These cuts are calibrated against the live fleet, whose scores span 4.1 to 9.7 with a median of 7.9. The original thresholds (low >= 7, high < 4) put roughly 80% of vaults in “low” and made “high” unreachable, so the label carried no information.

The 8.0 cut is deliberate. A verification-flagged vault scores at most 7.96 even with a perfect showing on every other dimension, so no flagged vault can ever be labelled low risk. That upper bound is real rather than theoretical: no protocol scores above 9, and no asset type scores above 9, so the remaining 0.78 of weight cannot reach 8.0 on its own.

Note

“Low risk” means the vault scores well across all dimensions. It does not mean the vault is risk-free. DeFi always carries smart contract risk, oracle risk, and market risk that are beyond the scope of this scoring system.

---

The 7 Dimensions

1. TVL Magnitude (18% weight)

Measures how much capital is locked in the vault. Higher TVL generally indicates more trust and lower risk of rug pulls or liquidity crises.

TVL (USD)	Score
>= $100M	10
>= $50M	9
>= $10M	8
>= $5M	7
>= $1M	5
>= $100K	3
< $100K	1

Caution

The spec declares tvl.usd a string; the live API sends a number. The SDK’s parseTvl() accepts either, so the flip is a non-event for callers, but code that assumes one shape will break when it meets the other. See Pitfall #8.

2. APY Stability (14% weight)

Measures how much the current APY diverges from the 30-day (or 1-day) historical APY. Large divergence suggests volatile or unsustainable yield.

Divergence	Score
< 5%	10
< 10%	8
< 20%	6
< 50%	4
>= 50%	2

The divergence formula is:

divergence = |apy.total - ref| / max(apy.total, ref)

Where ref is apy30d (preferred) or apy1d (fallback). If both are null, or either side is zero, the score defaults to 5 (moderate).

Note

Short-lived vaults often have apy1d: null and apy30d: null. Instead of penalizing them harshly or rewarding them unfairly, the SDK assigns a neutral score of 5. See Pitfall #18.

3. Protocol Maturity (18% weight)

Known protocols receive a fixed tier score based on track record, audit history, and scale. All 27 live protocols carry a tier; the table below is keyed by LI.FI’s unversioned protocol id.

Tier	Protocols	Score
Blue chip: multi-year record, deep audits, systemic scale	aave, morpho	9
	yearn	8
Established: long record or very large, well audited	spark-v2	8
	euler, fluid, pendle, etherfi-staking	7
	ethena, maple	6
Mid: real adoption, but young, thin, or unaudited	ipor, midas, kinetiq, hyperlend, upshift	5
Small or young: audited, but short record and limited scale	avant, cap, neverland, yo, usdai, nest, concrete, infinifi, auto-finance, ember	4
Examined and still unknown	apyx, hypurrfi	3
Unknown protocol (fallback)	anything unlisted	3
Examined, findings adverse	ample	2

Two entries are worth calling out, because they are where size and safety come apart:

• kinetiq scores 5 despite $785M TVL: listed 2025-07 with zero audits. Scale is not a substitute for review.
• ample scores 2, below the unknown-protocol fallback: listed 2026-03, zero audits, and lottery-style payout mechanics. Being examined can be worse than being unexamined.

Caution

Protocol ids are unversioned. aave-v3, morpho-v1 and euler-v2 match nothing and return zero results with HTTP 200. See Pitfall #19.

4. Redeemability (10% weight)

Whether the vault supports withdrawals. Non-redeemable vaults carry additional liquidity risk since you may not be able to exit your position.

Redeemable	Score
Yes (isRedeemable: true)	10
No (isRedeemable: false)	3

5. Asset Type (10% weight)

Whether the vault holds stablecoins, less any impermanent-loss exposure LI.FI tags.

Asset Type	Score
Has stablecoin tag	9
No stablecoin tag	5
minus impermanent-loss exposure	−3 (floored at 1)

6. Verification (22% weight)

LI.FI publishes verificationStatus on every vault and flags ~10% of the fleet (74 of 703). The field appears in no spec, changelog, or quickstart, and LI.FI’s own hosted MCP server does not expose it, but it is LI.FI’s own vault-quality signal, so it carries the single largest weight here.

Verification	Score
Not flagged	10
Flagged, reason includes apy_outlier	1
Flagged, any other reason	2

Caution

Sort by APY without checking this and a flagged vault lands at the top of your list. The highest advertised APY in the entire fleet is flagged apy_outlier. See Pitfall #21.

7. Reward Dependency (8% weight)

What share of the yield is paid in token incentives rather than earned. Incentives stop; lending fees generally do not.

Reward share of total APY	Score
0%. No rewards reported	10
< 20%	9
< 40%	7
< 60%	5
< 80%	3
>= 80%	2
Not reported (null)	7

Note

apy.reward is three-valued and the distinction is load-bearing: null means the protocol did not report, 0 means it reported none, and these are not the same claim. A null scores 7 as an opacity penalty, not a sustainability verdict. Morpho returns 0; Euler and Aave return null.

---

Weighted Formula

The final score is a weighted average:

score = (tvl * 0.18) + (apyStability * 0.14) + (protocol * 0.18)
      + (redeemability * 0.10) + (assetType * 0.10)
      + (verification * 0.22) + (rewardDependency * 0.08)

Weight rationale:

Dimension	Weight	Rationale
Verification	22%	LI.FI’s own quality signal. The only dimension sourced from the vendor’s judgment rather than inferred
TVL Magnitude	18%	Strong signal of trust and liquidity
Protocol Maturity	18%	Audit history and track record are the strongest safety indicators
APY Stability	14%	Volatile APY suggests high risk or unsustainable incentives
Redeemability	10%	Ability to exit is important but secondary
Asset Type	10%	Price risk matters but is user-preference dependent
Reward Dependency	8%	Sustainability signal, but a high reward share is a reason to look closer rather than to avoid

---

Example Calculations

Both examples are real vaults, scored against the live API.

A clean vault: STEAKUSDC (Morpho, Base)

Dimension	Raw Score	Weight	Weighted
TVL ($579.9M >= $100M)	10	0.18	1.80
APY Stability (divergence < 5%)	10	0.14	1.40
Protocol (morpho)	9	0.18	1.62
Redeemability (true)	10	0.10	1.00
Asset Type (stablecoin)	9	0.10	0.90
Verification (not flagged)	10	0.22	2.20
Reward Dependency (0% rewards)	10	0.08	0.80
Total			9.7

Label: low risk. This is the highest score in the fleet.

The highest APY in the fleet: USP (Pendle, Ethereum)

Advertised at 106.16% total APY. 92.40 points of that are rewards.

Dimension	Raw Score	Weight	Weighted
TVL ($216K >= $100K)	3	0.18	0.54
APY Stability	10	0.14	1.40
Protocol (pendle)	7	0.18	1.26
Redeemability (true)	10	0.10	1.00
Asset Type (stablecoin)	9	0.10	0.90
Verification (flagged apy_outlier)	1	0.22	0.22
Reward Dependency (87% of APY)	2	0.08	0.16
Total			5.5

Label: high risk: despite the best headline number on the platform.

---

Usage

SDK

import { createEarnForge } from '@earnforge/sdk';

const forge = createEarnForge({ apiKey: process.env.LIFI_API_KEY });
const vault = await forge.vaults.get('morpho:8453:_:0xbeef...');
const risk = forge.riskScore(vault);

console.log(risk.score);   // 9.7
console.log(risk.label);   // "low"
console.log(risk.flags);   // [], populated with plain-language reasons when risk is present
console.log(risk.breakdown);
// { tvl: 10, apyStability: 10, protocol: 9, redeemability: 10,
//   assetType: 9, verification: 10, rewardDependency: 10 }

CLI

earnforge risk morpho:8453:_:0xbeef0e0834849acc03f0089f01f4f1eeb06873c9

Prints all seven dimensions plus a Flags section naming why a vault carries risk. Requires @earnforge/cli 1.0.1 or later: 1.0.0 printed five dimensions and no flags.

React

import { useRiskScore } from '@earnforge/react';

function RiskDisplay({ vault }) {
  const { data: risk } = useRiskScore(vault);
  // risk.score, risk.label, risk.breakdown, risk.flags
}

MCP

Prompt: “What’s the risk score for vault morpho:8453:_:0xbeef…?”

The get-vault-risk MCP tool returns the full breakdown as validated structured output.

---

Filtering by Risk

Use the risk-adjusted strategy preset to filter for low-risk vaults only:

const top = await forge.vaults.top({
  asset: 'USDC',
  strategy: 'risk-adjusted',  // score >= 7, flagged vaults excluded
});

Note that the preset’s cutoff of 7 predates the label recalibration, so it admits the top of the medium band rather than low-risk vaults only. Filter on riskScore(vault).label === 'low' if you want the stricter reading.

Or use the suggest() engine, which weights allocations by risk score:

const allocation = await forge.suggest({
  amount: 10_000,
  asset: 'USDC',
});
// Higher-scoring (safer) vaults get larger allocations

suggest() excludes verification-flagged vaults unconditionally. Pass includeFlagged: true to opt back in.