Yeelds Vault Risk Methodology
Methodology v1.2.1· see the revision log below
Yeelds gives covered vaults a 0–100 score and a letter band, A+ down to D. It answers one question: given what you deposit and what the vault does with it, how risky is this? The grade is keyed on the vault's true exposure, not its label. This page publishes the actual numbers, so anyone can challenge a specific one. Accepted changes bump the version (see the log).
Higher = safer. 100 and A+ are safest; 0 and D are riskiest.
It's an opinion, not a guarantee.A grade is our relative risk read on its grading date, under the version it was scored with. A low or missing grade can mean ‘too new,’ not ‘dangerous.’ Every grade shows its date, version, full breakdown, and what we could not verify. Grades are researched(a curated seed, not live-computed) and shown beside the Pharos stablecoin signal — never blended into it.
The model: anchor + modifier, never an average
Risk compounds, so fatal flaws are multipliers— they cap the whole grade — while only soft signals are added (and bounded). A grade caps; it never averages. Averaging a fatal flaw against strengths is exactly how linear rubrics rated Aave-rsETH highly right up to its April 2026 exploit. This model is built so that can't happen.
score = clamp( round( (anchor_base + Σ soft_delta) × Π multiplier ), 0, 100 ) grade = band(score)
- Anchor— strategy exposure tier (S) × protocol tier (P) → base points.
- Soft signals— bounded additive deltas, clamped to [−16, +5] in total.
- Fatal risks— multipliers that scale the whole score.
- Output— score + letter + a breakdown of every component, plus anything we couldn't verify and the methodology version.
Axis A — Strategy Exposure Tier (S1–S5)
What does the vault actually do with your deposit? FLAT— you still hold a claim on the same asset (only contract risk is added). ADDS— you now hold a different exposure (new token, peg, or off-chain counterparty). AMPLIFIES— leverage multiplies the exposure, so it is capped.
| S-tier | Exposure | Examples |
|---|---|---|
| S1 | FLAT | single-asset lending/supply (Aave, Compound, Morpho-at-deposit) |
| S2 | ADDS, correlated | stable-stable LP; mature LST (stETH); a battle-tested fiat stable |
| S3 | ADDS, market | volatile LP; LRT (weETH/rsETH); Pendle PT; mechanism stable (USDe/GHO) |
| S4 | ADDS, counterparty | delta-neutral basis; RWA / private credit; CDP mint-and-deploy |
| S5 | AMPLIFIES | recursive leverage / looping |
- The dominant leg sets the tier. We read the vault's real allocations and grade the biggest true-exposure class. A risky-but-minority leg becomes a delta instead; once a leg passes 50% of the book it reclassifies the whole tier (a bounded delta can't stand in for a majority).
- Stable line is about mechanism, not age. A fiat-reserve stable under a recognized regime (MiCA, NYDFS) is S2; a mechanism or yield-bearing stable (CDP / delta-neutral / algorithmic) is S3. Peg youth is priced by a small delta, never the tier.
Axis B — Protocol Quality Tier (P1–P3)
| P-tier | Roughly | Examples |
|---|---|---|
| P1 institutional | ≥3yr live, ≥3 reputable audits, >$1B all-time TVL, timelock + active multisig, clean record | Aave V3, Morpho core, Spark |
| P2 established | ≥1yr, ≥2 audits, >$50M peak TVL, some governance, no open incidents | Euler V2, Pendle, Ethena |
| P3 emerging | new, smaller, or thin audit trail — the default for un-tiered protocols | newer entrants |
P grades the platform that actually holds the deposit(the lowest tier when layers stack). For a Morpho meta-vault that is Morpho (P1) regardless of who curates — curator quality flows through the curator delta and the governance multipliers, so operator risk isn't counted twice.
The anchor matrix (S × P → base points)
Each cell is a letter mapped to the mid-point of its band, so a vault with no modifiers keeps its anchor grade. The strategy class caps how high the grade can reach.
| P1 | P2 | P3 | |
|---|---|---|---|
| S1 FLAT | a+ (96) | a− (84) | b (72) |
| S2 ADDS-corr | a (90) | b+ (79) | b− (66) |
| S3 ADDS-market | b+ (79) | b (72) | c+ (60) |
| S4 counterparty | b− (66) | c+ (60) | c (53) |
| S5 AMPLIFIES | b (72) — cap | c (53) | d (30) |
Base-point lookup: a+ 96 · a 90 · a− 84 · b+ 79 · b 72 · b− 66 · c+ 60 · c 53 · c− 46 · d 30. S5 caps at beven on a P1 protocol — one terrible axis bounds the whole grade. A+ is reserved for S1×P1: an S2 anchor (90) plus the +5 soft cap tops out at 95.
Modifiers — two kinds
Soft signals → bounded deltas (total clamped to [−16, +5])
No single soft signal can swing the grade far.
| Soft modifier | What it captures | Range |
|---|---|---|
| Asset / collateral quality | deposit and underlying collateral grade (Pharos for stables, asset-class ladder otherwise) | −8 .. +2 |
| Curator quality | operator track record within normal bounds (a realized blowup is a multiplier instead) | −5 .. +2 |
| Withdrawability | thin withdrawable liquidity vs. utilization (a near-empty exit is a hard floor — below) | −5 .. 0 |
| Depositor concentration | top-5 holders > 80% | −4 .. 0 |
| TVL band | very thin → penalty; deep → no bonus | −5 .. 0 |
| Timelock | ≥72h → +1 · 24–48h → −1 · <24h / none → −2 | −2 .. +1 |
| Reward dependency | >70% of APY from one volatile reward token, or fast reward decay | −5 .. 0 |
Stables → Pharos.We map from the Pharos grade (baseline USDC at B+ = 0), so A− → +1, B → −2, C+ → −5, down to D → −8. A peg live under 2 years adds up to −3 on top. Severe active depeg is handled by a multiplier, not a delta.
Fatal risks → multipliers (the anti-averaging mechanism)
These scale the whole score, so a fatal leg can't be averaged away by strengths.
| Multiplier | Trigger | Factor |
|---|---|---|
| Toxic / depegged collateral | a material (≳15%) leg in dead or depegged collateral, or active depeg ≥1000 / ≥2500 bps | ×0.30 / ×0.50 / ×0.25 |
| Issuer-NAV opacity (v1.2) | share of the book priced by an unverifiable issuer-controlled NAV feed (see below) | ×0.50 / ×0.78 |
| Governance — rug-class | effective single key with no real timelock; self-referential collateral on a dominant leg | ×0.55 |
| Governance — red flag | named single-key control; anonymous keys; extreme holder concentration | ×0.78 |
| Realized incident — recent | fresh exploit/insolvency or unremediated losses on the curator's book | ×0.55 – 0.60 |
| Realized incident — old / made-whole | contained, >6mo old, no open losses; decays out over 12 months | ×0.82 – 0.85 |
| Active leverage | live leverage-liquidation surface (on top of the S5 cap) | ×0.90 – 0.95 |
The same fact never stacks — the worst multiplier wins.A leg that is both dead and depegged takes ×0.30, not ×0.30××0.25. The issuer-NAV multiplier and the ×0.30 toxic/depeg floor never compound either: the lowest one applies (NAV is the pre-depeg structural penalty; ×0.30 is the post-depeg realized one). Distinct facts do compound. The score rounds once, at the end.
Governance key-management (v1.1)
Operator and key failure cause most realized vault losses, so we read the on-chain control surface: who can move funds (owner / curator), the multisig threshold, the timelock, and any sentinel / guardian. The grade is driven by the setup, not reputation.
- An effective single key(an EOA, a 1-of-1 Safe, or any threshold-1 control) with no real timelock is a multiplier — ×0.55 (rug-class) when anonymous or with no protection, ×0.78 (red flag) when the operator is named.
- Proper multisig (M-of-N, M≥2) with a working timelock takes no penalty.
- A named, reputable operator is shown, but it never rescues a weak setup— a single key is a single key.
This is the dimension reputation/Lindy scores ignore, and it's cheaply verifiable on-chain. We surface the full control surface in the vault's Governance & Security panel.
Issuer-unverifiable NAV (v1.2)
The hardest risk to see: a chunk of a vault is priced by a NAV number the issuer controls and that you cannot independently check. The valuation can sit at par while the underlying is impaired — the loss is invisible until the issuer chooses to mark it down. We treat this as a serious, structural risk, sized by how much of the book it prices.
| Share of book on issuer-NAV | Treatment | Effect |
|---|---|---|
| > 50% (dominant) | multiplier ×0.50 | craters to D |
| 15–50% (material) | multiplier ×0.78 | ~1.5 bands |
| < 15% (minor) | a soft delta | nudge |
The STRICT bar: trust, but verify
Our rule: if there's nothing to verify, assume there's nothing there. A NAV leg is exempt from the multiplier only if it has one of:
- independent proof-of-reserves, or
- a regulated / independent ~daily NAV attestation, or
- an on-chain auto-marking rate that moves with the asset.
Self-reported “we mark it down,” a periodic self-audit, or a proof that reserves merely exist (without marking impairment) do not exempt. The test we apply: would you commit your own capital on the issuer's word alone?
Hard floors (v1.2)
Some flaws cap the grade regardless of how high the anchor is:
- Dealbreakers → cap at C or below: an unverified contract, blocked redemptions, a single key without a timelock, or no audit.
- Very low withdrawable liquidity → cap at C or below. Trapped capital is how a 2026 loss got locked in. Carve-out: a deep, healthy, high-utilization pooled reserve, or a vault that is locked by design (pre-deposit / migration), is not trapped capital and is exempt.
NAV refinements (v1.2.1)
Two clarifications keep the NAV rule proportionate — it should penalize unverifiable valuation, not the mere existence of a NAV mechanism.
- Market-priced PTs are not issuer-NAV. A Pendle PT priced by a market/discount oracle — it trades at a discount and has a live secondary market — is independently priced by that market, so the NAV multiplier does notfire on it. Its underlying's risk is priced as a normal collateral delta.
- We check attestation leg by leg. The exemption is applied per leg, and we band off only the share that is still unverifiable. A credit leg with a genuine daily attestation is exempt; a sibling leg with only an existence-style proof-of-reserves still fires.
And when one leg is both self-dealing and NAV-opaque, that's one fact — one multiplier, never two stacked.
Score → letter (fixed thresholds, never a curve)
| A+ | A | A− | B+ | B | B− | C+ | C | C− | D |
|---|---|---|---|---|---|---|---|---|---|
| 93–100 | 87–92 | 82–86 | 76–81 | 70–75 | 64–69 | 57–63 | 50–56 | 43–49 | 0–42 |
Fixed absolute thresholds, never a percentile curve — a curve would re-grade a vault when other vaults change, which is just noise for a curated catalog. It is fine for most of a curated catalog to score well; C and D stay meaningful for genuinely risky vaults so the scale keeps its teeth.
Missing signals & honesty
- A signal we can't verifyis treated as neutral and listed under “Not verified.” We never penalize a vault for our own data gaps.
- Caps and fatal multipliers only apply where we can positivelyclassify the risk — an unclassifiable strategy shows “strategy unclassified,” not “assumed safe.”
- Deliberate opacity is different from a data gap.No proof-of-reserves + issuer-controlled NAV + an unclassifiable strategy holds the grade back from publication entirely — we don't hand it a neutral score.
Worked examples
Real vaults from the calibration runs. Format: (anchor_base + Σsoft) × Πmult = score.
Steakhouse USDC — A (the clean case)
S2×P1 anchor 90 · clean collateral · no multipliers. (90 + 0) × 1 = 90 → A. Blue-chip collateral, top-tier curator, deeply liquid — no modifiers, so it keeps its anchor.
Galaxy USDC Quality — B− (a multiplier at work)
S2×P1 anchor 90 · depositor concentration −4 (99.8% in two wallets) · governance red flag ×0.78 (anonymous EOA keys). (90 − 4) × 0.78 = 67 → B−. Pristine collateral can't average away an operator red flag.
An issuer-NAV vault — D (v1.2 in action)
S4×P1 anchor 66 · soft sum at the −13 floor · ~80% of the book on an unverifiable issuer-NAV feed ×0.50. (66 − 13) × 0.50 = 27 → D. Under v1.0 the same vault scored C — the masked NAV leg was a soft delta. v1.2 makes it a multiplier, which is why it now caps.
Revision log
- v1.0 — 2026-06-10 — launch. The anchor + modifier model, assignment rules, and tiered multipliers, calibrated against the first Morpho grading run.
- v1.1 — governance key-management — added the on-chain control-surface signal (threshold × timelock × sentinel); an effective single key without a timelock is a multiplier.
- v1.2 — issuer-unverifiable NAV — promoted NAV opacity from a soft delta to a materiality-banded multiplier, added the STRICT attestation bar and the C4 hard floors.
- v1.2.1 — NAV refinements — the market-priced-PT guardrail (a market-priced PT is not issuer-NAV) and per-leg attestation (exempt leg by leg; band off only the unverifiable share).
Every rendered grade carries its own date and version; grades scored under different versions are never mixed in a sort. Research partners: critiques of any constant on this page are welcome — an accepted revision bumps the version, triggers a re-grade, and is recorded here.