Asset Composition (30%)

Category specific asset scoring, evidence quality, caps, staleness penalties, market LLTV haircuts, and Aave lending market construction.

Asset quality is the irreducible risk floor. The asset model scores assets by category against explicit dimensions, then applies non-compensatory caps.

1) Category Taxonomy

Category	Examples	Primary failure mode
native	ETH, WETH	Chain halt or liveness failure
fiat_backed_stablecoin	USDC, USDT, PYUSD	Issuer insolvency or reserve shortfall
cdp_stablecoin	DAI, USDS, LUSD, GHO	Collateral cascade or governance attack
synthetic_stablecoin	USDe, USD0, eUSD	Basis unwind or funding inversion
lst	wstETH, cbETH, rETH	Validator penalty and withdrawal risk
lrt	weETH, ezETH, rsETH	AVS validator penalty aggregation risk
wrapped_btc	WBTC, cbBTC, tBTC	Bridge or custodian failure
governance_token	LINK, UNI, ARB	Volatility and liquidity shock
defi_derivative	PT and YT wrappers	Pricing and settlement complexity
tokenized_treasury	BUIDL, BENJI	Issuer, custodian, and redemption risk
t_bill_backed_stable	AUSD class assets	Backing, attestation, and liquidity risk
rwa_credit	Tokenized private credit	Credit quality and redemption risk
commodity_backed	Gold-backed tokens	Custodian and redemption risk
tokenized_equity	Tokenized equity wrappers	Legal wrapper and market-access risk
wrapped_crypto	Wrapped crypto assets	Wrapper, bridge, or custody risk
unreviewed	Any unclassified asset	Unknown or insufficient assessment

2) Dimension-Based Scoring

The weighted formula depends on category. The live dimension set is listed below.

Dimension	Applies most often to
`peg_stability`	Stablecoins and pegged wrappers
`issuer_custody`	Issuer-backed and custody-backed assets
`redeemability`	Stablecoins, RWAs, and wrapped assets
`reserve_transparency`	Backed assets and off-chain collateral
`governance_controls`	Mutable wrappers and protocol-governed assets
`protocol_security`	Smart-contract and wrapper assets
`operator_quality`	LSTs, LRTs, and managed RWA products
`bridge_mechanism`	Wrapped and cross-chain assets
`dependency_depth`	Recursive or stacked wrappers
`liquidity`	All tradable assets
`volatility`	Governance tokens and non-pegged assets
`credit_quality`	Credit-fund and private-credit exposures
`default_rate`	Credit-fund and private-credit exposures
`recovery_rate`	Credit-fund and private-credit exposures
`subadvisor_concentration`	Credit funds with delegated underwriting
`redemption_terms`	Credit funds and products with gated withdrawals

Missing or provisional dimension values use a neutral conservative default of 5.0.

3) Cap Cascade

Final asset score is the minimum of the weighted dimension score and all active caps.

final_asset_score = min(
  weighted_dimension_score,
  overlay_cap,
  review_status_cap,
  hard_fail_cap,
  override_cap,
  staleness_cap
)

Review status caps

Review status	Cap
`reviewed`	10.0
`provisional`	9.0
`unreviewed`	7.9

Unreviewed category scoring remains stricter than the cap alone. The live formula also constrains unreviewed assets through liquidity-weighted fail-safe behavior.

4) Overlays And Market Haircuts

yield_bearing wrappers are constrained by wrapper and underlying dependency quality.
lockup overlays reduce the liquidity dimension.
Morpho and lending-market LLTV haircuts apply to exposure scores before portfolio aggregation.

Market LLTV collateral penalty

LLTV band	Multiplicative haircut
`>= 0.95`	10.0%
`>= 0.90`	5.0%
`>= 0.85`	2.5%
`< 0.85`	none

The published asset vector includes these market haircuts when applicable.

5) Hard-Fail Controls

Hard-fail flags apply strict caps regardless of weighted dimension quality. When a flag is active, the final asset score cannot exceed the listed cap, even if the rest of the evidence is strong.

Flag	Cap	Cooldown
`sanctions_exposure`	0.0	7 days
`active_depeg`	1.0	7 days
`redemption_paused`	2.0	3 days
`single_signer_upgrade`	3.0	14 days
`endogenous_collateral_high`	4.0	30 days
`proof_of_reserve_missing`	4.0	30 days
`unaudited_token_contract`	4.0	none
`no_recent_attestation`	5.0	30 days

Flags are event-sourced and may include trigger windows and cooldowns.

6) Staleness Penalties

Evidence freshness affects effective dimension values.

State	Treatment
stale	value multiplied by 0.92
expired	value reduced to 75% of the last value

Expired evidence has no score floor. When more than 50% of weighted evidence is stale or expired, the asset score is capped at 8.0.

7) Unresolved Address Fallback

If an adapter cannot resolve an exposure address, scoring falls back to strict unreviewed behavior.

asset score is constrained to a conservative low cap
explainability includes a machine-readable unresolved-address reason

8) Portfolio-Level Adjustments

After per-exposure asset scoring, vault-level portfolio adjustments may apply.

Adjustment	Score effect
concentration	Applied to the asset portfolio score
wrong-way risk	Applied to the asset portfolio score
category correlation	Informational only in the live score

Single-asset vaults are exempt from concentration and correlation penalties.

Post-composite asset-quality drag uses assetQualityAnchor, not always the same value as the published asset vector. The anchor includes category scores, caps, and review-status constraints, but excludes LLTV haircuts and portfolio-level adjustments.

This distinction matters for lending markets. A vault can show an LLTV-penalized asset vector below 8.0 while still carrying an anchor at or above 8.0 for the post-composite drag check.

Any exposure of at least 5% weight with a binding fail-safe cap forces the anchor below 5.0, which then limits the final vault score to the Edge ceiling.

10) Aave v3 Lending Markets

For single-sided lending markets on Aave v3, the asset vector is not the deposit asset alone. A depositor in an Aave pool is exposed principally to the token they supplied, and through the shared liquidation engine to the other active reserves in the same pool.

The Aave lending asset score is a 50/50 weighted composite.

asset_score_aave_lending
  = 0.50 * score(deposit_asset)
  + 0.50 * weighted_mean(
      score(other_active_reserves),
      weight = reserve_share_of_pool,
      filter = reserve_share >= 1%
    )

The reserve filter excludes dust reserves. Dimension scoring, cap cascade, hard-fail flags, staleness handling, and portfolio adjustments apply before the components are combined.