CRS·COLLATERAL RISK STANDARD
sister standard to RWALS  ·  V1.0 public draft
An open standard for everything pledged as collateral

On April 17, 2026, rsETH carried a 93% LTV. The standard says: ineligible.

Twelve months. Four major lending ecosystems. The same failure in four costumes: collateral admitted on judgment, monitored by nobody, repriced after the loss. CRS replaces judgment with a published, reproducible score — and committee repricing with automatic response.

One problem, four costumes

Everything below is from the public governance record of the past twelve months.

Aave

An estimated one hundred twenty-three to two hundred thirty million dollars in bad debt after a token wrapped four layers deep, bridged over a single-attestor route, was underwritten like native ETH.

Euler

The Elixir and Steam Labs collapse left users seeking treasury compensation — while risk stewards grind monthly manual parameter updates across dozens of expiring instruments.

Morpho

Depositors learned that curated vaults can freeze withdrawals, and that oracle deviation is handled by improvised timelocks instead of a standard.

Centrifuge

No oracle anywhere proves a physical asset actually exists and matches its on-chain representation. MANTRA and Tangible/USDR were the bill for "trust us."

Eight dimensions. Every one bought with real losses.

Each dimension is scored 0–100 from published, checkable sub-criteria. Weights are fixed and public.

D1weight 20%

Liquidity Depth & Exit Capacity

Can the protocol actually liquidate this at scale? Depth, venues, slippage at size, time to exit. The RWALS DNA.

D2weight 15%

Wrap & Rehypothecation Depth

Every wrapper is a new way to die. rsETH reached Aave wrapped four times — the outermost wrap is what broke.

D3weight 12%

Bridge & Cross-Chain Exposure

Shared-supply tokens inherit the weakest verification path on any chain. Kelp's Unichain route was a 1-of-1 DVN.

D4weight 13%

Oracle Integrity

Self-referential pricing — internal exchange rates, linear discount models — with no external market check.

D5weight 12%

Redemption Posture

Is there a par redemption path, at what notice, gated by whom? An AMM exit is not redemption.

D6weight 8%

Duration Risk

A 14-day principal token and a 365-day principal token are not the same asset. Spot collateral scores 100 here.

D7weight 12%

Backing Verification

Who independently verified the backing exists — reserves, custody, or the physical asset itself, on a fixed cadence?

D8weight 8%

Operational & Governance Risk

Admin keys, upgrade powers, curator concentration, withdrawal-freeze authority, regulatory posture, incident history.

Five hard gates that cannot be averaged away

Averages launder fatal flaws. A perfect liquidity score does not excuse a single-attestor bridge. Gates are never waived, for any asset, for any reason.

G1Effective wrap depth of four layers or moreINELIGIBLE
G2Any live bridge route securing the asset's supply verified by a single attestorINELIGIBLE, EVERY CHAIN
G3Off-chain or physical backing with no independent, recurring verificationCAPPED AT TIER 3
G4Price oracle is self-referential with no external market checkCAPPED AT TIER 3
G5Time-to-maturity under thirty daysWEEKLY LTV STEP-DOWN

One formula. No cliffs. No committees.

Bucketed frameworks create cliff effects between adjacent scores. CRS maps score to maximum LTV continuously. Protocols stay sovereign — governance may always set parameters below the ceiling. The standard never tells a protocol to take more risk; only where risk ends.

max LTV %  =  0.75 × CRS + 20   for CRS ≥ 40
max LTV %  =  0   for CRS < 40, or any fatal gate
Scores fall by machine — monitored triggers, evidence published at the moment of change. Scores rise only by published review. Risk is asymmetric; so is the standard.

The board

Reference scores, methodology V1.0, illustrative published inputs. Two rows are backtests — assets scored exactly as they were configured on the day a protocol listed them.

AssetCRS ScoreTierMax LTVGatesStatus

Commandment III, made literal: every number above is computed in your browser, right now, by the open scorer embedded in this page, from the inputs shown in each breakdown. View source — that is the methodology.

The Ten Commandments of CRS

Not values — constraints. Anyone can verify compliance with every one of them. A violation of any one voids the standard's claim to neutrality.

The methodology shall be public, complete, and free — forever.

No black box, no premium tier of truth. Rubric, weights, gates, and reference scorer: open source in perpetuity.

No issuer, protocol, or curator shall ever pay for a score.

The ratings-agency conflict — graded by the graded — caused 2008. It will not be rebuilt here. Scores cannot be bought, expedited, or sponsored.

Every score shall be reproducible by anyone, from published inputs.

Run the open scorer on the published inputs and you get the published score, bit for bit. If you cannot reproduce it, the score is invalid.

No strength shall hide a weakness.

Hard gates exist because averages launder fatal flaws. Gates are never waived — for any asset, for any reason.

Scores shall change when reality changes — not when committees meet.

Monitoring is continuous, downgrades are automatic. The standard does not wait for a governance call to acknowledge a fact.

Every downgrade shall ship with its evidence.

Trigger, timestamp, data source, and the exact rubric line invoked — published at the moment of change, permanently, on the record.

Disputes shall be heard in public and resolved on the record.

Any issuer or community may challenge a score. The challenge, the evidence, and the ruling are all published. Nothing is settled in private.

No single party — including the author — shall hold the power to bend a score.

Capture resistance by governance design: methodology changes require public proposal, open comment, and versioned publication. There is no override key.

The standard is free; only the rails are paid.

Sustainability without corruption. Methodology and scores are public goods. Revenue comes only from delivery — API, on-chain feeds, integration — never from the scores.

Every failure becomes a factor.

Each industry post-mortem must be scoreable retroactively — with the standard catching what was missed — or the standard amends itself in public. V1.0 is backtested against rsETH, wrsETH, Elixir/Steam Labs, the wstETH misalignment, and Tangible/USDR. Every future incident joins the test suite.

Free standard. Paid rails.

Commandment IX is the business model. The truth is free; the plumbing that delivers it in real time is not.

Free, forever

  • The full methodology — rubrics, weights, gates, version history
  • Published scores with their complete input sets
  • The reference scorer — open source, runs anywhere, including this page
  • The dispute process — open to any issuer or community