Methodology
How the registry is built — how protocols and feeds are chosen, how we stay neutral, and how every figure here can be checked against what actually happened on-chain.
The oracle-diversity principle
No single feed should be canonical for something as important as protocol risk. We treat risk feeds the way Ethereum treats price oracles: diversity is the point. We put what each one publishes side by side, attributed and verbatim, and let you weigh them. The aggregation is the value.
What we do
- Show each feed’s coverage status per protocol — covered, partial, or not yet covered.
- Show what each feed publishes, verbatim and attributed, with a link to source.
- Surface governance facts with a provenance tag on every datum.
- Sort and filter by neutral facts: TVL, category, feed-coverage count.
What we do not do
- Produce a composite or proprietary risk score of our own.
- Combine feeds into a single derived signal.
- Adjudicate when feeds disagree — we show the disagreement.
- Rank protocols by any risk judgement.
This constraint is binding and documented in the project charter. Changing it requires written agreement from the Ethereum Foundation.
Opinion vs. ground truth — the verification layer
The feeds publish assessments — diverse, subjective, shown side by side. Separately, the registry is stewarded by the team at Rails, which renders the full event-level on-chain history of each protocol, so every claim here can be checked against what actually happened. The feeds tell you what they say; the verification layer shows what is.
To stay neutral, the verifier is deliberately nota rated feed — a steward grading itself alongside the feeds wouldn’t be neutral. Rails is infrastructure, earns no rating column, and the relationship is disclosed in DISCLOSURES.md.
How we select protocols
The registry covers 20 protocols across 26 versioned deployments — the RFP’s seed protocol list, populated in full. That list names the top Ethereum DeFi protocols by TVL (or 24-hour volume where TVL doesn’t apply), prioritising those where user capital is directly at risk, and we adopt it verbatim: all 20 protocols across every category (Lending, DEX / AMM, Swap Aggregator, Yield / Vault, Liquid Staking), including each version the RFP names (Aave V3/V4, Compound V2/V3, Liquity V1/V2, Uniswap V3/V4/X). Rankings are re-verified against DefiLlama at build time, as the RFP instructs. We add no selection criterion of our own.
As the registry expands beyond the seed list, new members are admitted on the RFP’s own stated basis— TVL, or 24-hour volume for routing/intent venues where TVL doesn’t apply (CoW, UniswapX, 1inch, 0x, where funds don’t sit inthe protocol), prioritising where user capital is directly at risk — ranked on neutral DefiLlama facts. As live metrics move, this would naturally pull in large restaking and synthetic-dollar protocols; the basis is the RFP’s, never a risk judgement of ours.
The seed set already spans the full control spectrum, from large upgradeable lending markets to immutable CDPs like Liquity. We surface that as a verifiable property, never a selection input: an immutable protocol — core contracts no admin or governance key can change — has an empty control timeline, which our verification layer renders. Immutability is shown as a structural fact, never a mark of safety.
Versioned deployments are listed as their own rows (e.g. Aave V3 and V4, Uniswap V3/V4/X), so each carries its own governance and coverage; the 20 protocols therefore surface across 26 deployments. A protocol may appear in more than one category where it runs distinct products (e.g. Morpho lending vs. Morpho Vaults).
How we select feeds
The registry includes 16 feeds chosen to maximise methodology diversity — formal rating desks, dashboards, security-monitoring tools, and research. We also track whether each exposes machine-readable output (13 of 16today): the basis for our M2 automation — automate where possible, curate by hand where not. Each feed’s focus, type, independence, and coverage lives in the feed directory.
Curation is deliberate, and we document the calls. We scope each provider to where it actually publishes: a service that rates stablecoins rather than protocols is held for the asset-risk layer rather than padded into this matrix, and where one dashboard is published under two names it is counted once. The goal is a registry that reflects distinct methodologies, not an inflated provider count.
Corrections
Every datum is community-correctable. To fix a coverage status, a stale rating, or a governance detail, open an issue or PR on GitHub citing the source.