AVS scoring: professional operator TVL vs community TVL
How AutoLayer planned to rate each Actively Validated Service using two stake measures, and why the split between professional and retail backing is a useful risk signal.
AutoLayer Research · Published Oct 15, 2026 · 6 min read
Source note. The AutoLayer-specific claims in this article come from the team’s application to the Arbitrum LTIPP, dated 26 February 2024. It describes AVS scoring as a feature of AutoLayer V2, which was in preparation at that point. We describe it as “planned as of February 2024”, not as a live product. The background definitions are general and uncontroversial; we could not reach EigenLayer’s documentation when preparing this piece, so we have kept them short. Not financial advice.
The problem: yield without a risk label
Restaking lets a staker reuse staked ETH (or a liquid staking token) to secure additional services. In EigenLayer’s vocabulary those services are Actively Validated Services, or AVSs. Operators run the software for an AVS, and stakers delegate to operators. In return the staker may earn extra rewards. In exchange, the staker accepts extra conditions under which part of the stake can be penalised, which is slashing.
The difficulty is that each AVS is a different piece of software with a different task, a different team and a different failure mode. A staker choosing between them has no standard rating. The application argues that AVSs are “poised to become the linchpin of yields” and that evaluating them is therefore essential. Its answer is a risk score for every AVS shown in the AutoLayer interface.
The two inputs
The post defines the score through two measurements. Both are about how much money is backing the AVS, and by whom.
| Metric | What it measures | What it tells you |
|---|---|---|
| Professional Operators TVL | Amount staked on the AVS by institutional and professional operators. The post names P2P, Ankr and Nethermind as examples. | Whether firms that run infrastructure for a living have put capital behind the service. |
| Community TVL | Amount of retail money backing the AVS. | How much support the AVS has gathered from individual stakers. |
The stated reasoning for the first metric is that it produces “a risk profile based on factual data from funds deposited by professionals, rather than subjective opinions”. The second metric adds the retail side to complete the profile. The post also says that sudden changes in these scores can help users stay informed.
Why the split can carry information
Total value locked on an AVS is a single number. Splitting it in two says more.
Professional stake as a signal of diligence
Operators that run validators for a living can lose revenue and reputation if an AVS misbehaves. When such an operator commits stake to a new service, it implies someone with technical knowledge looked at it. That is not proof of safety. Professionals make mistakes, and they can also follow incentives that retail users do not see. But a service with no professional stake at all is at least a different category of risk from one with several established operators.
Community stake as a signal of reach, and of herd behaviour
Retail money can arrive fast, especially when rewards are advertised. Large community TVL with little professional TVL is a pattern worth noticing: it may mean strong marketing, or it may mean capital chasing yield without scrutiny. Read the other way, strong professional TVL with a small community TVL can mean an early service that professionals trust before the public has found it.
The ratio and the movement
The application does not publish a formula, weights or thresholds. We therefore do not know how the two figures would be combined into a single score, and we will not invent one. What the post does commit to is the inputs and the idea that movement matters. A score that drops because professional operators withdraw is arguably more informative than a static label.
A thought experiment with hypothetical numbers
Imagine two AVSs, A and B, each with the same total stake. A has most of its stake from three well-known operators and a small amount from retail stakers. B has almost no professional stake and a large retail base drawn in by high advertised rewards. These are hypothetical examples, not real services.
On a TVL ranking alone, A and B look identical. With the two-metric view, they look different at a glance: B is backed mainly by people who may not have reviewed the service closely, and if rewards drop, that stake could leave quickly. A is not safe because of its operators, but the evidence behind it is of a different kind. That difference in evidence is what the score is meant to surface.
What a score cannot do
A stake-based score has clear limits, and users should hold them in mind:
- It does not audit code. Nothing in the two metrics reads the AVS’s contracts or slashing conditions.
- It measures money, not correctness. A heavily staked service can still be flawed.
- Professional means self-identified or listed. The score relies on classifying operators as professional, and the post does not explain how that classification would be maintained.
- It describes the past. Stake can leave as quickly as it arrived.
None of this makes the idea weak. A transparent, data-based first filter is useful in a market where most decisions are made from a Twitter post. It just should not be mistaken for a guarantee.
How it fits with the rest of the V2 plan
AVS scoring is one of four planned V2 features in the application, alongside LRT scoring, compounding and liquification, and slashing bounties. They address different layers of the same risk:
- AVS scoring looks at the services being secured. This article.
- LRT scoring looks at the liquid restaking token itself and how it is reused as collateral.
- Slashing bounties aim to halt an AVS before a slashing event reaches stakers.
The post also describes a Vault Manager where each AutoLayer Operator validates its own set of AVSs, producing vaults “characterized by varied yields and risk profiles”. A score would give that choice a basis. Read more in slashing bounties and Virtual Operators.
How to use the idea today
You do not need a product to apply the logic. When comparing any AVS, a few questions follow directly from the two metrics:
- Which operators are running it, and are they names you can verify as established firms?
- How much of the stake comes from them versus the general public?
- Has either number changed sharply in recent weeks, and why?
- What exactly can be slashed, and under which conditions?
The fourth question is the one a stake score cannot answer for you. Read the AVS’s own documentation and the risk disclosure before committing funds. For the wider design of the product, see how AutoLayer works and the security page.
Planned as of February 2024
To keep the status clear: in the February 2024 application, AVS scoring is listed under “AutoLayer V2 Features”, which the team describes as in preparation. The post says the scores will be generated “for every AVS on AutoLayer”. It gives no launch date for this feature. Anything beyond that, such as a formula or a user-visible scale, is not in the source and should not be assumed.
Takeaway
Splitting AVS stake into professional and community TVL turns one blunt number into two readable signals. The approach is simple, based on observable on-chain data, and honest about being a risk indicator rather than a verdict. Its value will depend on how operators are classified and how the score is presented, details the source does not give. For the team’s wider background, read From Tortle Ninja to AutoLayer.