In this note, we will first comment on the incentives for generating evidence, and then speculate on the types of evidence that may be presented to the dispute resolution mechanism. This is necessary in order to determine what the typical court case may look like, which will in turn play a role in estimating some key court parameters.
We aim to decentralize evidence generation via the dispute resolution mechanism itself. Any party that submits a successful case to our courts will gain an economic reward (which comes from an operator’s bond) and is therefore incentivized to do so. The reward will be scaled as a function of the number of validators found guilty of white-labeling—providing even better incentives for cases that uncover larger operations.
Details are pointed out in the note Bonding requirements for operating the courts. Accusers will have at least two mechanisms to submit a case:
We expect to see models and classifiers that employ validators’ features (based on performance and networking data) in order to cluster these validators into their possible operator sets. For more details, check Heuristic approaches related to white-labeling detection.
From the note above (as well as discussions with the Lido team), we have noted that heuristic approaches can lead to a “cat-and-mouse” game, where white-label operators find ways to circumvent detection from these models, and model creators find new features or methods to continue to expose them. Models can become obsolete quickly in this battle of iterations, and they can be bolstered by technical advances that are not as widely used now but may flourish in the future (such as zkML for privacy or verifiable heuristic predictions)
Therefore, in order to deal with this dynamic landscape, the white-labeling protection mechanism should not tie itself to a specific model. Instead, it should encourage investigators from the outside community to generate their own heuristic tools by means of the aforementioned financial incentives.
In the beginning, and in order to bootstrap this mechanism, the Lido DAO can take advantage of preexisting collaborations with teams that are proficient in validator data analysis, such as Rated and Miga Labs. These teams have built models and crawlers that can be adapted to the task of white-labeling detection. Understandably, these teams may initially not want to disclose all the information about their models to other parties, since this could cause them to lose their technical edge.
To deal with this, we propose to kick-start this market by means of an experimental phase that rewards model generators. Details in the note Estimating probability parameters for economic analyses.
White-label agreements are expected to be private and founded on trust. As such, much of the relevant information will not come out to the public unless a party associated with the white-label operator (e.g. an operator’s employee) explicitly reveals the agreement, whether in the form of a contract or in private communications.
This could happen, for example, due to moral reasons. However, a stronger guarantee may be provided by the financial incentives described above.
In our note on bonding requirements, we pointed out how the reward $R(m)$ for an accuser will scale with the severity of the case, to the point where it can fund complex investigations for high-profile cases. This reward can be used to directly incentivize whistleblowers who have access to highly incriminating evidence. Moreover, the figure of the Lido prosecutor can be used to preserve the whistleblower’s anonymity.
Another form of evidence to consider is any “money trail” connecting to the white-label operator. Unfortunately, many of these white-label providers accept fiat and off-chain as a source of payment. References to these kinds of payments are likely to fall under the type of whistleblower evidence described before—it is not expected that an external party will get access to these.
However, some staking providers do accept on-chain payments. For example, StakeFish allows delegators to provide them with 25% of the execution-layer rewards as payment for running the validator. Other operators simply accept crypto payments to addresses controlled by them. An accuser may surface any of these payment events as supporting evidence, and the size of the payment can be used to estimate the severity of the white labeling.