Cats as Ethereum validators

The case for a lean Ethereum

Succinct verification is an accessibility technology. It empowers a cheap/weak machine to quickly verify any expensive/heavy-duty computation it cannot afford to perform. We are talking about several orders of magnitude in computational power difference. One of the potent use cases is scaling Ethereum. That is, more gas per block while honoring the 12 seconds finality. The current Ethereum design relies on validators that “re-execute” blocks to reach consensus. This is rather inefficient and there are active initiatives(like EthProofs) to improve the situation. Today, it takes 16 RTX 5090s(~1,677 TFlops) to prove 99% of 45M gas blocks. The achievement is commonly referred to as Real Time Proving(RTP). It allows the network to upgrade to a “prove once + verify by many” consensus model. Instead of (wasteful) re-executions, prove blocks once and let the public verify them. A win for decentralisation too since each proof would receive thousands of independent verifications. While 1,677 TFlops computing power is required to prove a block, verification takes less than a second on a low-end mobile phone. Kudos to recent advances in proving technology! The proof generation process is still centralised though. This is due to the large size of the intermediate objects, often hundreds of MBs in size, that arise and have to be passed around during the process. It is thus not trivial to distribute the task in a divide and conqueror fashion over the Internet. Let’s accept it as is for now and just stay focused on the big picture: to replace fat validators with lean machines for block validation while increasing network’s capacity.

Gigagas 2028

Ethereum yearns for Gigagas blocks. That is, to inflate block gas limit to 1B in three years. As of now, post Fusaka hardfork, the value sits at 60M. The plan is to increase it by 3x per year. So, by the end of 2028, Ethereum could handle 1B gas blocks. A gigantic stretch. The immediate question would be, can we handle 1B gas and generate proofs in time? The proving rate in certain applications(like Fibonacci) is improving by ~10x per year. The rate for serious ones i.e. zkEVMs is much lower. And it has still got some room before progress decay kicks in. Prove-friendly hashing and instructions are also being introduced to the EVM that would lead to an even smoother transition. We may also see surprise prove-maxi non-EVM target ISAs.

It is Q2 2026. A single RTX 4090 proves ~250KGas/s on Reth-RSP(unified blocks approach) via Succinct’s SP1 Hypercube. Reserve 2 seconds for initial block setup and 10 seconds is left for proving. Each 4090 can contribute up to 2.5MGas to proving. We have got 3 years and have to extract 4,000x more to secure Gigagas RTP on a single 4090. This translates to 16x annual improvements compounded for 3 years. Extremely hard but achievable. OK, how about going solo with an entry-level Macbook Air instead of 4090? Well, the poor M5 MBA does 4 TFlops compared to 4090’s 80 TFlops. We would need to introduce an additional 20x improvement on paper. Or link up to 30 MBAs in a p2p proving copium layout like the one we are doing with Wholesum. I would go with the p2p network option because it brings decentralisation guarantees as by-product. Setting up prover ASIC farms is another interesting angle but it is somewhat distant due to the frequent introduction of hardforks.

We have to innovate in 3 areas: the EVM, zkEVMs, and general proving including ASICs.

LTS Hardforks

Ethereum had two hardforks in 2025. This frequency makes it impossible to maintain dedicated ASICs for proving blocks. Ethereum is evolving into a succinctness-friendly environment. This implies more frequent upgrades to the core in the near future. This is not what hardware investors or engineers would accept. ASICs are always the fastest in any application domain. Two hardforks per year would render any ASIC design obsolete by the time circuits and field elements are etched onto the silicon. Proving is a young technology and its adoption by Ethereum has made the situation somewhat complex. Teams are really interested to come up with (decentralised?) ASIC proving farms and go big with scaling. Let them do it. The result would be an even more diverse proving landscape. I just hope that we see less frequent updates to the core as time goes by.

A notable example is PlayStation with its ~5 years release cycle. If adopted, one hardfork per 5 years would be great for AISCs to catch up and shine. It is also beneficial to other domains. Note please that general-purpose ASICs that can be reprogrammed mid course are not of my interest here. I would rather look for hardfork-specific ASICs that fall out of fashion when the next hardfork shows up. We have got three years to Gigagas era. And can fix a design spec, accumulate as many changes to the core as possible, and then announce the 5 years release cycle. Looking forward to an immutable Ethereum.

The Interface

New tools have clunky interfaces. Subsequent upgrades improve on humane factors like time expenditure, handling, cost, and etc. Computer is indeed a tool, a fast-evolving one. Alan Kay, one of the early computing pioneers, refers to this man-tool relationship as body part extension: axe handle as the extension of the arm, computer as the extension of the brain, etc. He, and others, then warn about the after effects of adopting newly-introduced tools. The famous quote attributed to him, “we shape tools and tools shape us”, implies the potential of tools to change the fabrics of society. Let’s project Ethereum scaling onto this plane of thought and speculate about it. Here, I would like to approach verificaion as an interface design problem.

Verifiably scaling Ethereum has two stages: proving and verification. We can outsource proof generation over the Internet(e.g. 16 RTX 5090s) in time for RTP. For verification, the immediate solution that comes to mind is to ask the public to verify proofs. Setup a web interface, supply proofs, and some push buttons to initiate verification. The whole procedure won’t take more than half a second. This guarantees a decentralised path to block validation. Unlike proving which demands extremely large compute capacity, verification is easy for any computer produced in the last 20 years. But wait, can we push the line a dozen million years back and get primitive computers to verify proofs? It is an optimization rabbit-hole. A race to find the next local minima. So, let’s see how it might unfold.

Cats as validators

Back to the future of cats securing Ethereum.

Cats are non-deterministic mobile maverick computers. They are narcissist and manipulate us despite being quantitatively weaker. They used to own Egyptians, own the Internet, and would inherit Ethereum. Computation-wise, they are several orders of magnitude weaker. Earlier, I claimed that succinct verification is an accessibility technology. We may never be able to get a cat to prove 1B gas. Verification, however, is much less involved compared to proving. What is needed to get cats involved is a couple of technological breakthroughs in neuroscience, BCI, coding theory, and communication complexity. We can then happily delegate proof verification to them and retire the entire human validator set of Ethereum.

Two sets of gadgets have to be invented: physical and software. For the cat, a small non-invasive gadget similar to earbuds could be mounted on the head to decode human requests. It would gather vital organ statistics en route. She should have the option of opting out. The design should be ergonomic and gentle so that her well-being is respected. The design should strictly follow Dieter Rams’ principles of good design too. Remember, the overall goal is to augment cats’ intellectual abilities in a cooperative fashion. The actual result of verification can either be a set of electronic messages or encoded mechanically and taped in special cabins spread out across neighbourhoods.

For the human part, a software package is all that is needed. A light-weight cat-human communication protocol that encodes human requests and decodes cat responses. It should monitor and collect cat responses, analyze them, and submit verification outcome to the network.

Without a proper detailed image of the cat brain we will not succeed. Mapping the entire brain and its messaging nuances is thus an essential prerequisite. Recent advances in neuroscience and AI are heart warming but not enough for us to start. I think at least 75% of our energy would have to be spent on studying cats. This is obvious from the description of the expectations we have from the to be invented gadget(s).

We would need a place like Janelia farm equipped with state of the art neural and computer labs. It should feel like Xerox PARC of 70s. During phase I, the first 5 years, we would hire a few founding scientists. To honor Bob Taylor’s legacy, only the best would be hired. The goal is to lay out the plan for phase II which would stretch to the next 10 years. In phase II the founding scientists would slowly grow their group and start the actual research work. I expect for each lab to grow to 16 members by the end of this period. The goal is to come up with blue prints for gadgets and protocols. The next phase, III, would take 15 years to complete. The objective is to build and synthesize gadgets and undertake early cat-human-verification testing. The final phase, IV, should take up to a year to finish. The goal is to publish and present products to the public. In total, I expect the project to take 31 years to complete. So, by 2027 + 31 = 2058, can we rely on cats to validate blocks? I hope so. This is going to be my life-time project.

A validator set composed entirely of cats is the climax of decentralization. You can’t bribe cats to collude with proof verification. This is a somewhat unsound claim though. If we have the technology to get cats to verify proofs, it should not be that hard to teach them to act maliciously.

May the gods in cats’ clothing bless Ethereum and restore the holy Saveh lake to her prime days.