[Uber tested withdrawing funds from its blockchain on weekends; Primus has also submitted a development proposal] @primus_labs @CantonNetwork Uber tested whether it could withdraw funds from its investment pool outside regular banking hours—specifically on weekends. In a demonstration announced on October 30, Uber purchased shares in Franklin Templeton’s money market fund using HSBC’s tokenized deposits and confirmed it could partially cash out those holdings even on weekends. The blockchain used for this transaction was Canton Network. Primus and HashCloak have submitted a joint proposal to integrate zkTLS data verification technology into the same Canton development fund. The requested amount is 1,303,500 Canton Coins (CC), equivalent to approximately $150,000 at the time of submission. The proposed development timeline is 14 weeks. Uber’s experiment focused on the process by which corporations invest funds and later retrieve them when needed. Primus’s proposal aims to enhance the Canton app with the ability to verify whether external financial conditions are met—without exposing the original data. For example, suppose a company needs to verify that its counterparty meets certain financial requirements before executing a transaction. Instead of receiving and storing full account statements, the system would verify a proof based on authenticated data from approved sources. Primus plans to generate proofs for web-based data, while HashCloak will develop tools enabling the Canton app to directly verify these proofs. The public proposal outlines that the Canton app will be able to independently validate mathematical proofs without receiving the original data, and that the resulting tools will be open-source and reusable by other developers. Implementing this would allow financial verification apps to validate proofs directly—rather than relying solely on Primus’s signature. Uber’s weekend fund usage is a completed demonstration; Primus and HashCloak’s proposal is currently under review. Both address distinct challenges within the same Canton ecosystem. The next step is determining whether external financial verification capabilities can be securely integrated into transaction infrastructure. What matters most after this is not the funding amount, but whether the verification functionality actually works. A public test demonstrating that the Canton app can directly validate Primus’s proofs without accessing the original financial data will serve as the key indicator of progress. #Primus #CantonNetwork

[Why Receiving an AI Judgment on Solana Doesn’t Mean Settlement Is Complete] @GenLayer Let’s assume Solana’s AI agent marketplace delegates quality judgment of tasks to GenLayer. Even if validators reach consensus on the results, determining which Solana account should receive the reward is a separate issue. The bridge code published by the GenLayer Foundation outlines a structure that transmits messages between GenLayer and Solana via LayerZero V2 and an EVM hub. Here, the location where the AI judgment is executed is decoupled from the location where the app’s state is updated based on that result. How Solana receives messages depends on the account configuration: If the receiving account is pre-determined, the registered receipt state can be updated directly via the DirectCpiReceiver path. Actual external program calls must be additionally implemented to match the receiving program. However, if the account varies per order or user, messages are first stored via StoreAndClaim, and the required account is specified during the subsequent claim_message call. Solana requires a predefined list of accounts to be specified before executing a transaction. Therefore, merely receiving a message via the bridge does not mean settlement has been completed for the user. The origin of the message is also independently verified. The public implementation is designed to validate trusted peer sources and prevent duplicate processing of identical requests using a unique message_id. However, preventing duplicates and ensuring proper message delivery does not constitute proof that the AI judgment itself is correct. When I developed PolicyGuard in Studio, I addressed the question of whether a document’s rules had actually changed. This bridge code confirms that connecting results to services on other chains requires designing not only the judgment criteria but also the message transmission, receipt, and settlement states. While the public repository contains bidirectional test logs between Base Sepolia and Solana Devnet, these cannot be accepted as operational proof for commercial Solana services using GenLayer AI judgments. Success in cross-chain AI applications requires three distinct proofs: Final confirmation of the GenLayer judgment, verification that the message originated from a trusted source, and actual state change in the destination account. Public code: https://t.co/N4zVSxjYG7 #GenLayer #GenLayerDOKDOPioneer

[What Happens When One ENS Name Manages Thousands of Accounts] @ensdomains @ethereum With the launch of ENSv2 on October 27, https://t.co/NvxB4rnXZr.eth and https://t.co/xwZsp1UT6i.eth can now be managed by different owners—allowing companies to grant specific teams only the permissions they need, without transferring full ownership of company.eth. This structure introduces a critical condition: a single domain can become a single point of failure for an entire service. If the parent name expires, all subordinate names beneath it become unreachable. ENS is known as a service that replaces long wallet addresses with human-readable names like alice.eth. In October, ENS disclosed it had approximately 905,000 .eth names and 34 million subordinate names. The total of over 36 million names includes both directly registered names and subnames issued by organizations. The model of using one name to manage multiple users and services is already in widespread use. ENSv2 restructures the foundation for managing these subnames. Each name can now maintain its own subname registry, and owners can assign granular permissions—such as registering new subnames, editing profiles, or managing subnames—based on specific roles. For example, an organization can issue individual names to employees or assign editing rights per team. While subnames existed before, ENSv2 integrates these management capabilities directly into the core contract architecture. On the flip side, responsibility becomes centralized. When resolving a name, ENSv2 traces up the chain from parent registries. If company.eth expires or its subname registry connection is severed, even possessing https://t.co/xwZsp1UT6i.eth won’t allow you to resolve its payment address. The ability to manage a subname is now tied directly to the parent name owner’s renewal and permission management. Expiration rules further intensify this responsibility. ENSv1 allowed 90 days of grace after expiration during which names remained resolvable. ENSv2 reduces this grace period to 28 days—and resolution stops immediately upon expiration. Renewing within the grace period restores functionality, but any lapse may disrupt connected services. When migrating existing names to v2, the expiration date is extended by 62 days to offset this difference. For example, if a v1 name had 100 days remaining until expiration, after migration it will have approximately 162 days left, and the point at which others can re-register remains around 190 days later for both versions. While re-registration eligibility aligns, v2 names become unresolvable during the final 28 days. The risk of losing the name to another party and the risk of temporarily disabling services now follow different timelines. For an individual’s personal nickname, this may simply be a reminder issue. But if an organization has issued thousands of subnames to customers, employees, or agents, a single failure to renew the parent name can disrupt access to multiple services. The more finely grained the role-based permissions, the more critical it becomes to assign clear responsibility for parent name renewal and establish emergency recovery procedures. Registration pricing changes are aligned with this design. The annual base price for .eth names of five characters or more has increased from $5 to $8—a 60% rise—while registering for six years or more reduces the annual cost to approximately $4.50. This pricing structure incentivizes long-term ownership over short-term use. However, as more accounts become linked to a single name, the cost of a failed renewal far exceeds the registration fee. Whether ENSv2 succeeds cannot be measured solely by new name registrations. More compelling indicators will be the actual adoption of subnames by organizations, the adoption of role-based permission delegation, and the growing number of apps capable of reliably resolving these names. ENS already supports name resolution for over 600 apps. Going forward, who takes responsibility for operating and maintaining these names will determine the protocol’s trustworthiness. #ENS #Ethereum

Questions AI research labs will receive before purchasing 12,480 data records @datafdn @SuiNetwork [Three stages of data purchase connecting DATA’s Trace with Sui and Walrus] Suppose an AI research lab is seeking data for voice training. If a seller presents a dataset of 12,480 records, the researcher cannot determine the actual volume of desired materials simply by counting total files or reviewing contracts. In fact, the audit example shown by DATA Foundation’s Trace—1.42TB in size—contains 6,140 videos, 3,770 images, 1,710 documents, and 860 audio files. For a lab seeking audio data, it would begin by examining the 860 audio files, representing approximately 6.9% of the total. These figures are not actual transaction results but demo data from Trace—useful for illustrating what information buyers should review. After filtering by file type, researchers must verify the provider’s identity, consent and terms records, and the integrity of delivered files. On the same demo screen, a KYC verification rate of 96.4% refers to contributors, while a latest terms adoption rate of 91.8% reflects the scope of consent version application. Since these metrics use different criteria, buyers must separately assess the provider’s authentication status and terms records per file. The collaboration between DATA Network and Sui, announced on October 12, 2026, aims to link this review process directly to original files and usage rights. At the time of announcement, Trace had registered over 250 million AI data contribution records and processed over 3 million receipts daily. The Sui Foundation invested $30 million in DATA Network, and DATA plans to gradually migrate from its independent blockchain to a Sui-based protocol. [Five steps preserved in a 19.1KB receipt] A single Kled receipt published on Trace contains the SHA-256 hash of a 19.1KB HEIC image, resolution 196 × 348, provider KYC status: Verified, and upload/signature/registration history linked to an on-chain transaction. The upload time was October 6, 2026; provider signature and Trace registration occurred on October 7. This page displays registration history and verification items—not the original image. This receipt’s lifecycle consists of five stages. The “Captured” stage—referring to image capture—is marked “Not recorded yet.” While detailed data includes fields for file occurrence time (occurred_at) and upload time (uploaded_at), these alone cannot confirm the actual camera capture moment. Terms of use are marked “Not recorded” under Origin and “Current” under Identity & Consent versions. These distinctions help buyers break down their review criteria precisely: The hash serves as an identifier to verify what file information was registered; signatures and on-chain records provide evidence of who registered the data and when. Provider authentication status and terms version are separate criteria. The researcher’s right to use data for training must be determined based on actual consent content and licensing agreements. [Transforming one receipt into ten thousand reviews] On Trace’s Verify interface, users can paste up to 10,000 data IDs at once or submit larger lists via CSV. Trace aggregates receipts from a dataset to summarize provider identity verification status, terms version, media type, and modification history, allowing users to drill down into individual records. In the earlier demo of 12,480 records, combining videos and images yields 9,910 entries—roughly 79.4% of the total. A lab seeking audio data and another seeking video data would examine entirely different subsets from the same sales listing. Displaying provider verification rates alongside terms adoption rates enables buyers to select files matching their purpose and then locate corresponding verification records. Trace also includes records from Numo (handling audio and video), oto (handling conversational audio), and other providers. Receipts registered by each app are bundled into Merkle roots and anchored on-chain. Thus, daily processing of over 3 million receipts does not equate to 3 million individual on-chain transactions. This architecture enables registration of massive contribution records while preserving audit trails for individual datasets. [How review changes when records connect to original files] Currently, Trace stores receipts and associated metadata but does not host original files. In announcing its collaboration with Sui, DATA Foundation proposed integrating Walrus as the default storage layer for its application, linking encrypted references to original files with Trace records.For example, if the laboratory receives a file purchased from a provider, it can design a procedure to obtain the reference file from Walrus and compare its hash. This provides evidence to determine whether the content was altered during transmission and how the subsequently processed file differs from the registered reference. Since the reference itself is initially submitted by the provider, this comparison serves as proof of file identity but does not automatically guarantee the authenticity of the original filming conditions or the original consent. Data stored in Walrus is publicly accessible by default, so private originals must be encrypted prior to storage. Sui ecosystem’s Seal combines client-side encryption, Move policies, and a key server to implement conditions under which decryption keys may be granted. A service can be designed to provide keys exclusively to license holders, but it does not include functionality to retroactively revoke already issued keys. Legal usage scope must be separately documented in license and consent records. Trace serves as the layer for searching registration and consent histories; Walrus stores encrypted reference originals; Sui and Seal enable datasets to be treated as programmable assets and enforce access conditions. Trace’s registration and auditing functions are already operational; integrating Walrus-based original storage and Seal-based purchase access control into this service is planned for future implementation. [Decision to Migrate from a Custom Chain to Sui] DATA Network has explained that it is migrating to Sui to reduce the burden of maintaining an independent L1 and enhance scalability. The existing network will continue operating until the transition is complete; $DATA balances will be transferred 1:1, with locking and vesting conditions preserved. The adoption of Walrus and the network migration will proceed incrementally; detailed timelines will be announced separately. DATA will migrate existing registration records from Trace and data provider applications already in operation into the Sui ecosystem. Sui can represent datasets and licenses as programmable objects, while Walrus can store large-scale originals. This shift also frees up development resources previously dedicated to chain operations, allowing greater focus on products for data registration, licensing, and purchase verification. The success of this combination cannot be measured solely by how quickly registration receipts accumulate. More direct product outcomes include whether actual purchasers can select desired files, verify provider authenticity and consent records, compare received files against reference originals, and use data under contractually defined access permissions. The 12,480 demo audit screens demonstrate how purchasers interact with datasets. The 19.1KB receipt shows how thoroughly one can trace the registration history of a single file. The collaboration between DATA and Sui aims to connect these two audit pathways with original storage and access control. If data suppliers for AI training can present records and purchasers can jointly review those records alongside the actual files, then hundreds of millions of registration receipts can become infrastructure usable within individual purchase contracts. #DATAonSuiQuest

[Why Kaito Open-Sourced Aura] @KaitoAI @KaitoStudio_ Even if someone on X says, “I’m confident in this trade,” it’s hard to know whether they’ve actually put money on the line. This is the information gap Kaito aims to close with Aura. It seeks to connect what people say on social media with their actual on-chain activities into a single, verifiable record. Aura is Kaito’s reputation system. It links the influence built on X with verifiable trading activity to present and evaluate a user’s full history. Kaito introduces Aura as a feature that assesses verified trading behavior and assigns value to real-world actions. Users can connect their accounts on the Aura page to review their history; accumulated metrics over the past 48 hours may be recalculated. For example, imagine someone who frequently posts predictions on Polymarket. You can see how widely their posts are shared through X reactions—but whether they actually took positions is separate information. Kaito Pulse displays connected trading data alongside X posts, while Aura uses these activities as components of reputation. Aura Arc maps relationships between connected activities across multiple services and social accounts in a graph format. Service connections are user-selectable, and the scope of publicly available data varies by partnered apps. In short: Pulse shows activity history directly on social feeds; Aura treats that history as reputation; and Aura Arc attempts to extend this relationship model to other services. Kaito has committed to offering Aura Arc for free and enabling its use by external products. This decision to make it free may be strategically calculated: the more widely such shared data on who has built trust in which domains is adopted, the greater the demand will become to identify suitable individuals or execute collaborations. Studio helps brands select creators and manage contracts and payments; Katalyst links campaign outcomes—such as clicks, sign-ups, and deposits—to rewards. For Kaito’s Jumper campaign introduced on October 22, 2,065 creator applications were received, 69 were contracted, and they generated over 760,000 impressions across 119 posts. This demonstrates Kaito’s ability to mobilize people and run campaigns. However, Aura also faces challenges. Verifying trading records is only the beginning. If users only connect successful accounts while hiding losing ones, individual records may be authentic—but overall reputation becomes distorted. A person skilled in trading isn’t necessarily skilled in technical analysis or security research. Ultimately, the ability to distinguish between who is famous and what someone has actually accomplished in specific domains is essential. If such domain-specific reputations genuinely improve corporate decision-making, InfoFi’s scope could expand significantly. Key metrics to watch include: the rate at which brands rehire creators, the number of new customers acquired through campaigns, the diversity of activities connected by Aura, and whether data selection bias decreases over time. Kaito’s next competitive edge will lie not in measuring how much attention people receive, but in proving that the histories behind that attention serve as credible foundations for real-world decisions. #Kaito #Aura #InfoFi

[Did AI recommend my brand? An AI that investigates this starts at $5] @termix_ai The TermiX AI Agent Marketplace features an unexpected service: Its name is GEO Specialist. It investigates how often a specific brand is mentioned by other AIs, where it is recommended, and which sources are cited. The starting price is $5. Imagine a team building security tools on BNB Chain. Even with a well-designed website, someone asking an AI for recommendations on where to audit smart contracts might only see other services appear. This is a problem that arises when customers narrow down options based on AI responses before manually reviewing search results. This field is called GEO (Generative Engine Optimization). Google itself explains that its AI summaries and AI mode responses may differ in structure and linking. Traditional SEO principles still apply, and no special markup is required for visibility in AI search. So, can we directly measure how often a brand appears in actual customer queries? Select ten customer questions that omit the brand name, then repeatedly submit the same questions to two AI services over three days to gather 60 responses. Record three key details: 1. Was the brand name mentioned? 2. Was it included as an actual recommendation? 3. Were the cited links appropriate as supporting evidence? The way questions are phrased can significantly affect results. If you ask, “What is our company?” and the brand name appears, it’s like giving the test-taker the answer. But if the brand name emerges when someone asks, “Recommend somewhere to solve this problem,” it more closely mirrors how a new customer discovers it. On TermiX’s Agent Marketplace, this kind of research is listed as a paid service. The structure of agent-to-agent transactions and on-chain settlements connects not only to smart contract development and data analysis—but also to investigating how brands are discovered within AI search environments. There’s a fundamental difference between someone who already knows the brand and searches for it directly, versus someone encountering it for the first time via an AI recommendation. When measuring brand visibility in AI search, it’s not enough to track how often a brand is mentioned—you must also record the specific questions through which it appears. #TermiX #GEO

[What Must Change Before XO Can Launch in Korea] @xomarket launched the Pulse PnL Panache (PPP) event on October 20, where participants trade whether BTC will rise or fall within five minutes, competing for rankings based on profit and loss (PnL). A total of 3,000 USDC and 30,000 CP will be distributed among the top 10 participants. To qualify for ranking, users must meet a minimum trade size of 5 USDC per trade and a cumulative trading volume of 100 USDC after registration. The event is scheduled to end on November 3. However, even if Korean users meet these participation conditions, their trading activities are not automatically legal. In August, access to Polymarket was blocked in Korea, and in September, 26 domestic users were investigated for alleged gambling, with 18 referred to prosecutors. Although no convictions have been finalized yet, it has become clear that permitting access to overseas services does not equate to legal authorization under Korean law. Before XO can enter the Korean market, it must first determine which types of contracts can legally be permitted—not merely whether access from Korea is technically allowed. PPP’s BTC 5-Minute Up/Down contract settles based on the price of a cryptocurrency asset. In contrast, contracts wagering on elections or sports outcomes differ fundamentally in the nature of their underlying events. A 2026 study on Korean commercial law also concluded that contracts based on political or sports events are difficult to classify as derivatives under the Capital Markets Act. BTC price-based contracts require separate legal review regarding their underlying asset, transaction structure, and settlement mechanism. Simply using a price as an indicator does not automatically make a contract a lawful financial instrument. Following this distinction, actual pathways for legalization must be established: It must be determined whether such products can be lawfully treated under current regulations, who is authorized to offer them under what licensing framework, and how they avoid conflict with laws prohibiting gambling or speculative activities—supported by official regulatory assessments and legal grounds. If existing regulations cannot accommodate such contracts, separate legislative discussions may be necessary. Being designated under a financial regulatory sandbox does not automatically exempt a product from criminal liability under gambling laws. Finally, product design must be considered: Until legal foundations are secured, restrictions should be applied individually to Korean users’ USDC trading, event participation, and cash-based rewards. Meanwhile, non-monetary features—such as market data access or analysis of settlement rules—could potentially be offered separately. XO’s terms of service already prohibit use by users in jurisdictions where activities are illegal under local law. Here lies a key variable determining XO’s future competitiveness: The ability to apply differentiated access permissions based on country, contract type, and user behavior may be as critical as the capacity to launch more prediction markets. Even if a legal pathway opens for BTC price-based products, it does not mean all political or sports prediction markets and reward-based contests will be permitted simultaneously. The real launch for Korean users will occur only when the scope of permissible contracts, the legal qualifications of operators, and the specific domestic conditions for trading, settlement, and rewards are publicly disclosed and actually implemented. #PredictionMarkets

If I receive my salary in USDC, can my employer see my entire balance? @Americanfort_io Let’s assume a freelancer receives USDC every month to the same wallet address. If the employer enters that address into a blockchain explorer, they can view all other deposits and the full balance of that address. The address you share to receive payments becomes a clue that enables tracking of your entire financial activity. AmericanFort’s AmericanFortress aims to solve this by separating the public identifier from the actual address used to receive funds. First, you register a @username called FortressName. Instead of sharing a complex wallet address, your counterparty simply enters that name—during supported payment flows, a new stealth address is generated and used to send assets. A stealth address is designed so that the public name cannot be easily linked to individual incoming transactions. This way, you can keep using the same name while minimizing repeated exposure of your actual receiving address. What about when payment proof is needed? The upcoming SafeSend feature is planned to allow users to selectively share only the necessary payment details—not their entire transaction history. For example, a freelancer could prove receipt of a specific invoice, or a company could provide auditors with only the transactions they need. While all transactions remain recorded on-chain, this approach reduces the risk of linking all personal financial activity to a single public address. The currently released macOS wallet is in beta. Names registered in beta are for testing only and cannot be used to send real assets; they will not migrate to the live environment. An SDK for integrating this functionality with external wallets and services is also in preparation. Imagine receiving money without ever sharing your wallet address directly—and only providing proof of specific transactions when needed. If AmericanFortress’s privacy-focused payments are successfully implemented, they could transform everyday financial activities like stablecoin salaries or freelancer settlements.

[Saturn invested in annual 15% preferred shares; calculated based on funds received by the company, this amounts to 17.2%] @saturn_credit The $SATS preferred shares issued by Strive on October 20 had a total offering amount of $630 million. However, calculating the annual accumulated dividends on 7 million shares yields $105 million—approximately 16.7% of the total offering amount. Saturn also participated in this IPO using its own treasury funds. The discrepancy begins with the $10 par value. While SATS carries a 15% annual dividend rate, the calculation is based on a stated value of $100 per share. The offering price was $90. Therefore, the annual dividend accumulated per share is $15, and the ratio relative to the offering price is: 15 / 90 * 100 = approximately 16.7% We can also calculate from the issuer’s perspective. Strive disclosed offering fees of $3 per share, totaling $21 million. Excluding this from the total offering amount, the net proceeds to the company before other issuance costs amount to $609 million. 105,000,000 / 609,000,000 * 100 = approximately 17.2% This ratio illustrates how substantial the annual accumulated dividend is relative to the capital actually received by the issuer. It does not reflect investment returns accounting for price fluctuations, reinvestment, or actual dividend payment timing. Strive also disclosed in its SEC filing its plan to set aside $22.50 per share—equivalent to 18 months of dividends—in a reserve account. For the full 7 million shares, this totals $157.5 million. This is a measure to secure initial funding for high fixed dividends. Priority structure applies: SATS is subordinate to SATA preferred shares but senior to ASST common shares. Dividends accumulate, but actual cash payments require board declaration and legally available funds. The first regular dividend payment date is scheduled for November 9. This transaction leaves different calculations for investors and issuers. Investors accept price volatility, payment risk, and subordination to higher-priority securities in exchange for high dividend yields. The issuer secures capital for Bitcoin purchases or working capital but assumes long-term dividend obligations. Saturn’s participation carries distinct significance: This investment is made from Saturn’s own treasury funds, separate from USDat or sUSDat user assets. It represents Saturn’s selection of Bitcoin-based credit products as a genuine capital allocation target, with no announcement of changes to existing user product collateral or payment terms. What remains to be monitored are: actual dividend distributions starting in November, the market price of SATS, and the management of reserve funds and issuer liquidity. Only when these three factors align can we assess how sustainable the high dividends of digital credit truly are. #Saturn

[Proven Ethereum with a single GPU—but the real-world test is different.] @zksync proved 7,200 consecutive Ethereum mainnet blocks using just a single GPU. 99% of these were proven within 10 seconds, and all blocks were proven within 12 seconds. Yet, according to EthProofs’ real-time proofing performance metrics—supported by the Ethereum Foundation—only 2 out of 8 evaluated provers met the recent weekly eligibility criteria. These two results cannot be directly compared, as they measure different targets and environments. Rather, the gap between them reveals the next challenge for ZK technology. Summary: - Airbender Prime demonstrated the feasibility of matching Ethereum’s ~12-second block production rate using just a single GPU. - In real-world prover operations, however, proof speed alone is not enough—uptime, proof failure rates, and hardware requirements must also be considered. - Long-term, zkEVM competitiveness will likely hinge less on raw proof throughput and more on accuracy verification, proof cost under worst-case conditions, and operational reliability. On October 13, ZKsync announced Airbender Prime: a system that proved 7,200 consecutive Ethereum mainnet blocks using a single NVIDIA RTX PRO 6000 Blackwell GPU. Since Ethereum produces blocks approximately every 12 seconds, 7,200 blocks correspond nominally to about 24 hours of block data. The significance lies in the fact that—when including time needed to prepare the witness data—99% of blocks were processed within 10 seconds. If proof generation is faster than block production, verifiers can confirm execution results cryptographically rather than re-executing every transaction—dramatically reducing computational overhead. However, performing a performance test on 7,200 blocks differs fundamentally from sustaining continuous proof generation in a live network environment. EthProofs publicly measures the performance and operational status of multiple systems generating real-time proofs for Ethereum blocks. As of the October 26 dashboard snapshot: - 8 provers under evaluation; only 2 met recent weekly eligibility criteria. - 69.7% of evaluated block slots had proofs generated within 10 seconds. - 26.2% of provers were classified as offline. These figures are not from Airbender Prime’s single-GPU test—they represent aggregated operational data across multiple provers, with varying hardware, execution environments, and eligibility standards. Yet they underscore a critical point: to be useful in a live network, proof systems must deliver not just high individual performance but also operational reliability. Even if a prover averages only 5 seconds per proof, frequent downtime or inability to process certain blocks makes any verification architecture relying solely on that system inherently unstable. It’s essential to verify whether alternative implementations can step in when one prover fails—and whether proof delays accumulate when difficult blocks arrive consecutively. Hardware requirements also require independent validation. The Airbender open-source documentation states that GPU resources are used for base and recursive proof stages, while the final proof stage requires CPU and approximately 150 GB of RAM. It remains unconfirmed whether this configuration matches the exact setup used in the latest Prime benchmark. Therefore, the result of “one GPU” cannot be taken as definitive evidence of total system memory requirements or operational cost. On October 15, @VitalikButerin praised the Airbender team’s achievement and outlined three next steps for zkEVM: First: Mathematically verify that the computations used in proving match Ethereum’s actual execution semantics. Second: Impose strict upper bounds on proof cost relative to block gas usage. Third: Verify the cryptographic security of the proof system itself. The second condition directly relates to real-world operational costs. What if someone deliberately constructs a block containing operations that are computationally trivial to execute but extremely expensive to prove—while still staying within the allowed gas limit? The block executes normally with standard fees paid, yet generating its proof requires disproportionately high computation and time. If such scenarios are possible, even an average proof time of a few seconds becomes meaningless: predicting the network’s maximum sustainable load becomes impossible. The economic viability of real-time proving depends not only on average proof cost but also on how much resource is needed under worst-case conditions. EthProofs’ roadmap is already moving in this direction.The current performance optimization phase is complete, and the security verification phase is underway. Next, the integration of proof into the delay proof, obligation proof, and protocol layers is scheduled. This sequence matters because creating a fast proof system is fundamentally different from making Ethereum trust that proof within its consensus process. The announcement of Airbender Prime provided strong evidence that hardware overhead for real-time proofs can be significantly reduced. The next milestone cannot be evaluated solely by faster metrics. What matters is how long high proof success rates can be sustained in real-world conditions, whether proof costs remain bounded even in worst-case blocks, and whether different implementations can accurately prove identical Ethereum execution results. ZK proof technology is transitioning from demonstrating fast computation to becoming infrastructure that networks can reliably depend on. #Ethereum #zkEVM