On December 21st, APRO tweeted mentioning that their weekly data verification volume reached 89K, and AI Oracle calls also totaled 89K. These two numbers are completely consistent, which is quite interesting. It indicates that each data verification now generally involves AI processing, and it's no longer just a simple price feeding.
Let's look back two months ago, at the end of October and the beginning of November, when APRO had just launched on Binance. At that time, the verification volume data fluctuated between 77K and 107K. The statistical criteria might not have been very stable, but now it has stabilized at the level of 89K, indicating that the network has entered a relatively mature operational state.
What does 89K mean? On average, that translates to 12,714 verifications per day over a week of 7 days, equating to 530 verifications per hour and nearly 9 verifications per minute. Essentially, a data verification occurs every 6-7 seconds. For an oracle project that has only been online for just over a year, this level of activity is already quite significant.
But the more critical question is, what are these 89K verifications doing? Traditional oracles like Chainlink mostly verify prices, such as how much BTC costs and how much ETH costs—simple and direct. However, within the verification volume of $AT , a significant portion is handling complex unstructured data.
From a technical architecture perspective, #APRO is currently running a three-layer verification system. The bottom layer is the data collection layer, which collects raw data from over 40 blockchains and more than 1400 data sources. This layer is fully automated, with crawlers and API interfaces working 24 hours a day.
The second layer is the AI verification layer, which is APRO's most core competitive advantage. It deploys multiple independent LLM nodes, each running its own AI model to understand, analyze, and cross-verify the collected data. This layer is not just simple numerical calculation but involves semantic understanding and logical reasoning.
The third layer is the consensus layer. After multiple AI nodes provide their judgments, the PBFT Byzantine Fault Tolerance consensus protocol is used to compile the statistics. If the majority of nodes reach a consistent conclusion, that result will be accepted. If there are significant disagreements, it triggers manual review or stricter verification processes.
The design of this three-layer architecture addresses a fundamental flaw of traditional oracles, which can only handle structured data. @APRO-Oracle introduces AI to understand PDF documents, parse news reports, analyze social media discussions, and even comprehend images and videos. This capability is critical for the tokenization of RWA (real-world assets).
For example, a real estate tokenization project needs to verify whether the ownership of a certain building is clear. The traditional approach is to manually review property certificates and appraisal reports, which are PDF documents that are inefficient and costly. APRO's AI can use OCR to recognize document content, use NLP to understand legal terms, and then provide a credibility score.
Furthermore, if the property involves disputes or mortgages, this information may be scattered across court announcements, news reports, and government websites. AI can proactively search for relevant information, establish connections, and then comprehensively assess whether this property is suitable for tokenization.
Such complex data processing may involve calling dozens or hundreds of data sources and running several rounds of AI reasoning before arriving at a conclusion. Therefore, although it superficially appears to be just a single verification, the computational load behind it is enormous.
#APRO's deployment of this system on BNB Greenfield is a very thoughtful choice. Greenfield is a decentralized storage network launched by Binance, which has native interoperability with traditional IPFS or Arweave. Data can flow efficiently between the storage layer and the computation layer.
Specifically, when processing data, APRO's AI nodes will store all original data, reasoning processes, intermediate results, and final conclusions on Greenfield. They then use cryptographic signatures to ensure that this data has not been tampered with. Anyone can trace back and verify what data the AI made its judgments based on.
This fully traceable design is a lifesaver for prediction markets, as they fear unfair judgments the most. If the platform states that an event occurred while users claim it did not, both parties will hold their ground. The traditional resolution method is arbitration, but arbitrators can also be biased.
@APRO-Oracle's multi-node LLM consensus mechanism makes the judgment process completely transparent. Each AI node independently analyzes events, cites evidence, and provides a reasoning chain. All this information is made public on Greenfield, allowing users to verify whether the AI's logic is reasonable.
Moreover, due to the multi-node consensus, the biases of individual AI models will be diluted. For instance, if there are 10 nodes, 8 of which believe an event occurred and 2 that do not, the judgment will follow the majority opinion. However, the reasoning processes of those 2 minority nodes will also be preserved for community review. If issues are found with the majority, disputes can be initiated. $AT Another innovation introduced in this system is heterogeneous model consensus, where not all nodes run the same AI model, but simultaneously deploy models from different vendors such as GPT, Claude, and DeepSeek, each having its own training data and inference methods.
If different models can reach a consensus, the credibility of the result will be very high because the biases and blind spots of these models are different. A judgment that all of them can agree on is likely to be correct.
The difficulty of this heterogeneous consensus lies in how to align the outputs of different models, as the answer formats provided by each model may vary and their expressions differ. #APRO uses a standardized output protocol to force all models to provide judgments and confidence in a unified format.
From the distribution of 89K verification instances, it should be concentrated in several core scenarios. Firstly, RWA data verification: Lista DAO currently protects $600 million of real-world assets, which need to be updated daily for valuation, reserve verification, and risk monitoring. Each operation must call the oracle.
Secondly, there is the data demand of AI Agents. @APRO-Oracle has integrated with more than 25 AI frameworks, including mainstream platforms such as DeepSeek and ElizaOS. These frameworks run hundreds or thousands of AI trading bots, each of which may query market data and make decisions every minute.
There are also event judgments in the prediction market. Although prediction markets are still niche, each event's judgment may involve hundreds of data verifications, as evidence must be collected from multiple sources to cross-verify its authenticity. Thus, a single case can generate a large volume of calls.
The ATTPs protocol also accounts for a portion of this 89K verification volume. This protocol is a data transmission standard specifically designed for AI Agents. It uses zero-knowledge proofs and Merkle trees to ensure that data is not tampered with during transmission and supports selective disclosure. AI can prove that it knows certain information without revealing specific content.
This privacy-preserving data verification is particularly useful in cross-protocol risk assessment scenarios. For instance, a lending protocol may want to know a user's borrowing record on other platforms without exposing the query behavior. It can return encrypted results through ATTPs, with the nodes not knowing the query content, and only the querying party can decrypt it.
From a technical implementation perspective, APRO's current weekly verification volume of 89K is already approaching the upper limit of the comfort zone of the current architecture. If it wants to continue expanding, it may need to increase the number of nodes or optimize the consensus mechanism.
The good news is that Binance, as an investor and listing platform, will provide APRO with substantial infrastructure support. The block time of BNB Chain is only 3 seconds, and TPS can support thousands of verifications per second. As long as APRO's node network can keep up, technical bottlenecks will not become a limitation for growth.
Moreover, #APRO employs a hybrid architecture that performs complex calculations in an off-chain Trusted Execution Environment (TEE). On-chain, it only does the final consensus and settlement. This design significantly reduces gas costs, ensuring that even if the verification volume grows tenfold, the costs remain controllable.
From an ecological perspective, this 89K figure will continue to grow, as @APRO-Oracle supports more and more chains, increasing from 27 chains in November to over 40 now. Each additional supported chain brings new verification demands from the DApps on that chain.
Moreover, AI applications have significantly accelerated in the second half of 2025, with various autonomous trading bots, AI research assistants, and intelligent market makers operating on-chain. Their demand for data is rigid. If they can seize this wave of AI application explosion, the verification volume may double within a few months.
Overall, the number 89K in weekly verification volume is supported by a complex three-layer verification architecture, from data collection to AI analysis and decentralized consensus. Each layer contains technical content. APRO is redefining what verifiable AI data is through BNB Greenfield's storage, multi-node LLM intelligence, and ATTPs' privacy protection. If this direction can be successfully executed, the application boundaries of oracles will be significantly broadened.
