Next technical objective: increasing shardchain performance

LLM-based content moderation in Toncenter API Toncenter API now automatically filters inappropriate token images (NFTs and Jettons) using an LLM-based moderation system. Following the reduction in TON transaction fees, spam and inappropriate content distributed as NFTs have become more noticeable on the network. Building on its existing manual moderation tools, Toncenter has introduced an LLM-based image analyzer. Images flagged as inappropriate are now served by the API in blurred form. No changes are required for services already using Toncenter API. Guidelines for Services Toncenter API v3 metadata includes the following fields to help services handle potentially unsafe or malicious tokens: - is_nsfw — marks tokens that have been identified as unsafe or potentially unsafe; - is_scam — marks tokens associated with scams or spam; - extra._image_big, extra._image_medium, and extra._image_small — contain links to processed token images. Images are resized to standard presets and automatically blurred when a token is marked as unsafe. To better protect users, we recommend using an opt-in approach when displaying tokens in wallets, explorers, and other services. By default, show only well-known or trusted tokens and their images. Other tokens and their images should be displayed only after the user explicitly chooses to enable them. Please note: LLM-based filtering provides an additional layer of protection, but no automated filtering system can guarantee 100% accuracy. Both false negatives and false positives may occur.

No text

Pending API and Streaming API Performance Improvements We’ve shipped a major performance update to the Pending API, Streaming API and trace processing pipeline. Key improvements: — Moved action classification to C++ and integrated it directly into the Trace Emulator. — Moved external message TTL tracking into the Trace Emulator, removing a separate service from the pipeline. — Reduced Redis reads across trace processing and Streaming API. — Made confirmed → finalized trace promotion significantly cheaper by avoiding full trace reprocessing. With these changes, performance increased by a factor of 2.5. The system demonstrated stability during testing on a testnet at 1,500 TPS on the blockchain. No changes are required on the part of customers.

TON performance update With the August 17 update, TON mainnet validators transitioned to a new collator-based architecture. The update has been running stably on mainnet. The rollout of planned limit increases began on August 27. The first change raised leader_window from 4 to 16. Single-shardchain performance tests Two tests were conducted on mainnet to measure single-shardchain performance using a workload that simulated jetton transfers: — August 25, approximately 10:30 UTC: an input load of approximately 1,200 TPS. — September 4, approximately 11:30 UTC: an input load of approximately 1,500 TPS. In the latest test, a single shardchain processed 600–650 TPS in sub-second mode for this workload. As the input load increased further, the shardchain split into two. After split operation latency increased, but the block production rate remained stable. Next steps Work to increase single-shardchain performance will continue. The current target is 1,000–2,000 TPS per shardchain. As a reminder, single-shardchain throughput is not the throughput limit of the entire TON blockchain. TON can run many shardchains in parallel.

Updated staking smart contract lineup Single-nominator (v1.1) — Direct validation with one validator and one nominator. The nominator receives 100% of the rewards. Nominator-pool (v2) — Private staking pools with a small group of nominators and validators. Liquid-staking — User-facing staking products with many depositors. Issues jettons representing staked Gram for use in DeFi. Used by @StakeeBot and Tonstakers.