把 10 个 Jev demo 仓库的架构思路拼成 Solana 交易流水线 nerve

h100envy @h100envy原推发布于 2026.09.24

在 X 上打开原推

JEV OPENED THE API. EVERYONE RUSHED TO BUILD DEMOS. I TOOK THE 10 BEST REPOS AND WIRED THEM INTO A TRADING PIPELINE. repo: http://github.com/h100envy/nerve did not copy a single line of code. pulled one architectural idea from each repo and mapped it to nerve. agent-desktop reads the desktop accessibility tree and decides where to click. i took the pattern of reading structured state instead of screenshots. pool scanner reads pool state directly from chain, no ui parsing. typesafe-mario plays mario without screenshots. reads emulator ram, gets structure, decides. same principle: typed state in, typed decision out. jev in nerve gets pool json and returns verdict json. no text. jev-drone controls a drone. stabilization and safety below, jev makes decisions above. in nerve this is the reflex layer. deterministic checks (bundle, sniper, honeypot) fire before jev even sees the pool. safety does not wait for the model. OneVOneJev plays an fps. every tick: movement, aim, fire, jump. four decisions per call. in nerve jev answers four typed questions per request: liquidity, team, narrative, timing. one call, four scores. 190ms. jev-trader market makes on monad testnet. model latency 81ms. i benchmarked nerve on solana: 190ms per pool. not hft, but faster than any agent framework chaining prompts. prism does not place orders itself. identifies state (toxic flow, mean reversion, pressure) and hands off to the strategy. nerve works the same way. jev does not trade. jev scores. the decision to buy is made by the user or the executor node. neo4jev stuffs jev into a knowledge graph. at each node decides which edge to take. in nerve this is the narrative map. grok builds a narrative graph from ct, jev decides which narrative is accelerating, which is dead. jev-curate filters training data. checks quality, relevance, risk of each row before letting it into the next training round. in nerve the memecoin analyzer does the same thing with tokens. description, twitter, telegram, whois, creator wallet history. filter before entry. canny prevents coding agents from lying about being done. looks at outputs, diffs, tests and judges whether to believe them. in nerve this is the adversarial checker. gets everything the other nodes wrote and looks for reasons not to buy. told to be harsh. killmyidea scores startups from multiple angles and outputs KILL, FIX or SHIP. the scoring matrix in nerve does the same thing. collects scores from all nodes, weighs them, outputs pass or skip. no appeals. 10 repos. 10 patterns. zero lines copied. architectural ideas only, mapped to trading. the best way to learn to code on a new model is to break down what others already built and apply it to your own problem.

这条是引用着另一条写的：x.com/h100envy/status/2099876352849813580