大型語言模型交易代理在生產環境中實際做了什麼：一份來自兩個機群、為期六個月的群體規模紀錄

摘要

我們提出一份連續、群體規模的測量紀錄，記錄自主語言模型交易代理在生產環境中的運作，涵蓋兩個具有同一設計血統的系統：DX Terminal Pro（3,505 個使用者出資金庫，於 2026 年 2 月至 3 月的 21 天期間在 Base 迷因幣市場交易真實 ETH）以及 DXAP 實盤 alpha 艦隊（全歷史 500 至 599 個使用者建立的代理，91 至 117 個同時活躍，交易 Hyperliquid 永續合約，2026 年 6 月至 8 月）。此紀錄橫跨約六個月、750 萬次單一模型呼叫，其中約 30 萬次鏈上操作，另有 231,638 個多工具回合，產生 14,596 筆成交。本文的核心為四項發現。第一，操作層對行為的決定程度高於策略文本中所寫的任何內容：風險滑桿可解釋槓桿（每級 +0.425），代理固定效應吸收 60% 的變異，而排行榜渲染邊界會因果性地引導選擇（前三名切點處的迴歸不連續為 1.75 倍）。第二，部位規模設定對波動率視而不見：在每一個波動率六分位中，槓桿中位數都是 5.0 倍，而單一姿態滑桿格（占部位簿 11%）就承受了 62% 的清算。第三，代理幾乎未捕捉到它們曾觸及的上漲空間：43.2% 的部位在 24 小時內曾出現至少 +300 個基點的有利偏移，然而其中 49.3% 以負交易報酬平倉；機械式括號單可為每個部位挽回 +39.0 個基點。第四，兩支艦隊皆未展現方向性優勢。DXAP 艦隊並不獲利，且落後於配對的 Hyperliquid 散戶基準（來回交易勝率 41% 對 50%）。在 416 個擷取的生產情境上，前沿模型的配對重播競賽發現，在此時間範圍內決策品質在統計上無法區分，而選擇穩定性在不同模型家族之間差異顯著。每一項主要結果都經得起按日分群的推論、置換虛無檢定，以及共同費用重述；本文最後以一套 17 條規則的方法學正典作結，那是以我們自身的撤稿換來的。

English

We present a continuous, population-scale measurement record of autonomous language-model trading agents operating in production across two systems with one design lineage: DX Terminal Pro (3,505 user-funded vaults trading real ETH in Base memecoin markets for 21 days, February to March 2026) and the DXAP live alpha fleet (500 to 599 user-created agents all-history, 91 to 117 concurrently active, trading Hyperliquid perpetuals, June to August 2026). The record spans roughly six months, 7.5M single-model invocations with about 300K onchain actions, and a further 231,638 multi-tool turns producing 14,596 fills. Four findings carry the paper. First, the operating layer determines behavior more than anything written in strategy text: a risk slider explains leverage (+0.425 per level), agent fixed effects absorb 60% of variance, and a leaderboard render boundary causally routes selection (regression discontinuity 1.75x at the top-3 cut). Second, sizing is volatility-blind: median leverage is 5.0x in every volatility sextile, and one posture-slider cell (11% of the book) holds 62% of liquidations. Third, agents capture almost none of the upside they reach: 43.2% of positions saw at least +300 bps of favorable excursion within 24h, yet 49.3% of those closed with a negative trade return; a mechanical bracket recovers +39.0 bps per position. Fourth, neither fleet shows a directional edge. The DXAP fleet is not profitable and trails a matched Hyperliquid retail benchmark (41% vs. 50% roundtrip win rate). A paired-replay league of frontier models on 416 captured production scenarios finds decision quality statistically indistinguishable at this horizon, while choice stability differs sharply across model families. Every headline survives day-clustered inference, permutation nulls, and a common-fee restatement; the paper closes with a 17-rule methodology canon bought with our own retractions.