
Build on @arc | @Polymarket trader | @Aster_DEX
America
Joined October 2021
- Tweets2.6K
- Following2.4K
- Followers237
- Likes2.4K
Zen retweeted
Winner announcement in 24 hours! We'll randomly draw 100 eligible winners with help from @grok. Make sure you follow us, like & repost to be in the running. Good luck! 🍀
Zen retweeted
PAPER is now trading 24/7 on Hyperliquid.
$PAPER is the native token of @papertrade_xyz, a synthetic perpetuals platform on HyperEVM, minted to traders from realized losses and stakeable for a share of platform fees. The contract trades as a Pre-Launch perp that converts to a normal perp.
Zen retweeted
Build long-running agents with more control over agent execution.
New capabilities in the Agents SDK:
• Run agents in controlled sandboxes
• Inspect and customize the open-source harness
• Control when memories are created and where they’re stored
Zen retweeted
Deposit Wallets launch tomorrow May 4, 12:30 UTC. Up to 30 min downtime to fix the ghost fills issues users have been experiencing
Things will get better and better throughout the day as the rollout spreads
⚠️ Devs: New API key orders now use ERC-1271, update payloads + clients
UI Users: nothing changes.
docs.polymarket.com/trading/…
Zen retweeted
让大家享受最后一年意义的工作,Chad Hurley 写的这句话是多么有诗意的一个说法,为什么在中文语境里就变成了贩卖焦虑和断章取义?语文到底是谁教的?我认为技术主义者本质上都是反人类主义者,他们不断 push 人类向下一个神性的世界前进,所以他们的话语中无处不存在着悲观,但行动中无处不显露着乐观,如果这一点都无法感同身受的话,就不要写代码了,也不配!
Zen retweeted
Aster 社区抽奖第二弹🚀💛
48小时极速冲浪!
很多没来香港的群友们都想要我们的周边,
搞个幸运数字快速抽奖宠一下群友~
规则超简单⬇️:
☑️关注@Ccweb3hub ;
☑️点赞并转发此条推特;
☑️评论区晒你的Aster邀请码(图片)
🔸数字尾号:1/3/7/8
🔸字母尾号:B/F/G/T
🎁奖励:
选8个符合的第一个评论的小伙伴送充电宝+充电器套盒一份!
兄弟姐妹们冲起来!
#Aster #ASTER #AsterDEX
(抽奖结果将在下周公布,获奖者私信发放)
Zen retweeted
今天正式发布了我的第 13 个 vibe 产品 tuwa.ai
这款产品比较特殊,它是一个电话服务。准确来说,tuwa 是一个 AI 电话网络,连接着世界上超过 100 种不同语言的人们和互联网上的 agent。
任何人都可以不下载 tuwa 而使用它,你只需要拨打免费的转接热线电话 +1 888 886 2968,告诉它你需要打给哪个号码,tuwa 就会帮助你拨打对应的号码,你说自己的母语,对方听到的却是TA的母语,反过来也是一样。兼容任何电话,对方不需要安装应用。固话、手机,世界上任何一个角落,都可以。
tuwa 支持 100 多种语言的实时翻译,你甚至可以在打电话时随便切换语言和对方对话。除此之外,tuwa 还支持语音克隆,每一通电话,都会让你的 AI 语音听起来更像你。
当然,我也为它设计了方便的 web app,如果你想,可以不通过转接电话而使用 web app 拨打,并设置自己喜欢的声音,使用外呼 agent 拨打电话,连接自己的 agent(例如 openclaw 或者 codex/claude code)并让他们自由的呼入与呼出。
外呼电话 agent 是我最喜欢用的 tuwa 功能,只需要交待清楚事情,比如完成餐厅预订,它就会在你希望的时间主动拨打对方的电话,说明来意,达到目的,并记录和翻译所有对话内容。
tuwa 的使用和收费都很简单,每月免费额度,固定套餐,按需付费。
这个产品的命名灵感来自于日语的「通話 tsuwa」最初,我只是想设计一个能帮我预订餐厅的电话服务,但后来,我在 vibe 的过程中慢慢意识到,世界上仍然有很多人无法体验 AI 带来的变化与便利,而电话,是连接他们最简单与自然的方式。我希望 tuwa 能帮助外语普及率低,偏远地区和第三世界国家的人们体会到这一点。
Zen retweeted
昨天转发了一篇用 Deribit 波动率曲面推导 Polymarket 二元期权定价的思路。一位做 PM 自动化交易的大哥@jtrevorchapman 告诉我,他几个月前试过类似的路径,结论是在 15 分钟尺度上,波动率曲面能给的信息,市场微结构信号可以更快更准地捕捉到。
我赶紧去看了大哥的实盘,发现他当前使用的交易系统表现非常优秀,盈利因子 4.29(每亏 $1,能赚回 $4.29),胜率 93%,而且他有问必答毫无保留。
热心哥的核心系统是一个三层架构:记忆 → 信号 → 防御
1. 记忆层
每个 session 开始,先扫描最近 30 +个已结束的 session,找出最像当前盘面的历史 session。比如当前 BTC 价格在某个位置、波动率是某个水平、市场情绪是某个状态,在过去类似的session中,最后结果是 UP 赢得多还是 DOWN 赢得多?如果是 UP 多,那么系统会带着这个“历史经验”(先验偏差)进入下一层。
2. 信号层
这层关注实时数据,每秒钟运行8-12条规则,然后独立投票UP / DOWN + 置信度(0-100%)。这些规则中热心哥认为 CVD(累计主动买卖量差)是预测力最强的单一指标。其他还有预言机报价距离、 Binance 动量、订单簿不平衡、UP / DOWN 代币价格走势(反映 PM 上所有参与者的集体判断)。
所有规则按置信度加权,得出方向和综合置信度,如“方向 YES,综合置信度约 65%”。
3. 防御层
方向确定后,用五个因子算出一个 0-1 系数,直接乘到仓位上,1 就是全仓执行,0 就不做。所以同样一个"65% 置信度 YES"的信号,根据防御层的评估,仓位大小也会有区别。这五个因子如下:
- CVD 同不同意?如果信号层说 UP,但 Binance 上的净卖压很大(CVD 不同意),仓位直接大幅压缩。CVD是热心哥极其看重的因子,在session混乱的情况下(预言机报价反复穿越基准价格 5 次),CVD 甚至有一票否决权(系数降为 0)。
- 距离基准价格多少?如果 BTC 现价和基准价咬得很紧,说明随时可能翻盘,需要压缩仓位。
- session 剩多少时间?如果只剩最后两三分钟,任何突发波动都来不及反应,风险陡增。
- 当前 session 预言机报价反复穿越了基准价格几次?超过 5 次就算混乱。混乱市场里信号可靠性大打折扣,仓位要激进压缩。
- 当前入场价的利润空间够不够?入场价越高,对胜率的要求也越高,所以系统在高价位时对信号可信度要求极高。比如 $0.95 入场时,只允许最无歧义最干净的信号通过。
除了核心系统之外,热心哥还提过几个额外细节:
- 只做 BTC,不做 ETH 市场,估计是 ETH 的信号不稳定
- 用 40 天、每 200 毫秒记录一次的数据做回测,发现在 session 剩余 6 分钟的时候指标才开始可靠。他用这个来衡量反转风险,过滤掉不该进的session。
限于时间没有全部整理完,他时间线上还有很多细节,推荐去翻翻。原文来自这篇回复:
teapawt.net/jtrevorchapman/status/…
Replying to @BryanBrygrant
three-layer system: Memory > Signals > Defense.
Layer 1 Session Memory (the prior)...
before ANY signals fire, the system already has an opinion. It scans the last 30+ completed sessions and asks: 'when past sessions LOOKED like this one, which side won?"
It computes a YES win rate from historical data. This becomes a directional "prior"... a starting bias before the current session even begins. If history says YES wins 65% of the time, the system enters the session already leaning YES.
Layer 2: Real-time Signal Voting
Every second, 8-12 signal rules evaluate current market conditions. Each rule independently votes YES or NO with a confidence score (0-100%).
Example:
"Chainlink oracle is $150 above the ptb -> YES, 70% confidence"
"Binance 60s momentum is strongly negative -> NO, 55% confidence"
'5m CVD shows heavy buying pressure -> YES, 80% conf"
These votes get aggregated: winning direction is determined by majority vote weighted by confidence. The final confidence score is the weighted average so if 8 rules say YES at 60-80% and 2 say NO at 40%, you get ~65% YES confidence.
Layer 3: defense dampening/sentinel
A 65% confidence signal DOESN'T mean a 65%-sized trade. The Sentinel risk system asks five questions:
Does Binance flow (CVD) agree with this trade? If not -> dampen hard
How far is the oracle from the ptb, price to beat? Tiny spread -> dampen
How much time is left? Early session = safer; late session with thin margin = risky
How choppy is this session? 5+ reversals = chaos -> dampen aggressively
What's the profit margin at this entry price? 0.95 entry = only 0.05 margin -> need near-perfect WR
These five factors produce a composite risk score (0.0 = safe, 1.0 = maximum risk) which gets converted into a dampener (1.0 = full size, 0.0 = skip entirely).
so... a trade that starts as "65% confident, $5 position" might get dampened to $1.50 because CVD disagrees and the session is choppy. Or it might get boosted to $7 because all 5 risk factors are green and it's a clean trending session.
The sIGNAL picks direction. The SENTINEL picks size. Defense generates the alpha, not offense.
Zen retweeted
预测市场终于有公开的全量 orderbook 数据了。
每小时更新,记录所有市场的每一次挂单变化和成交,30M 行/快照,Parquet 格式免费下载。Polymarket / Kalshi / Opinion 都覆盖。
成交记录只能告诉你「发生了什么」,挂单簿才能告诉你「为什么发生」。做策略回测的人都知道这个数据有多稀缺。
I just published a data dump of full order book data from @Polymarket
The data is maximally granular. There is no filtering whatsoever. Every order book change and trade is saved. Across all markets
Updates are hourly. Each snapshot contains ~30M rows. Snapshots are downloaded as parquet files. Each file is approx. 500MB-1GB large.
The data dump is already 2B+ rows large and growing fast. But this is just part 1/3. Coming soon is a much bigger dump that also includes @Kalshi / @opinionlabsxyz / @trylimitless etc
I started collecting this data because I noticed I couldn't get it from Dome API. Their historical order book data was filtered limiting its usefulness. Also now with the acquisition there's a lot of uncertainty about whether they will continue operating




























