alexrue

不知名技校本水硕在读|因为太菜只能尝试伪科学撸毛养活自己

[alexrue @alexrue27]

(En version here)
"The Ambiguous Rules of Polymarkets, Irrational Pricing, and Inspiration from UMA Attacks"

I. Starting with the "Alien Files": A Living Lesson in Progress

About a week ago, Polymarket launched several contracts around themes such as "Trump declassifies UFO files" and "Has the US confirmed the existence of extraterrestrials?" The two most typical ones were:
Trump declassifies UFO files in 2025?
Trump declassifies UFO files before 2027?
The rules for both are almost identical, except the deadlines are December 31, 2025 and December 31, 2026, respectively.

Fueled by the UFO topic, Trump's aura, and the narrative of "aliens finally making an official announcement," some markets were driven up to nearly 99% implied probability, while the 2027 market price was significantly discounted—logically creating a strange phenomenon where "a longer window actually results in a lower probability."

Even more outrageous is the fact that the 2025 market was repeatedly submitted as "YES" based on a so-called "declassified document" from September 2025; yet, when discussing the 2027 market, Polymarket/UMA attempted to use a vague English phrase in the rules to "exclude" the same document from the longer-term market.

This controversy has severely damaged Polymarket's image as a "truth market" and exposed three structural problems:
1) The rules are vaguely worded, allowing key clauses to be arbitrarily interpreted afterward;
2) Pricing contradicts basic probabilistic logic, revealing serious behavioral biases and room for manipulation;
3) The settlement mechanism, reliant on UMA governance, leaves room for governance attacks that could "buy the truth."

II. Mismatch Between Headlines and the Essence of the Event:

The Misleading Narrative of "Trump Declassifies"

Let's look at the headlines of these two markets:

Trump declassifies UFO files in 2025?

Trump declassifies UFO files before 2027?

For an average participant, such headlines easily induce the following intuitive understanding:

"If Trump, during his term, personally authorized/approved the declassification of a batch of documents related to aliens/UFOs, then it's YES."

In other words, the headlines paint a psychological picture:

This is an event strongly linked to Trump's personal actions;

This is a dramatic event with a strong "whistleblower" feel—"Trump has finally unleashed his secret, revealing aliens."

However, looking further into the rules, it's a completely different story:

This market will resolve to "Yes" if the Trump administration declassifies any previously classified files pertaining to extraterrestrial life and/or unexplained arial phenomena by...

Several questions immediately arise:

What exactly does "Trump administration" refer to?

Does it refer to the declassification decisions made directly by Trump and his cabinet?

Or does it refer to all documents "declassified by any federal agency through normal procedures during the term of this administration"?

If it's the latter, then theoretically, any release of UAP-related materials by a department (such as AARO) during normal workflow could be considered "Trump administration declassifies"—even if Trump himself was completely unaware of it.

The personified verb "declassifies" in the title further reinforces the misleading nature.

The title isn't "UFO files are declassified under Trump,"

but rather "Trump declassifies…."

In English, this personifies a highly bureaucratic process as "Trump personally lifting the lid." This approach panders to the emotions of the media and individual investors, but deliberately downplays the institutional context of the event and the normality of the declassification process.

The actual "declassification action" is most likely just routine updates by agencies like AARO.

Based on the operational methods of the AARO website, it's clear that the so-called "UAP-related documents" are likely just routine "unexplained phenomena investigation + regular archiving," with the vast majority categorized as balloons, birds, and drones, and the remainder listed as "unexplained." This is entirely different from the public's perception of "official recognition by an extraterrestrial civilization."

However, in Polymarket's headline narrative, these are packaged as "Trump declassifies UFO files"—not only attributing the event to Trump personally but also implying that he has a shocking secret to reveal.

Therefore, from a communication perspective, this headline design has a strong tendency towards narrative amplification and misleading:

It packages what is essentially a mundane bureaucratic process as a dramatic presidential decision;

It simplifies what is essentially "declassification of any UFO/UAP-related documents" into "Trump personally declassifying alien files";

It simplifies the "official information + media consensus" principle into a sensational question mark sentence, misleading uninformed retail investors into believing they are betting on a "historic moment."

This laid fertile ground for subsequent pricing distortions and settlement disputes.

III. Double Standards Regarding Timeframes: Is the same September 2025 document only "valid" for the 2025 market?

Let's look at the core of the controversial rule:
Announcements of declassifications that are not implemented within this market's timeframe will not count.
This was originally intended to exclude a scenario where an official prematurely informs the media: "We intend to declassify certain documents";

but the actual declassification (classification change, public release) is delayed beyond the timeframe.

Therefore, the rule emphasizes:

"Announcements that are merely 'announced' but not actually declassified within the timeframe are not counted."

However, in the UFO market controversy, Polymarket/UMA's interpretation of this statement clearly demonstrates a double standard regarding timeframes.

1. Timeline Recap

Real-world event:

September 2025: A so-called "declassified document" related to UFOs/UAP appears, which is used as key evidence by the YES side.

Two Markets:

2025 Market
Title: Trump declassifies UFO files in 2025?
Creation Date: April 18, 2025
Deadline: December 31, 2025

2027 Market
Title: Trump declassifies UFO files before 2027?
Creation Date: November 5, 2025
Deadline: December 31, 2026

In practice, the logic was handled as follows:

For the 2025 Market:
The September event occurred after the April creation and before the December deadline ⇒ It was naturally included in the timeframe, becoming the main basis for a YES result.

Regarding the 2027 market: The same September document was interpreted as:

"Events occurring before this market was created (September < November 5th) are not included in this market's timeframe and are not counted."

In other words, the timeframe was implicitly defined as the period "from the market's creation date to the deadline."

This was never explicitly stated in the rule text, yet it was used afterward to exclude the impact of the same event on longer-term markets.

2. Why is this a serious rule rewrite?

(1) Contradicts the intuitive meaning of the title

"in 2025" is naturally understood as "the entire year of 2025";

"before 2027" is naturally understood as "anytime before 2027".

No normal reader would intuitively understand this as:

"only declassifications that happen after this market is created and before the deadline."

(2) Rewriting the prediction market into a monster that "only predicts future sub-periods"

Following this logic,

even if a genuine UFO file declassification occurred in January 2025,

if the market only went live after April 2025,

this actual declassification would be treated as "never having existed."

This deviates from the fundamental mission of prediction markets—to aggregate information about the true state of the world.

Markets should not "forget" events that have already occurred, nor should they forcibly erase facts by using creation time.

(3) Most crucial: The same real-world event is treated completely differently in two markets

In the 2025 market: this September document is the core evidence for YES;

In the 2027 market: the same document is excluded by the "time interval" clause.

In other words, within the same world, two contradictory "versions of fact" exist on the blockchain: In one contract, the document is treated as "the Trump administration declassified UFO-related documents"; in another contract with a longer timeframe and identical rules, the system pretends the document never happened.

This is a classic example of: exploiting ambiguity in the rules to rewrite the timeframe after the settlement phase, serving a preference for rulings on a specific market outcome.

If Polymarket/UMA genuinely believes that:

"the timeframe must start from the market creation date,"

this condition must be clearly stated at market creation:

"from the market's creation until December 31, 2025/2026,"

instead of "picking out" a vague term from the rules after a dispute arises,

to explain why the same September document is only counted in the 2025 market and not in the 2027 market.

IV. Irrational Pricing:

When prices are no longer bets on "events" but on "arbitration,"

based on the two problems mentioned above (misleading titles + double standards in timeframes), market prices will naturally become severely distorted.

However, in reality, a structural inversion has emerged: "2025 = 99%, 2027 ≈ 95%."

The reason is that most participants are actually betting not on "whether the real world will be declassified," but rather on:
how UMA/Polymarket will ultimately interpret the rules and settle the 2025 dispute.

Regarding the 2025 market, many believe that "under immense public pressure and the interplay of existing positions, UMA is more likely to force a 'YES' interpretation for 2025."

However, regarding the 2027 market, expectations are different:

There's still plenty of time;
The controversies have already surfaced;

UMA is unlikely to repeat such a radical interpretation in the future;

Therefore, the 2027 market will not "automatically follow 2025 and become 'YES'."

Thus, the market with higher prices is actually the one that people are more likely to believe will be "determined to become 'YES'."

This is not information aggregation, but speculation on the "governance outcome." The closer the price is to 100%, the less certain the event is, and more certain it is that:

"Everyone is more convinced that this will be arbitrated as YES—even if this YES contradicts the original rules and common sense."

V. UMA Oracles and Governance Attacks of "Buying the Truth"

All these markets on Polymarket ultimately settle through the UMA Optimistic Oracle:

After the market closes, anyone can submit an "answer" and deposit a security deposit;

If no one challenges it within a short time, the answer automatically takes effect, and the market settles accordingly;

If someone challenges it, it enters the UMA governance layer, where UMA token holders vote to decide "which answer is correct."

Theoretically, this mechanism relies on an optimistic assumption:

"As long as a majority of voters are willing to uphold the truth,

then attackers would need to pay an extremely high price and therefore would not easily manipulate the results."

However, there have been publicly documented cases of **"UMA governance attacks"**:
Attackers hold a large number of UMA voting rights;

After a high-value market event concludes, they use concentrated voting to change events that should have been YES to NO (or vice versa);

Ultimately, they tamper with the on-chain "truth" at extremely low cost and profit from applications like Polymarket.

In this structure, highly emotional markets like UFOs/aliens, with extreme information asymmetry and a jumbled set of rules, are almost ideal targets for governance attacks:

The title provides attackers with narrative ammunition for emotional mobilization—"Trump is going to reveal aliens";

The ambiguous rules provide attackers with ample room for interpretation—"Is this considered a decryption? Is it an announcement or implementation?";

The double standard of time intervals is a convenient tool for rewriting the results afterward—"This deal can only be considered 2025, not 2027."

Therefore, in these scenarios, the so-called "prediction market" is more like:

"Whoever controls the discourse and governance has the ability to write their preferred version of history on the blockchain."

VI. Conclusion: When the "Truth Market" Meets "Narrative Economics"

Polymarket imagines itself as a truth machine that "makes prices tell the truth";

However, in the extreme narrative race of UFO documents and aliens, we see a different picture:

Headlines are deliberately personified and dramatized, reinforcing the emotional link between Trump personally and alien conspiracy theories;
Key rules are vague, especially the timeframe, which is later interpreted as an implicit lower limit "from the creation date," creating a double standard in time intervals;
Market prices are more about betting on "how the UMA will settle this dispute," rather than betting on "whether declassification has occurred in the real world."

@Polymarket
#predictions #UMA

[Polymarket @Polymarket]

BREAKING: Trump projected to declassify the UFO Files this year.

[alexrue @alexrue27]

@Justinhk1208 哈哈哈好！我就偷懒直接ssh连aws的日本vps了，但是这种如果不做加密处理是不是容易被识别封ip

[alexrue @alexrue27]

老哥说的大实话，中转的成本什么的我也门清，虚拟卡挂了一堆，cc基本都是反代出来的，aws都被大厂包圆了，要么是走低价渠道或者接入模型什么的，然后还有利润来源是售卖客户信息供国内厂商蒸馏，截取信息等等

[Kai @xkajon]

这摊浑水我搅定了

为什么发这篇东西？

我就是要推动更多人直接下场做中转站老板，别特么去当什么代理。自己搭站、自己控渠道、自己定价，哪怕起步再惨，那也是你自己的生意。这算是我对这个行业做的一丁点贡献，也算给国内中转站出海收割老外的路子上添把柴。

说实话，我看着国内一些很有潜力的代理被这么收割，心里真不舒服。哪怕你现在觉得自己不厉害，那又怎样？谁不是一步步爬过来的。

我刚开始碰中转站那会儿，惨得一批。一天流水一二百块钱，刨去成本剩不下几个子儿。为什么惨？渠道成本高得离谱，信息差大得吓人，两眼一抹黑全靠自己瞎撞。后来怎么起来的？路子跑通了，渠道商一个一个主动找上门来，信息差磨平了，利润率直接从脚底板窜到百分之八十。渠道源头一旦卡准，钱就跟捡的一样。

所以别觉得这事儿离你远。很可能你前脚把站搭起来，后脚就有渠道商摸过来给你供货——我手里一大半渠道都是这么主动找过来的。

但前提是，你得先坐上那张桌子，而不是跪在桌子底下等人扔骨头。

天策那套玩法是什么？拉一帮流量矩阵号，铺天盖地招代理，一条链接百来个浏览，成交一单不到五十块，层层分下去代理到手二十五。这叫生意？这叫披着AI外衣的传销过家家。一帮兄弟跟着干一两天，挣个四五十块，还得替他扛客服骂、背中转站炸锅的锅。他敢拍胸脯说自己的中转跟官方一模一样？鬼扯。那些暗病、那些小毛病，开过站的人心里门儿清。

我从最早碰中转站到现在，经手流水七十多万。这里头的猫腻、封号节奏、利润卡点，我比他们清楚一万倍。

之前随手写个搭建教程，那算客气的。这次是看到那几个矩阵号蹦跶——仔细一琢磨，背后搞不好全他妈是同一个人，天策，搁那儿自导自演呢。

行，既然你这么爱演，那我就不客气了。

什么VC、美元VC分析师、05年美元VC分析师、混币圈的，搞中转站，身份叠得跟千层饼似的。我就问一句：你手上到底有多少技术储备？

TLS指纹识别，特征流量识别，你搞得定吗？Claude那边每天都在迭代更新的特征识别系统，你拿什么扛？你要是真敢拍桌子说自己全绕过去了，那他妈可太精彩了，我第一个给你鼓掌。

但你要是绕不过去，就别拿那套花里胡哨的身份牌糊弄底下代理。

接下来两天，我直接把所有源头写出来。Claude的全链路、ChatGPT的供货端、Gemini的拿号口子，闲鱼上二十块一个月GPT Plus的底细，五六百Claude Max 20x的真实成本——拿货二百出头我转手三百五——这些你们听都没听过的号池源头，我一锅端。

天策，还有你底下那堆号，关注列表里还没我吧？现在点上。

等我忙完这两天边写边发，学着点。

[alexrue @alexrue27]

@diamondhandjs 归去，也无风雨也无晴🥺

[alexrue @alexrue27]

@0xAA_Science 🥹好高级

[alexrue @alexrue27]

@Michael_Liu93 还有能承受极大杠杆和波动的心理素质😭

[alexrue @alexrue27]

@LeiIsRealGoat 果然智商250和50能挣钱

[alexrue @alexrue27]

@leifuchen 没有没有我可菜了，还得老师多带带🧐，刚空下来准备好好看看

[alexrue @alexrue27]

Merton跳跃可能需要考虑一下是用garch还是泊松，以及潜在的解释理论依据等等，当然也可以使用别的跳跃模型

[leifu _/ @leifuchen]

大家可能都听说过 BS 模型不适用于加密期权定价，但到底有多不适合可能缺乏定量的认识。Kończal 在 2025 年发表的论文《基于加密期货合约的期权定价》使用 CME 的 BTC/ETH 期货期权数据对比了 6 个定价模型，BS 模型的误差是最优模型的 3.5-5.5 倍。

论文的核心发现:
- 对加密期权而言，能处理跳跃的模型碾压不能处理跳跃的模型。价格突然跳变才是加密市场的核心特征，因此捕捉价格的突然跳变比精确建模波动率的连续变化更重要。

- BS 模型的误差远超其他模型，几乎不能用于实际定价(尤其是远期期权)，原因是加密期权的隐含波动率大约是 S&P 500 的 4–6 倍，而且收益率分布有厚尾和偏度，完全偏离于 BS 的正态假设。

模型选择建议:
- 跨币种选 Merton 跳跃扩散模型（4 参数，两个币种都排前列）
- 分币种优化：BTC 用 Kou，ETH 用 Bates（MAPE 仅 1.9%,全场最优）

论文用三个指标衡量模型定价 vs 市场价的差异:
- MAE（平均绝对误差）最符合直觉,每个期权的定价偏差取绝对值，求平均。BTC 上 Kou 的 MAE 是 258,意思是平均每个期权偏了 $258。

- RMSE（均方根误差）先平方再开方，所以大的偏差会被放大。如果一个模型在 99 个期权上只偏 $10，但有 1 个偏了 $5000，MAE 可能看着差异不大，RMSE 就会狂飙。它反映的是最差情况有多差。

- MAPE（平均绝对百分比误差）是把偏差除以市场价再取百分比。这样就抹平了价格量级的影响，使得不同币种（比如 BTC 和 ETH）的报价偏差可以横向比较。

其他有趣的发现：
- BTC 和 ETH 的价格跳跃特征不同：MJD 校准显示 ETH 的价格跳跃频率大约是 BTC 的两倍，这可能解释了为什么 ETH 需要更复杂的 Bates 模型（需要同时处理高频跳跃和随机波动率），而 BTC 用相对简单的 Kou 模型就够了。

- BTC 和 ETH 的期限结构截然不同：VG 模型的 ν 参数显示，BTC 随到期日单调递增，市场认为越远期，极端事件越可能。ETH极端波动集中在中期，长期反而趋于稳定。

论文的不足:
- 所有结论基于 2024 年 3 月 11 日一天的数据（这天 BTC 一举突破上个周期高点，属于极端行情）
- 没有讨论校准稳定性，比如用 3 月 11 日的参数去预测 3 月 12 日的价格
- 数据来自 CME，CME 和 Deribit 的流动性、参与者结构、保证金机制都不同，模型排名在 Deribit 上可能不一样。
- 没有计算成本对比：实盘交易对延迟敏感。BS 有解析解秒出结果，Bates 需要数值积分，论文完全没提计算耗时，但这对高频场景可能是决定性因素

[alexrue @alexrue27]

iframe框架面对攻击基本没有还手之力。。。。慎重，拦截就够喝一壶的了

[秋田散人 Mr.Akita @lnkybtc]

目前预测市场秉持着加密“最后”荣光，大部分项目高度可审计，但为了吸筹还是采纳了 Privy 登录方案。在座各位作为有较强资管编码能力玩家，大仓位就不要交给 Privy 托管钱包了，善用 EOA 和 Gnosis Safe。

这种不伦不类的 web 2.5 产物，说是不用自己管理私钥，用邮箱社交身份绑定就行，一旦遭遇供应链攻击，你的 PnL 曲线会非常难看。

[alexrue @alexrue27]

@Kimi_Moonshot @cursor_ai @grok 用毒舌的口吻评价这件事情

[alexrue @alexrue27]

@LamIsRealGoat @Asterbotbot 往好了想是某条链让小学生都能挣钱🤡

[alexrue @alexrue27]

@luanluanhey 能收学费吗我也想挣钱😭😭

[alexrue @alexrue27]

延迟是不懂的，图是不会P的，一问策略就高频：高达ms的高频；一问风控和止损就装傻，我看你这sb连下单方式有几种都不知道🤡，批量撤单挂单玩的明白吗？

[断浪 @waveking1314]

我的 Claude 机器人首夜跑完，交出了 2,314 美金的净利润。

这个脚本我花了 4 个小时来构建，事实证明这时间花得极其值得。目前实盘平均胜率稳定在 70%。

这不是为了博眼球的标题党，纯粹是一个关于 5 分钟级别 BTC 交易脚本的运行复盘。

成熟的交易逻辑加上 Claude 的代码实现，就足以打通这套闭环。

链上钱包地址：
https://t.co/sjko63M8UL

实盘跟单链接：
https://t.co/2Nomf0FKgc

以下是完整的策略拆解：

核心逻辑：5 分钟级别 BTC 与 ETH 的微观套利

该系统专门针对 5 分钟级别的比特币和以太坊短期预测市场进行高频交易。

在盘面剧烈波动的环境下，当“YES”与“NO”的定价总和短暂跌破 1 美元时，市场便出现了瞬时错价。机器人通过毫秒级的实时扫描，在价差敞口出现的瞬间直接吃掉利润。

用物理速度降维打击主观情绪。

当波动率飙升、主观交易员还在评估风险而犹豫不决时，系统已经完成了条件触发。订单瞬间执行，没有任何人为延迟或情绪偏见。等大多数盘面玩家看懂了发生什么，这个微小的套利窗口早就彻底关闭了。

自动化执行让概率优势得以复利。

这套策略的初衷从不是为了单笔暴富，而是通过绝对理性的机器执行，反复收割市场上的微小价差。高频的运行机制允许机器人全天候运作，随着时间推移，这些微弱的单笔利润最终堆叠成了极其可观的绝对收益。

这套框架跑通后，接下来要做的只剩下优化代码细节、扩大资金规模部署，然后让它安静地运转。

[alexrue @alexrue27]

Attention is all your need？
Residual?

[Kimi.ai @Kimi_Moonshot]

Introducing 𝑨𝒕𝒕𝒆𝒏𝒕𝒊𝒐𝒏 𝑹𝒆𝒔𝒊𝒅𝒖𝒂𝒍𝒔: Rethinking depth-wise aggregation.

Residual connections have long relied on fixed, uniform accumulation. Inspired by the duality of time and depth, we introduce Attention Residuals, replacing standard depth-wise recurrence with learned, input-dependent attention over preceding layers.

🔹 Enables networks to selectively retrieve past representations, naturally mitigating dilution and hidden-state growth.
🔹 Introduces Block AttnRes, partitioning layers into compressed blocks to make cross-layer attention practical at scale.
🔹 Serves as an efficient drop-in replacement, demonstrating a 1.25x compute advantage with negligible (<2%) inference latency overhead.
🔹 Validated on the Kimi Linear architecture (48B total, 3B activated parameters), delivering consistent downstream performance gains.

🔗Full report:
https://t.co/u3EHICG05h

[alexrue @alexrue27]

@xingpt 感觉是钓鱼然后全杀了

[alexrue @alexrue27]

@jakevin7 还得是大佬啊，这反风控水平🥹

[alexrue @alexrue27]

@nbblock 被迫的吧，毕竟他的预测市场是数一数二的流动性啊

[alexrue @alexrue27]

@xingpt codex pro记得一起买🤓