Polymarket is misleading its traders. Again.
Some top accounts appear to get 50–65% Taker Rebates, far above what their volume should allow
While checking top @Polymarket traders on @graviahq , I noticed their Taker Rebates were suspiciously high. So I compared them with the weighted volume (wV) these accounts actually generated.
The result: NONE of them reached Obsidian tier (see screenshots). Most only hit Platinum, which is a 32% taker rebate.That's already a solid, well-earned rebate.
Per Polymarket's own rules, Obsidian requires $10M+ in volume over 30 days, and in practice more because of how wV is calculated. So why do certain traders get rebates beyond what their tier allows?
If they have custom fee terms, they likely also have uncapped referral deals, which adds another +15%. That puts total fee rebates at ~65%.
Polymarket previously said it ran an incentive program with select market participants to test fees. It also said the testing was over and all perks were removed.
So why do some traders STILL have an edge?
Accounts (not all):
0xc53375ff94e96100f2b30a4b5775db35218d69a9
0x3048d65321be3497164cdfc2996f94f98a2e7537
0x0ca4e9d27ff826a109ecbf74754fb66b93f277a6
0x0c7c5204404e9d5402d258fedac59c7212bae4cb
0xeebde7a0e019a63e6b476eb425505b7b3e6eba30
@Polymarket @PolymarketDevs , care to explain?
Hello @Polymarket @PolymarketDevs @devjoshstevens
As a trader on your platform, I've run into an issue with co-location verification. The process currently takes quite a long time, and there's no way to tell whether a verification has been rejected or is still under review, since right now you only send approvals
Also, co-location applications are submitted through Google Docs. That's convenient and fast, but it raises data security concerns and isn't a very professional approach.
It would be great if you used a third-party provider, such as Sumsub (the most popular one, as far as I know), to handle verifications and approvals. At the very least, users wouldn't have to worry about the security of their personal data
It would also help to speed up processing, or at least send rejection notices when a verification doesn't pass. Ideally, there would be a dedicated page showing application status and progress. For example, users could complete verification once and then add their accounts directly through a dashboard (Opus 5.5 can help with that 😁)
I hope you'll improve this part over time. I recently needed additional accounts for my bots, and the same issue that bothered me during my previous verification has come up again
This is what strategy overfitting looks like
There's no clear logical reason to pick this particular filter or this specific time mark
Not to mention that this isn't scalping
The reasons why any AI model from any company will lose on the financial market are very simple:
1. AI is a black box. You can't trace the full path of its reasoning and the logic behind its conclusions. Any even remotely sound system requires solid cause-and-effect relationships, which AI can't deliver because of how it's trained: it simply predicts the next token
2. For AI to "predict" outcomes, it has to be trained on specialized financial data, not on a mountain of text from every field imaginable. That's a narrow bottleneck
3. Even if you train AI on financial data, you won't be able to use it in strategies. Any model can predict the short horizon with reasonable accuracy, but over longer horizons it becomes a coin toss (if you're a trader(gambler), that'll suit you just fine)
4. High-frequency firms do use AI, but only to generate probabilities and predictions with some variance; after that, the math takes over. Under no circumstances would anyone give an AI direct trading access. Anyone who understands ML knows that out-of-sample results can be unpredictable to some degree (again, it's a black box)
If AI can trade stocks so well, why are all these AI-managed portfolios at Fidelity, Schwab and elsewhere struggling to even beat the S&P 500?
And where are the AI portfolios winning the U.S. Investing Championship?
Backtests and predictions are easy. Show me the audited, real-money results. That’s where the bullshit stops.
The answer is because it can't. There is zero evidence of any enduring or consistent results from purely AI or computer manager portfolios that can even outperform the S&P 500. Fact.
Rovac retweeted
This thread is 100% gonna blow up and I am gonna look like a salty c****again but @icobeast pissed me off so now it's gonna get ugly
Kalshi fakes their crypto volume and I can prove it
NOTHING pisses me off more than watching a company treat its own customers like complete fucking idiots
Grab some popcorn cause I dissected every piece of evidence one by one until there’s basically no way for him to respond without making this look even worse
Oh and FYI, I left every source at the end so y’all can check the receipts yourselves 😘
1/x
That all sounds good until you realize that AI for trading is a black box, and nobody in the financial industry uses either LLMs or Jev for trading, and nobody will. For that, they use ML like RN or transformers that are trained on specific data, not a random data like any AI from any lab
The only way to use any AI for trading is if you have $10M and you want to fuck it all away down to 0
卧槽，JEV 的出现很可能就是自动智能交易的起点。
昨天JEV在AI 圈刷屏，今天 Monad 工程师就用JEV做了一个自动交易机器人。
实时读取 MON/USDC 价格，JEV 不负责写分析报告，也不跟你解释一堆逻辑，它只做最核心的事，判断 Buy 还是 Sell，再给出置信度，然后直接把交易发到 Kuru 的链上订单簿执行。
这一下我突然明白 JEV 为什么要做成一个不会说话的模型了。
交易根本不需要 AI 每 300ms 给你写一篇小作文，它需要的是：
行情进来→判断→下单→新行情→重新判断，而且这一整套循环必须足够快、足够便宜。
而 Monad 现在刚好每 300ms 出一个区块，Kuru 又是链上订单簿。高性能链负责执行，JEV 负责决策，两个东西拼起来，已经有点AI 原生交易系统的味道了。
未来的交易 Agent，不一定需要一个会写研报的超级大模型，它更需要一个能在极低延迟下连续做几百万次判断的大脑。
昨天我 提交的WL今天也通过了，接下来我先尝尝咸淡，看看到底怎么个事。
22% of Markets on Polymarket Are Subject to Manipulation Daily
Based on the data and formulas published in the article “Settlement Manipulation in Prediction Markets,” I decided to check the percentage of markets that were subject to manipulation during the period from June 8 to July 17.
The percentage of markets subject to manipulation is growing. According to the calculations, in June the average number of such markets was 8.9%, whereas in July it was already 11.1% an increase of 2.2%.
Judging by actual trading, there is 1.5–2 times more manipulation, which could amount to 22% of the total number of markets. The calculation does not account for situations where manipulation occurred earlier than 10 seconds before the market’s close, nor for failed attempts at such manipulation.
Will TWAP60 help solve this problem? Partly.
Let's suppose a situation where the price of the underlying asset is $60,000 and the target price is $60,010 meaning that to change the outcome, we need to raise the price by $10. The question is: by how much do we need to raise the price of the underlying asset, and hold it there, in order for the outcome to change?
Below is the calculation.
Seconds until market close: amount by which the price of the underlying asset needs to be raised
60s: $10
30s: $20
20s: $30
15s: $40
10s: $60
2s: $300
1s: $600
The number of manipulations will naturally drop, but won't disappear entirely. They will become more expensive to execute, and fewer people will be able to do it but manipulators will also earn more. Thanks to TWAP60, markets will become more confident in outcomes and will press harder toward the edges of the market, which will allow buying cheaper and earning more money for a small handful of people with large capital.
POLYMARKET NEED TO PAY OUT $300M to the gamblers speculators who bought shares at $0.003 just to keep them on its market, only to never make that money back in the future because in that case every gambler would have to spend $3k on fees (100k gamblers).
Or even better, just go ahead and shave the evil UMA manipulators and the whales who are buying up No shares basically a global redistribution of wealth to the proletariat.
Is that really how the fuck you imagine this?
The fairest decision on Polymarket’s side would be to resolve the market as No, but compensate those who held positions before May 31, 11:59 PM, because people genuinely bet on this happening
As for the speculators who bought shares after the market had effectively ended, their accounts should be banned, because they weren’t originally betting on the event happening and are now squealing like stuck pigs
For all the degens who voted "Yes" on the MSTR market, do I understand your logic correctly: if we have a market that resolves on a specific date, and then a week later it turns out that the event actually happened during that period, we should go back and reconsider the outcome of the market?
Would you personally pay the winnings out of your own pocket to those people?
Precedents like that would completely fuck up every market that has a resolution date tied to a specific point in time.
Fuck up the latency to 2–3s, bring everything back to normal, then say we improved latency х5✅
Friday Update - Week of May 18th, 2026:
- CLOB latency: Still the team's top focus. Today's release cut latency by roughly 5x, with more in the pipeline and another substantial one coming next week
- Taker rebate program: Released, kicks off May 29. Details: docs.polymarket.com/trading/…
- Unified TypeScript and Python SDKs are live (beta): Both SDKs went out yesterday under one consistent surface across discovery, market data, trading, account data, and realtime. Docs: docs.polymarket.com/dev-tool…
- Coming soon: Unified API, Rust SDK, and React SDK
- Gamma DB migration completed: Faster reads, lower replica lag, and more room to scale. Markets and events data will be fresher and quicker to update
- API Gateway: First implementation slices live on Gamma endpoints, with the Developer API Keys boundary and observability foundation in place
- MFA on withdraw: In final review
- Large World Cup feature: End-to-end testing lined up for next week. More details soon
- Gamma / Data API: builderCode exposed on /v1/builders/leaderboard and /v1/builders/volume in data-api v0.15.10. Activity feed market-title fix shipped. Public API additions for structured sports fields, team ordering, and event sport metadata. OpenAPI migration advancing across more route groups
