海盗鸭

美股跌的时候,债市那边更值得看一眼。30 年期美债收益率周三盘后又跳 6 个基点,上到 5.2% 以上,贴着 2007 年以来最高。10 年期 4.70%,2 年期 4.28%,曲线明显走陡。UOB 的说法是近端紧缩预期降了,长端的通胀和财政担忧起来了。

「多地高分生扎堆填报中本贯通」上热搜，评论区直接吵成两派。

一派说这是清醒：中考分流压力下，用一个不算低的分数换一条七年直通本科的确定路径，比在普高卷三年再赌一次高考划算太多。
另一派说这是提前把人生赛道锁死：十五岁选的专业，等二十二岁毕业，行业可能已经换过一轮了。

我偏向前者一点点——这个年代，能把不确定性换成确定性本身就值钱。但前提是：那个专业你至少不讨厌。

如果是你家孩子，这个志愿你会不会填？

上海地铁多条线路今天全线停运，早高峰算是被直接掀桌了。

地铁一停，暴露的是通勤的单点依赖：平时四十分钟的路，网约车排队几十位、公交在积水里挪、共享单车骑出去等于送人头。对那些住郊区、每天靠一条线进城的人来说，这不叫「晚点」，叫直接出不了门。

停运本身是对的，把人扣在家里，比让人挤在风雨里安全得多。真正该改的是另一套逻辑——把「到不了岗」当成员工个人问题的那套考勤规则。

今天你是怎么解决出行的？

A sticking point in Clarity is the president’s own crypto business. After Senate Banking Committee Democratic staff reviewed the revised ethics rules from July 22, their conclusion was that the loophole remains: money can still be taken from World Liberty Financial tokens, stablecoin reserves, and the memecoin split through intermediaries and authorization agreements. The Office of Government Ethics disclosed that Trump’s crypto income last year was about $1.4 billion.

OpenAI calms employees during internal meetings. CFO Sarah Friar said that annualized revenue for July has already surpassed that of the entire second quarter, and added, “The second quarter isn’t bad.” The backdrop is that Anthropic disclosed annualized revenue of $47 billion in May, with roughly $10 billion for the full year of 2025. This year, its valuation has already overtaken OpenAI.

The Federal Reserve held steady 9-3, keeping interest rates at 3.5%-3.75%, marking the fifth hold. All three dissenting votes were in favor of raising rates: Hammack, Kashkari, and Logan. The three voted for a rate hike at the same time; the last time was in September 2016. After Warsh took office, he has removed forward guidance from the post-meeting statement.

The problem with this Bitcoin cycle isn’t that “someone is dumping.” It’s that three layers of buying power went out at the same time—yet their speeds are completely different.

ETF outflows are the pace of the allocation players, measured over quarters. Advisors and institutions change their allocation ratios through meetings and decisions—starting is slow, turning around even slower.

Perpetual futures buying weakening is the pace of leveraged players, measured over hours. It’s the most sensitive to price: it comes quickly and leaves just as fast, and usually accounts for most intraday volatility.

On-chain capital stagnation is the middle layer, measured over weeks. It reflects real ownership transfer.

In normal times these three layers are out of sync: leveraged players manufacture volatility; on-chain funds absorb it; allocation players set the long-term slope. When one layer goes quiet, the other two can hold up for a while.

When all three layers stop at once, what happens isn’t a crash—but something else: pricing power shifts from the “marginal buyers” to “the least willing sellers.” The support around $64,000 is essentially made up of long-term holders’ “staying put,” not of someone truly putting fresh money in.

This condition is easy to misread as stabilizing. What it really means is that liquidity has thinned—and thin cuts both ways. With the same amount of capital, selling into a drop can dig a bigger hole, and buying into a rise can pull an even steeper angle. So at times like this, the rebound magnitude shouldn’t be taken as evidence that demand has recovered.

In Fidelity’s Q3 report there’s a related benchmark: this correction has lasted 203 days, whereas similar phases in 2018 and 2022 lasted 298 days and 299 days respectively.

By this ruler, time may be even more worth watching than price.

Right now, are you waiting for the price—or waiting for time?

USO has risen 87% this year, outperforming the crude oil futures it tracks. The reason is that the Iran war has turned the oil curve into a deep backwardation, with near-month contracts expensive and far-month contracts cheap. Rolling positions every month doesn’t lose money—it actually makes money. Starting mid-July, the curve turned back into contango. On August 5, USO fell 5.19% in a single day. For the same fund, over the past decade it has risen only 51%.

Miners sell electricity to AI—this bill is unsolvable in the short term; in the long run, it’s a wager on an opportunity.

First, look at the drivers. MARA CEO Fred Thiel put it in one sentence: with the same unit of electricity, the money made from powering AI is far higher than from mining. Around BTC at $64,000—down from nearly half of last October’s peak—hashprice has fallen to a historical low of about $34–35 per PH/s. Institutional miners are facing volatile returns against 15-year fixed leases.

The scale isn’t small. In its first-quarter report, CoinShares counted over $70 billion in AI/HPC contracts disclosed across the industry, and by May nine listed mining companies had already announced a pivot. Bitfarms changed its name to Keel Infrastructure in April, and the CEO said, “We are no longer a bitcoin company.”

But there’s an irreversible mechanism at work. Building an AI facility costs $8–15 million per megawatt, while mining only needs $0.7–1.0 million. This price difference isn’t just money—it’s options. Mining operations can turn on and off in response to prices at any time, while those who sign multi-year leases can’t. Miners’ most valuable asset is flexibility, and now they’re selling it off to buy certainty.

The core of the objection raised by Dragosch, head of research at Bitwise Europe, lies here: the realization of AI compute demand may be slower than the investment cadence, and he believes BTC is approaching the end of its downside leg. If that combination plays out, miners could miss the restoration of mining economics while their capacity is locked in.

In Q2, total network hashrate fell 5.8% to 1004 EH/s. Is this the loss of security—or is it a form of clearing?

Cramer’s take: this rebound in the software sector shows how quickly Wall Street can turn around—AI infrastructure stocks that have taken a beating may be next. You can compare the same week’s market performance: the Semiconductor ETF fell more than 4% in a single day, and Vertiv dropped 17% in one day because revenue was down by more than $100 million.

Get $217 Back for Every $1,000 Note That Matures: A Treasury-Mode Bill That Finally Landed in Retail

According to CryptoSlate, a batch of structured notes linked to Strategy (MSTR) from Goldman Sachs is set to mature soon. Based on MSTR’s July 24 closing price, for every $1,000 face value, about $217 will be returned—equivalent to 22 cents per $1. The final numbers are subject to GS&Co.'s calculations.

These products are worth understanding because they’re being sold in large volume to people who want a little exposure to BTC but don’t want to buy crypto directly.

Structured notes typically look like this: an upside cap—no matter how much the underlying rises, you only receive a fixed coupon (or a capped participation rate); a downside buffer—if the decline stays within a threshold, you don’t lose, but once it’s broken through, you take the full 1:1 downside loss.

Translated into options language: the person buying the note is essentially selling a put option, and the coupon is the premium you receive.

And that’s how the entire leverage chain closes: when the coin price rises → the treasury company’s stock trades at a premium to net asset value per share → issuance of more stock and convertibles to buy even more crypto → the investment bank packages this high-volatility exposure into a “fixed-income product with coupons” and sells it to retail → when the coin price falls, the volatility bill is settled starting from the far end of the chain.

Note: there’s no fraud here—the risk disclosure is written in black and white. The only issue is that the 22-cent result requires the underlying to drop to a certain extent, and at the moment of purchase, nobody believes it will fall that far.

In the future, when you see something like “linked to [some] stock, 15% annual coupon,” ask yourself first: am I collecting rent, or am I selling insurance?

"The bad news is gone, so it’s time to pull up"—what’s wrong with this saying, and why it’s worth taking apart carefully

At the last monetary policy meeting, the Fed kept its interest rate at 3.5%–3.75%. According to Decrypt, Chairman Kevin Warsh didn’t provide any new signals about when rate cuts or hikes might happen. After that, both BTC and ETH ended up trading sideways. At the time, many people thought, "The bad news is priced in—this rebound is coming." The result was neither a crash nor a rally.

Why does "priced in means rebound" so often fail? Because it mixes two different things into one: expectation adjustment, and liquidity changes.

There are two main pathways through which rate decisions affect prices. The first is the expectations path—markets price in outcomes in advance, so what drives the move is never simply "what the interest rate is," but rather "how far the actual outcome deviates from the market’s bets." When the result fully matches expectations, the energy from this path is zero: it’s neither bullish nor bearish. The second is the liquidity path—factors such as the size of bank reserves, the Treasury’s account balance, and changes in reverse repos determine, at the margin, how much money can flow into the highest-risk asset categories. This path operates on a weekly and monthly basis and has little to do with the declaration on a given day.

So the scenario where "the bad news is gone" truly holds only has one case: the market originally priced in an even worse outcome, but reality is not as bad, forcing shorts to cover. When the decision contains no new information, prices revert to being driven by the second path—meaning they continue to grind along slowly with liquidity.

Next time you see someone say, "No matter what happens tonight, it has to go up," you can rephrase it with two different questions: How big is the gap this time between the outcome and the interest-rate futures pricing before the meeting? Are the liquidity indicators this week expanding or contracting? If you can’t answer, that’s emotion—not logic.

Which path do you trust more?

Data center power and cooling provider Vertiv fell 17% in a single day. CEO Gio Albertazzi’s explanation was: revenue growth didn’t meet targets due to temporary project scheduling and supply-chain factors, not weakening demand—“this is temporary.” On this AI infrastructure chain, the market has little patience for the word “temporary”; it now has only one trading day.

Morgan Stanley will launch Ethereum and Solana ETPs (MSSE / MSOL) on July 28 with a 0.14% fee—lower than Grayscale’s Ethereum product (0.15%) and Franklin’s Solana product (0.19%). Media headlines are unanimously framed as a “fee war.”

But fee rate is probably the least important number in this story.

What truly determines returns is: actual return = staking yield × staking allocation × time actually staked − fees.

MSSE’s documents state a plan to stake 50%–80% of holdings. Gross staking rewards are about 95% allocated to the trust and shareholders, with 5% going to service providers such as Figment and Galaxy—sounds generous. But the same document hides a crucial detail: as of July 6, the Ethereum validator activation queue is about 2.71 million ETH, with an estimated wait time of 47 days. During the queue period, the ETH earns nothing.

For Solana, the prospectus says the binding time is 2 to 3 days.

For the same new inflow, MSOL starts generating yield almost immediately, while MSSE might sit idle for about a month and a half. The more frequent the fund flows in and out, the heavier the drag. A 1-basis-point fee advantage and a 47-day zero-yield window are not on the same scale—one makes the headlines, the other stays in the footnotes.

Morgan Stanley’s real weapon isn’t the fee rate: it’s nearly 16,000 financial advisors and around $2.6 trillion in client assets. On day one, the two products together saw roughly $38 million in total trades—not exactly a blockbuster—but distribution takes effect on a quarterly schedule, not on a daily basis.

Would you switch funds for 1 basis point?

What happens after the Dow drops 1,000 points in a single day? CNBC goes through historical samples: in the following week, performance is usually still weak, but the average returns at one month and three months are positive. The sample size is small and it doesn’t establish a pattern. Still, every time there’s a big selloff, this table gets revisited again.

Two Ethereum news stories on the same day may look unrelated, but they’re actually about the same thing.

First: In this wave of perpetual contract hype, trading activity basically doesn’t happen on Ethereum’s mainnet. L2s like Arbitrum and Base take most of it, while specialized chains like Solana and Hyperliquid capture high-frequency execution. Ethereum is being repositioned as the settlement layer and the collateral layer—not about who’s faster, but about being that “last place for settlement.”

Second: The Ethereum Foundation added security researcher pcaversaccio to its board of directors, with a one-year voluntary term. That brings the board from three to four members. This happens after the Foundation cut 20% of its staff in June and reorganized its R&D into five units.

In parallel: the technical narrative is giving up the idea that “execution happens here,” shifting to “security and settlement happen here.” Meanwhile, the organization puts a security researcher on a four-person board while shrinking the R&D side. Two expressions of the same bet.

The bet itself is clear: the moat is neutrality and security, not throughput.

But there’s a step that must be proven—after execution moves away, how does value flow back? L2s will have to keep paying for settlement and data availability, and ETH really needs to be put to large-scale use as collateral. If these two don’t hold, then “the settlement layer” is just a polite phrase.

Do you think the settlement-layer narrative is the moat—or just the steps to climb?

The Federal Reserve kept rates unchanged on Wednesday, the Dow fell more than 2% that day, marking its worst single-day performance since April 2025.

The pressure is not coming from policy interest rates, but from the long-term U.S. Treasury yields that jumped after the meeting—rates stayed the same, but long money moved first. The U.S. stocks took the hit first—what about you on the crypto side, has your positioning changed?

Lockheed Martin wins a $58.6 billion U.S. government contract to produce Patriot air defense missiles. What does $58.6 billion really mean: it’s more than the total market cap of many S&P 500 constituents, and this is just a single contract—not an annual budget. The bills of geopolitics ultimately end up in the order books of a few companies.

MarketWatch breaks down the numbers: Since the Iran conflict began, the U.S. oil fund USO has risen 58%, more than double the increase in the WTI crude oil futures over the same period. The same barrel of oil, two different numbers. The difference comes down to the roll-over—crude futures have to switch contracts every month, and the market structure determines whether this month-to-month roll costs money or makes money.

On July 28, Saylor posted that the biggest threat to Bitcoin is not external attacks, but an internal “change of the rules”: consensus rules are like the constitution, and any modifications must be rare, conservative, and done out of necessity rather than ambition. He specifically called out and opposed BIP-110, covenants, and increasing the block size.

There was another piece of news in the same week as well: researchers proposed a new scheme aiming to keep wallets compatible with existing address formats even in the post-quantum era.

Put the two together, and the question becomes concrete—when it comes to quantum, does it count as “necessary”?

Look at the numbers. Project Eleven’s statistics for this year’s February show that about 6.9 million BTC are in addresses where the public keys have already been exposed, representing roughly one-third of the circulating supply. The definition cited by the BIP-361 document is that as of March 1, over 34% of Bitcoin already had exposed public keys.

Breaking it down is even more interesting. Of those 6.9 million coins, about 4.99 million (72%) come from address reuse—once you spend money there, the public key gets written to the blockchain. Strictly speaking, this isn’t a protocol flaw; it’s a usage habit. Without changing any protocol today, simply switching to a new address fixes it.

What truly can’t be fixed is the other ~1.72 million P2PK coins, where the public key is directly embedded in the output script. Of these, about 1.1 million belong to Satoshi Nakamoto. Migration requires the private keys; without them, no protocol upgrade can save the situation.

So the real structure of this debate is: the part that can be fixed doesn’t need to touch consensus, while the part that would require changing consensus is exactly the hardest portion to fix.

Would you accept a hard fork once to defend against quantum threats?