When the Books Show 3x, You Can Only Take Home 30%

Deep thoughts on AI and aspirations —— ByteDance Deep Thinking Circle

In fall 2023, friend.Tech was the hottest product in crypto. Spend some ETH to buy a KOL’s Key, get into their private chat, and sell the Key anytime. Simple design, almost everyone’s account showed 2-3x returns, screenshots everywhere.

But between the price displayed on screen and the value you can actually take out lies a wide gap. Understanding this gap is worth far more than the product itself, because it appears in every closed-pool product where “everyone is making money,” and it will appear again and again.

The Moment You Enter, You’ve Already Lost Nearly 20%

First, the rules. Friend.Tech uses an extremely simple pricing curve: Key price equals the square of holder count divided by 16000. More holders means higher price, and the rate of increase itself accelerates—each new entrant pushes the price higher than the last, a textbook Ponzi curve. Every transaction charges a 10% fee, split equally between the protocol and the Key owner.

On the surface, trading cost is 10%. Run the actual numbers and you’ll see this figure is misleading. Suppose you enter with 1.1 ETH. Buying requires paying 10% fees, so you can only purchase Keys worth 1 ETH. When you sell, you pay another 10%. From the moment you hit buy, your position’s realizable value locks in at 0.9 ETH—the exit fee is guaranteed, just deferred.

Round trip, real loss is 1 minus 0.9 divided by 1.1, about 19.2%. Price needs to rise 22% just to break even.

The official explanation is “high fees encourage long-term holding.” This doesn’t hold up: if the goal is encouraging holding, taxing sellers would suffice. Taxing buyers 10% too doesn’t encourage holding, it just increases extraction. To judge a mechanism’s true intent, look at who it charges, not what it claims.

Three People Splitting One Cow: The Illusion of Book Value

The real sleight of hand is on the second layer. A thought experiment: Zhang, Li, and Wang pool money to buy a cow, a duck, and an egg, agreeing whoever exits first gets the cow, second gets the duck, last gets the egg. All three believe they “own” a claim to one cow, but there’s only one cow. Average across the six possible exit sequences, each person truly owns one-third of a cow, one-third of a duck, one-third of an egg.

Friend.Tech uses this exact illusion when everyone’s books show 2x. This product has only one counterparty: the pool. The TVL in the pool is the actual liquidatable capital, while the displayed Key price is a book figure calculated by the curve. For instance, when Key count is 40, price is 0.1 ETH, book market cap is 4 ETH, but the pool actually contains only about 1.38 ETH.

Plotted this way, once Key count exceeds roughly 20, realizable value as a percentage of book value stabilizes around 30%, never rising higher. This 30% has two implications. First, buyers entering during the flat curve segment lose about 70% of expected value instantly, on top of the two fee layers. Second, the Room Value displayed on screen is inherently over 3x the realizable value. So-called “everyone made 2-3x” is largely just mistaking a 3x book illusion for money.

Every point of paper profit you earn comes from diluting the expected value of the next buyer. This accounting doesn’t require a collapse to hold true—it’s happening every moment.

The Peak Doesn’t Wait for Collapse, It Walks Down on Its Own

Next comes the most elegant and cruel part of this structure: it self-collapses without any external shock.

The reasoning chain goes like this. Assume growth caps at N participants, then buyers entering after position N-M are guaranteed to lose money. Everyone can see this information, so rational actors won’t buy after N-M. But if nobody buys after N-M, then buyers after N-M-L can’t recover either, so even earlier positions find no bids. Layer by layer backward, equilibrium price continually migrates downward. This is the classic “beauty contest” reasoning from game theory: everyone is guessing when everyone else will stop, so the stopping point keeps advancing.

The actual micro behavior: once net inflow slows, expensive Keys become unprofitable first, speculative capital rotates to cheaper Keys, repeat cycle, individual Key price peaks step down one by one. From a book value perspective breaking even looks easy—5 ETH spent needs only 27 new buyers; from an expected value perspective, 1 ETH spent needs 115 new buyers to break even—while new buyer supply itself is drying up.

Making matters worse are bots. Friend.Tech is infested with bots camping on new Key issuance in the low-price zone. When the equilibrium price for new launches drops to a certain level, it falls directly into bot arbitrage range, shaving another layer off ordinary users’ expected value.

The popular mutual-buying etiquette in that circle (you buy my Key, I buy yours) can’t hold either. Two people mutually holding is indeed stable, like exchanging hostages; add more participants, front-running becomes profitable, once the chain of suspicion forms, final equilibrium is everyone loses together. This etiquette looked solid during the upswing only because paper gains masked the continuous drain of expected value; once inflow stops, front-running immediately becomes the dominant strategy.

The Business Model Is Actually a Pump

Add up all the friction and the product’s true face becomes clear.

Based on data at the time: at about $36M TVL, cumulative fees had reached $24M, with the protocol taking $12M. Even by the most conservative measure, inflow was about $48M, with a quarter converted to protocol revenue in roughly two months. Project forward: book market cap about $110M, if daily turnover is 5%, about $16.5M gets extracted monthly, nearly 45% of TVL. All net deposits flow continuously to the protocol.

For scale comparison: NFT marketplace OpenSea charges 2.5% royalty plus 2.5% fee round trip, here it’s 20% round trip, four times the friction. At this extraction rate, debating “whether this product is a Ponzi” becomes secondary—it’s first and foremost a machine continuously pumping away existing capital, the Ponzi structure is just the shell that keeps the pump running a bit longer.

This arithmetic can be applied to audit any closed-pool product, three numbers suffice:

If any of these three numbers significantly exceeds comparable markets, the displayed yield isn’t worth trusting.

Its Intended Exit, and Why It Failed

To be fair, this product’s design seriously considered an exit. The exit was logically sound: if real services emerged behind Keys—hosts continuously producing content, maintaining relationships, giving holders actual benefits—these “utility holders” would become subordinated claimants, not rushing to sell, genuinely improving remaining holders’ expected value. Back to the thought experiment: if Zhang commits to exiting last, what Li and Wang hold changes from “pick one of three” to “pick one of two,” gold content immediately rises.

Subsequent facts gave this hypothesis a complete test, result: falsification. After the airdrop landed, users concentrated on selling and exiting, activity collapsed, product heat cooled within about a year. Utility demand didn’t grow, but speculative demand did jam the exit channel.

My judgment is one layer more pessimistic than “utility didn’t grow”: speculative cold-start actively blocks the path to utility economics. Speculation’s pricing logic is front-running and flipping, utility users’ pricing logic is service valuation—two logics on the same curve cannibalize each other. A “group membership” bid up by speculators to several ETH repels people actually wanting the service—the service itself might be worth tens of dollars monthly, but the entry ticket has been priced by gamblers to another order of magnitude. Early speculators raise the barrier for all later entrants, the product quotes real customers at peak price. Cold-start borrows speculation’s power, then pays it back with interest.

The boundary of this conclusion must be stated clearly. The 30% gold content estimate is built on the assumption “all buyers are speculators”—if utility demand truly exists, the percentage would exceed 30%. So to judge whether a similar product has hope, don’t look at narrative, look at two observable variables: issuer’s self-held percentage, and the proportion of non-speculative holdings (those who paid for usage rights and don’t plan to flip). If these two percentages don’t rise, book value remains forever illusory; if they do rise, even closed pools can grow real business.

Back to the initial question: facing any closed-pool product where “everyone is making money,” discount the displayed price by 70%, then decide whether to enter. 70% isn’t a precise figure, it’s a reminder: when your counterparty is a pool rather than a market, the number on screen never promised to be your money.