The thing I keep relearning is that a narrative almost never dies from being wrong. AI infrastructure, DePIN, real world assets, whatever the theme of the moment is, the story is usually still intact when the tokens are down sixty percent. What actually kills the trade is crowding. Everyone who was going to buy the idea already bought it, and now there is no marginal buyer left to sell to. The story survives, the flows do not, and the two things come apart in a way that is genuinely hard to feel from the inside because everyone around you still believes.
So I stopped trying to time the top of a hot sector by vibes, because my vibes are a coincident indicator at best and usually a lagging one. Instead I try to measure how crowded the trade has gotten, since crowding leaves a paper trail. It shows up in funding, in attention, in how the leaders and laggards are priced relative to each other, and in how many new tokens are being minted to ride the same wave. None of these are perfect. Together they give me a read I can actually act on.
The four things I actually look at
Funding rates across the sector are the first tell, and the most honest one. When a narrative is early, perp funding is choppy and sometimes even negative because plenty of people are fading it. As it matures, funding on the sector's names goes persistently positive and starts clustering high. Everyone is levered long the same idea in the same direction. I do not care about one token's funding on one day. I care whether the whole basket of names tied to the theme is paying to be long, week after week, with the aggregate sitting near the top of its own historical range. That is leverage stacking on one side of a boat, and boats like that capsize on news that would have been a shrug three months earlier.
Social dominance share is the second. Not raw mention counts, which just track price, but the sector's share of total crypto conversation. Every big narrative goes through a phase where it eats an outsized slice of the attention pie relative to its actual market cap. When a theme that is a modest fraction of total market cap is generating a much larger fraction of the talk, you are near the part of the cycle where the marginal participant is a tourist, not an allocator. Attention is a resource with a ceiling. Once a narrative is most of what people are talking about, there is not much more attention left to capture, and attention is what was funding the bid.
Valuation dispersion between leaders and laggards is the third, and it is the one most people ignore. Early in a run, capital is selective. The one or two best names in the theme go up and the junk lags, so the spread in valuations is wide. Late in a run, the junk catches up. The nineteenth best token in the sector is suddenly trading at a multiple that only made sense for the leader, because buyers have run out of good names and are reaching for anything with the right tag. When dispersion compresses and the laggards start printing leader-like valuations, the smart money has usually already sold the leaders to the people buying the laggards.
New launches tagged to the theme is the fourth, and it is the supply side of the same story. Founders and market makers are not stupid. When a narrative is hot, the rational move is to ship a token that fits it, because the tag alone buys you a valuation. So the rate of new launches carrying the theme's label is a direct read on how much fresh supply is being manufactured to meet demand. When you go from a trickle of new names to a flood, the sector is getting diluted in real time, and every new launch is competing for the same finite pool of buyers and attention.
Building the composite read
I turn these into a single crowdedness score because any one of them will fake you out. Funding can spike on a two day squeeze. Social dominance jumps on a single headline. What matters is when several of them light up at once. My rough process looks like this.
- Score each of the four on a zero to one hundred scale, where the number is roughly where that metric sits inside its own trailing range for this sector, not an absolute level. High funding relative to this narrative's own history, not relative to some universal threshold.
- Average them, but weight funding and dispersion a little heavier, because in my experience those two lead and social dominance tends to confirm rather than warn.
- Watch the slope more than the level. A composite grinding from the fortieth percentile to the eightieth over a few weeks is more actionable than a reading that has been pinned high for a while, because a pinned high reading often means the unwind already started and you are looking at a hollow price.
When I look back at the late stages of the big narrative runs, the AI wave, DePIN, the RWA push, the pattern rhymes even though the exact numbers differ. Funding across the basket goes persistently positive and sits high. Social dominance runs well ahead of the sector's share of market cap. Dispersion compresses as the laggards close the gap to the leaders. And new launches carrying the tag go from occasional to constant. You rarely get all four screaming on the same day, but you typically get three of them elevated together in the weeks before the theme rolls over. That cluster is the signal. Any single one alone is noise.
Turning the read into an exit
A score is useless if it does not change what I do, so I tie it to actions ahead of time, while I can still think clearly.
- Composite in roughly the top quartile of its own range, with the slope still rising, is when I stop adding. No new capital into the theme. I let winners run but I quit averaging up.
- Three of the four metrics elevated at once, with dispersion compressing so the junk is pricing like the leaders, is my trim trigger. I take the position down in steps rather than trying to nail one exit, because I will be early and I would rather be early in tranches than late in one panic.
- New launches flooding in while social dominance rolls over from a high is my rotate signal. Supply is climbing while attention is fading, which is the worst combination for anyone still holding, and it is usually the cleanest tell that the marginal buyer is gone.
The failure mode I want to warn you about is confusing crowded with over. Crowded can stay crowded for a while, and if you dump the whole position the first time the score hits the top quartile you will leave a lot on the table and hate yourself. That is exactly why I scale out on a rising, clustered read instead of trying to call the precise top. Being early and partial beats being exact and all in, because being exact is not a thing you get to be.
The other honest caveat is data quality. Sector tagging is messy, funding is uneven across venues, and social metrics are noisy and easy to game. I do not treat any of this as precise. I treat it as four independent witnesses, and I act when three of them tell the same story. Pulling those four feeds together, funding across a basket, dominance share, the leader to laggard spread, and launch cadence, is fiddly by hand, which is a chunk of why we built the scorecards and sector views inside Blockcircle the way we did. The point of the whole exercise is not to be right about the top. It is to have decided your exit criteria before the sector was crowded, so the decision is already made when everyone around you still thinks it is early.