Attention Is Not Volume: Two Numbers, One Confusion

Volume counts value that moved through a pair. Attention counts people who chose to look. Neither implies the other, they are not in the same units, and only one of them can be read from the chain. Most confusion about launches comes from using the countable quantity as a stand-in for the one that matters, and the four cases below are where that substitution breaks.

Two quantities, two units

Volume, or turnover, is denominated in value: SOL, dollars, whatever the quote asset is. It is a sum over trades that occurred inside a window and it is exact, in the sense that everyone computing it over the same window and the same venues will get the same answer. Its exactness is why it appears in every ranked list.

Attention has no unit anyone can agree on. If you tried to define it you would have to choose between people who saw something, people who stopped on it, people who read it, and people who remembered it an hour later. Those are four different populations, all of them invisible to a blockchain, and the choice between them is a research design rather than a measurement.

The gap is not a technical limitation waiting for better tooling. Chain data records that an address signed a transaction. It does not record why, where the person read about the token, whether a person was involved at all, or how many other people saw the same thing and did nothing. No amount of indexing recovers information that was never written down.

What volume actually counts

Turnover is a sum of trade values, and everything that produces a trade contributes to it equally. That includes discretionary buying, arbitrage closing a price gap between two pools, routed legs of a single user request landing as several swaps, market-making inventory adjustments, and activity produced deliberately by a team or a vendor.

None of those categories is hidden. All of them are in the transaction record, distinguishable to varying degrees by patterns in signers, spacing and symmetry. What turnover does is add them together and present the total as one number, which is exactly what a sum is supposed to do and exactly why the number cannot answer the question people ask of it.

A second property makes turnover harder to reason about than it looks: it is a flow, and flows are measured against a stock that nobody puts next to them. Nine hundred SOL of turnover through a pool holding nine hundred SOL and the same figure through a pool holding nine SOL describe completely different situations, and the ranked list shows the same cell for both. Dividing by depth is the cheapest correction available and almost never appears on a screen.

The one thing turnover reliably tells you is that a pair was used and that fees were generated for whoever provided the liquidity. Compared against the pair's own history it also tells you that something changed. That is a starting point for an investigation, and treating it as the conclusion is the error this whole note is about.

What measuring attention would require

Suppose you wanted to measure attention properly. You would need a count of distinct people, not addresses, who were shown the token; a count of how many of those stopped; some record of what they did next; and a way to tell whether the same person was counted twice across two surfaces. Every one of those requires data held privately by platforms, and even they see only their own slice.

What a team can actually gather is much smaller and still worth having. In space you operate, you can count joins, replies, questions and returns. Those are countable events produced by people, which is more than any chain metric offers, and they are biased toward the people already close enough to be in your room.

The correct description of those numbers is proxy, not measurement. A proxy used honestly is stated as a proxy, with its bias named. A proxy presented as a measurement becomes a number in a deck, and the deck outlives the caveat every time.

The four divergence cases

Silent volume is the case that generates the most bad analysis, because it is the one that looks like success on a screener. Loud stillness generates the second most, because it looks like failure on a screener while being the healthier of the two positions if the people talking are real and stay.

Each case also decays differently, and the decay is more informative than the snapshot. Silent volume ends abruptly, because whatever produced it was funded and funding stops. Aligned activity fades gradually as the event that caused it recedes. Loud stillness either converts into transactions over days or dissolves into people drifting away, and which of those happened is visible within a week without any special tooling.

The fourth case deserves less contempt than it gets. Empty is the ordinary state of almost every token almost all of the time, including tokens that later do something. Reading emptiness as a verdict causes teams to abandon work that had not yet had time to reach anybody, and the alternative reading, that nothing has happened yet, is usually the accurate one.

A worked case of complete disagreement

Proxies, and how to use them without lying

Proxies are unavoidable and are fine as long as their status is stated. The rule that keeps them honest is that every proxy is reported with the thing it cannot see, in the same sentence, so the caveat travels with the number instead of being stripped off downstream.

• Distinct signers over a window: cannot see whether one person controls several addresses.
• Chat joins: cannot see whether the joiner ever reads anything, and includes accounts that join everything.
• Replies from new accounts: closer to a person, and biased toward the loud minority who post at all.
• Returning members across days: the strongest cheap proxy, and blind to everyone who lurks silently.
• Holder address counts: cannot distinguish distribution from one participant fanning out across addresses.
• Pooled depth added by others: real commitment, and it measures capital rather than people.

Read together these give a shape rather than a number, and a shape is what the situation actually has. A team that reports a shape can be argued with productively. A team that reports one headline figure has made the discussion about the figure.

There is one more discipline that costs nothing and is skipped almost universally: recording the proxy at a fixed time each day rather than whenever somebody remembers to look. Proxies read opportunistically are read when they look interesting, which builds a series that only contains peaks. A series sampled on a schedule contains the dull days too, and the dull days are what make the peaks legible later.

The same applies to the denominator nobody writes down. Twenty replies is a different fact in a room of eighty people than in a room of four thousand, and a team that records only the numerator will eventually conclude that engagement collapsed when what actually happened is that the room grew. Every proxy in the list above is worth storing alongside the population it came from.

Why the two get conflated

The conflation is inherited. In markets with known participants and regulated reporting, turnover genuinely does carry information about breadth, because producing it requires many parties who cannot trivially be the same party. That inference is reasonable in the setting it came from and does not survive the move to a permissionless chain where anyone can create addresses and pay fees.

Language does the rest of the work. The word activity sits comfortably in a sentence about markets and in a sentence about people, and it slides between the two without anyone noticing the switch. A pair with high activity and a community with high activity are describing unrelated things, and the shared word makes the substitution feel like a restatement rather than a claim.

It is also reinforced by the surfaces themselves. Ranked lists put turnover in the largest column because turnover is the quantity they can compute reliably, and a number displayed prominently acquires an authority it never claimed. The list is honest; the reading of it drifts.

The commercial layer completes the loop. A Solana volume bot platform exists because activity-ordered screens are a real distribution surface, and the category is straightforward about producing activity. The thing to hold onto is that producing the input to a sort is not the same as producing interest, and any description that blurs those two is describing a capability nobody has.

What to do with the distinction

1. Report two numbers, never one. Turnover next to distinct signers, over the same window, every time.
2. Name the case. Aligned, silent volume, loud stillness or empty. Naming it forces a claim you can be wrong about.
3. Count what you own. Joins, replies, returns. These are people-shaped and you do not need permission to gather them.
4. Keep the caveat attached. Write the blind spot in the same line as the number, so it cannot be separated later.
5. Ask for methodology before results. Any figure without a window, a venue list and a deduplication rule is a headline.
6. Revisit in a week. Silent volume decays when a budget ends; loud stillness sometimes turns into something. Time separates them.

The methodology point deserves emphasis because it is the only part of a vendor relationship a buyer can actually verify. Asking how volume campaigns are measured gets you a description you can check against the chain yourself. Asking what results to expect gets you a number that cannot be checked against anything, before or after.

The honest ceiling of measurement

There is a limit past which nothing here improves, and it is worth stating rather than implying. Attention is a decision made by people, in private, using information that mostly never touches a public record. The best available instrumentation is a handful of biased proxies and the transaction history, read carefully, from sources such as a public explorer like Solscan.

That ceiling is not a reason to stop measuring. It is a reason to stop promising. A team that knows it is working with proxies will describe its position in ranges and shapes, will notice when two indicators disagree, and will not spend a budget on the assumption that a number went up because people cared. A team that has confused the two will do the opposite, confidently, until the divergence becomes impossible to ignore.

Questions readers send in