What Ranking Signals Reward, And What They Hide

A ranked list rewards exactly what its sort field measures and nothing else. Turnover rewards value moved, ignoring how many people moved it. Percentage change rewards small denominators, which means small pairs. Transaction count rewards frequency, which is cheap. Recency rewards nothing but being new. Every failure mode people find mysterious in these lists follows directly from that arithmetic.

A sort is an argument

When a product chooses to order a list by turnover, it is asserting that value moved is the most useful single fact about a pair for the person reading. That is a defensible position and it is a position, not a neutral act. A different product ordering the same pairs by transaction count is asserting something else, and both lists are honest descriptions of the same market.

The reason this framing matters is that it removes the temptation to look for intent behind rankings. A token appearing high on one list and low on another has not been favoured or suppressed. It has been measured twice by two different rulers. Most of the folklore about lists being rigged dissolves once the sorts are read as arguments rather than as verdicts.

It also sets the boundary of what can be known. The sort field is often published. The full ordering logic, including tie-breaking, exclusions and any review layer, usually is not. This note describes the arithmetic of the fields, which is knowable, and refuses to guess at the rest.

The signals, field by field

Turnover, and what it cannot see

Turnover is the most robust of the common signals because it is denominated in value. You cannot inflate it by splitting one trade into ten, since the ten legs sum to the same total. To raise turnover you have to move value through a pool, which is a real commitment of capital and fees rather than a formatting trick.

Its blindness is structural. Turnover counts value, not participants. A pair where four hundred people each traded a small amount and a pair where four wallets cycled capital repeatedly can report the same figure and occupy the same row. The number is correct in both cases; it simply does not contain the distinction that a reader usually wants.

This is the gap that produced activity operates in, and it is worth being plain about the mechanism. An automated Solana volume bot raises the quantity a turnover sort ranks, which is a mechanical effect on a mechanical list. It has no effect on the separate question of whether anybody reading the resulting row decides to look further, and no honest description of the category claims otherwise.

The reader's defence is the second number. Distinct signer count over the same window, read from the transaction record, separates the two cases that turnover merges. It is not proof, because coordinated wallets are indistinguishable from independent ones on chain, but it converts a single ambiguous figure into two figures that can disagree.

Percentage change is a size filter

A percentage-change column looks like a performance ranking and behaves like a size filter. The arithmetic is unavoidable: change is a ratio, ratios grow as denominators shrink, and the shallowest pairs have the smallest denominators. A change-sorted list will therefore be dominated by the least liquid pairs on the venue almost all of the time.

This is why those lists feel unreliable to people who trade from them. The list is not promising that the top row is a good trade. It is reporting, accurately, that the top row moved the largest percentage, which in a thin pair can be caused by an order that would be invisible in a deep one. The list did its job; the reader inferred something the list never claimed.

The window compounds it. A one-minute change sort is almost pure noise measurement, because in one minute the only pairs that can post large percentages are ones where a single order dominates. A twenty-four-hour change sort behaves far more sensibly, not because it is smarter but because a full day of trading is harder for a single order to define.

Transaction counts and the cheapness problem

Count-based signals reward frequency. Each additional swap raises the count by one regardless of whether it moved meaningful value, and on a chain with low fees the marginal cost of an additional trivial swap is small. That combination makes count the easiest common signal to fill with activity that has no economic content.

Some surfaces mitigate this by applying a minimum trade size before a swap counts, or by counting unique signers instead of transactions. Both are improvements and both are usually undocumented in detail, so a reader cannot tell which mitigation a given list applies. Where the surface does not say, the honest read is that a count column may be measuring frequency alone.

Holder-delta columns share the same weakness in a different costume. An address is not a person, creating one costs almost nothing, and a single participant can hold the same token from a dozen addresses without doing anything unusual. A rising holder count is consistent with distribution widening and equally consistent with one wallet fanning out, and nothing visible on chain separates the two cases reliably.

There is a second, subtler problem. Aggregated routing means one user action can land as several on-chain swaps across different pools. A count that does not deduplicate routed legs will overstate activity for tokens that route through several venues, which quietly favours certain pairs for reasons that have nothing to do with interest.

Recency lists and the conveyor belt

A list ordered by pair age is a conveyor belt. Everything arrives at the top, moves down at a constant rate and leaves. Nothing a token does changes its position, because the sort field is a clock. Appearing on such a list is automatic and carries no information whatsoever.

What a recency list does change is who is watching, and that is a real effect worth separating from the ranking itself. The people reading a stream of brand-new pairs are a specific and unusual audience: fast, sceptical, and expecting most of what passes to be worthless. A token gets one moment in front of them and cannot earn a second by improving, because the sort has already moved on and will never return the pair to the top.

Recency lists are still useful to the people who read them, because they are the only way to see something in its first minutes. But teams routinely misread presence there as a result. It is not a result; it is the passage of time, and every token that exists appears there for exactly as long as every other token does.

Composite scores and the honesty gap

A composite blends several signals with weights the platform chose. In practice composites behave better than single fields, because each component's failure is diluted by the others. A pair that tops a change sort through pure thinness will usually not top a composite that also considers turnover and participant counts.

The cost is inspectability. A single-field sort can be checked: you can compute turnover yourself and see whether the ordering matches. A composite with unpublished weights cannot be checked at all, and a reader is left trusting that the platform's opinion is reasonable. That is often fine, and it should be understood as trusting an opinion rather than reading a measurement.

The desk's rule here is simple. If a surface publishes its weights, quote them. If it does not, describe the list as a composite of undisclosed weighting and stop, because the alternative is guessing at numbers and presenting the guess as knowledge, which is how most published ranking guides go wrong.

A worked league of five pairs

Round invented figures, describing no real pairs. Five pairs over the same one-hour window, ranked by four different signals.

The lesson is not that any list is wrong. It is that a team asking why they are not trending is usually asking about one list while measuring themselves against the properties another one rewards. Pair D would look like a failure to anyone watching a change column and like the healthiest pair on the venue to anyone counting signers.

How to read any list you are handed

1. Find the sort field. If the surface does not name it, treat every conclusion you draw as provisional and say so out loud.
2. Find the window. The same field over five minutes and over a day produces two lists that share nothing but a title.
3. Ask what the arithmetic favours. Ratios favour small denominators. Counts favour cheap actions. Value favours capital.
4. Look for the missing column. Whatever the sort cannot see is where the interesting cases hide, and it is usually participant count.
5. Check a second surface. Two independent indexes agreeing is weak evidence. Two disagreeing is a fact about the indexes worth knowing.
6. Write down what you concluded and why, so that in a week you can tell the difference between what you saw and what you assumed.

Venue structure matters here too. After a token migrates from a launch curve to a pool, its activity is ordered by the same sorts but produced by a different pricing engine and split across whichever pools exist, which is why teams thinking about post-migration presence look at a Raydium volume bot as a separate question from anything curve-stage. The list does not care where the flow came from; the reader should.

The boundary of this analysis

Everything above describes arithmetic that anyone can verify by computing the same fields from public data. It stops at the point where platforms stop publishing. No weights are asserted here, no thresholds are quoted for any specific product, and no claim is made about review layers or exclusion rules, because those are not documented in enough detail to describe honestly.

It also makes no promise. Understanding what a sort rewards tells you how a list behaves. It does not tell you how to get on one, and the step from a placement to a person deciding to care is not a ranking problem at all. Public sources on how the chain itself works, including the developer documentation on solana.com, describe the data these lists are built from far better than any secondhand summary of a ranking.

The most useful outcome of reading this is a smaller vocabulary. Not we are trending, but we are eleventh on a one-hour turnover sort on one index. The second sentence is checkable, survives a week, and cannot be turned into a claim about the token by anyone who reads it later.

Questions readers send in