Revenue per transaction is the L2 metric that survives incentives

Base charged $0.001 to $0.002 per transfer and Arbitrum One $0.004 to $0.005 in Ethereum.org’s L2 directory snapshot of 1 to 6 October 2026. That is a threefold gap in the price of the same action. Total revenue can be bought with a campaign. Revenue per transaction cannot, because the campaign has to pay the price again on every unit.

Why totals are easy to move

Incentive programmes pay people to transact. Every subsidised transaction still prints a fee line, and in most designs the fee is charged and then rebated, or the user is paid in a token the venue issues.

A campaign therefore raises the total while it lowers the average. Revenue per transaction is a division, and the denominator grows faster than the numerator when a subsidy is doing the work. A venue reporting a record revenue month may be reporting a fee policy that no user would choose without the rebate.

This is not fraud. It is the difference between a level and a rate. Level metrics respond to spending, and rate metrics respond to pricing.

The same revenue, different user counts

Take one million dollars of revenue. On Base at $0.0015 per transfer that is about 667 million transfers. On Arbitrum One at $0.0045 it is about 222 million.

Same money, three times the user count. Reverse the calculation and it becomes a screen. If two venues report the same revenue and one charges a third of the fee, the cheaper one either moved three times as many transactions or earned its revenue somewhere other than transfers.

The published ranges widen the spread. At Base’s $0.002 ceiling, a million dollars is 500 million transfers. At Arbitrum’s $0.005 ceiling it is 200 million.

That is the metric’s job. It converts a revenue headline into a count, and a count can be checked against blocks and against the venue’s own fee page.

The fee is policy, not cost

OP Mainnet displayed $0.00 in the same snapshot. Zero is a price, not a cost. The data cost of a rollup is decoupled from mainnet gas by blobs, so the marginal cost of a transfer is small and the posted fee is a decision about market share.

The directory is therefore a map of pricing strategies rather than a cost table. Scroll sits at $0.002 to $0.003, Starknet at $0.005 to $0.008 paid in STRK, and Linea at about $0.024. Linea’s floor is roughly 12 to 24 times Base’s, for the same kind of transfer in the same week.

Total value locked sees none of this. Base held $14.6 billion and Arbitrum One $12.3 billion. Both are stocks, and a stock barely moves when the price of a unit falls by a tenth of a cent. That is the argument in L2 fee revenue, not TVL, and the per-transaction figure is its sharpest form. The same logic explains why a sequencer that can reorder or delay a transaction is a different product at the same price.

What the number cannot tell you

The directory quotes a simple transfer. A swap or a bridge burns more gas and carries a different fee, so per-transaction revenue computed from a simple-transfer price is a floor for the blended network figure, not a measurement of it.

The figures also sit near the reporting floor. $0.001 and $0.005 differ by five times, and both round to fractions of a cent in a headline. Precision at that scale needs the raw fee, not the rounded quote.

And none of it describes duration. Base and OP Mainnet both take seven days to withdraw to mainnet. Arbitrum’s dispute window is 45,818 blocks, about 6.4 days. ZKsync Era advertises a minimum of three hours. A cheap transaction on a seven-day rail is cheap in the wrong dimension when the position has to move.

Verdict

Revenue per transaction survives an incentive campaign because it is the one number the campaign cannot pay down without paying it again on every unit. It also inverts the usual ranking. On the 1 to 6 October 2026 snapshot, a million dollars of revenue on Base implies 500 million to 1 billion transfers and on Arbitrum One 200 to 250 million.

That gap is not a quality judgment. It is a statement about fee policy, and fee policy is what a venue controls when it decides whether to rent activity or earn it. Ask for transfers alongside revenue, divide one by the other, and compare the answer with the fee on the page. If the campaign is doing the work, the rate will fall while the level rises, and only one of those two numbers is being paid for by users.