We measured the entire x402 economy. It is smaller than one developer salary.

"Each" is the average monthly revenue per listing in that category. Two categories carry the protocol: social data and web search. Together they are 12 percent of the listings and 61 percent of the calls.

Average monthly revenue per listing, by category, ranked. Email and messaging pays about fourteen times what media generation does per listing, and it is one of the smallest categories on the protocol. Nothing here clears four dollars.

The axis on that chart runs $9,000 to $10,700 and does not start at zero, so the slope is magnified; that is the only way to see a 5 percent band at all, and you should read it knowing that. What did not move across any of the sixteen readings: the median listing still earns three cents a month, about 93 percent still earn under a dollar, and the ten busiest listings still take roughly two thirds of every paid call.

Two things did move, and the seventh reading is what made them reportable. The count of listings clearing $100 a month sat at exactly 12 for three readings, then 11, then 8 three times running, then 10 five times, and now 11 for four readings running. Separately, paid calls reached their highest recorded level at that reading, 345,274 against 339,178 at the 24 July baseline, while estimated volume sat about 5.8 percent lower. More requests, less money: that is price compression rather than demand drying up.

Volume is above where this series started, for the fifth reading running - but the climb stopped at reading 15 and has not resumed. Reading 8 was the first to move up, $9,392 against $9,178, and it did so on flat calls. This page published that as data and explicitly not as a reversal, and said the next reading would decide. It then rose seven readings running - $9,392, $9,411, $9,577, $9,704, $9,989, $10,215 and $10,386 - before falling to $10,322 at reading 15 and falling again to $10,309 at reading 16. That is the first back-to-back fall since the slide into the reading-6 trough, though the second one is small: 0.1 percent, against 0.6 percent for the first. At $10,309 it is still 5.8 percent above the $9,747 baseline, against 5.9 percent below it at the trough, so the decline this page spent seven readings describing is still fully given back. What has stopped is the climb, not the recovery. Paid calls set a record at every one of the last nine readings including this one, ending at 370,464, so both falls came with more traffic rather than less. Distinct seller hosts read 1,484, a third consecutive fall from 1,534 two readings ago, against 1,445 at reading 4. Reading 8's flat-call claim did not survive; the growth came with traffic, not instead of it.

We published a test in advance. It failed at reading 15, we kept the bar where it was, and reading 16 failed it again. The measure is revenue per paid call, which is just volume divided by calls and is the thing compression actually predicts. Labelled by reading, because reading 2 has no call count: $0.028737, not recorded, $0.02823, $0.02738, $0.02679, $0.02679, $0.02658, $0.02720, $0.02674, $0.02676, $0.02692, $0.02754, $0.028075, $0.028367, $0.028072 and $0.027827. Compression is real as a description of readings 1 to 7. It then sat flat to within half a percent across readings 9 to 11 and rose at readings 11, 12, 13 and 14, and on the strength of that this page argued compression was unwinding and stated what would refute it: the next reading had to come in above $0.028367. Reading 15 came in at $0.028072 and failed. Reading 16 was asked to clear the same $0.028367, not an easier number, and came in at $0.027827: $0.000540 under the bar, and $0.000245 under reading 15 as well, so this is a second failure and a second step down rather than a repeat of the first. The average paid call is now 4.7 percent above the reading-7 trough and 3.2 percent below the 24 July baseline, having been 2.3 percent below it one reading ago and 1.3 percent below it two readings ago. What makes this worth reporting rather than shrugging off is the direction of the two inputs, twice running: paid calls set another record while volume fell again, which is the combination compression predicts and the opposite of what unwinding predicted.

The test for reading 17 stays at above $0.028367, and that is deliberate. The convention here is that each reading has to beat the one before it, which would put the new bar at reading 16's own $0.027827. We are not doing that. Reading 16 just failed $0.028367, and adopting the number it failed at would be lowering the bar to meet the result, for the second reading in a row. So the bar has now been held across two consecutive failures. It moves when a reading clears it, not when a reading misses it, because a test that follows the data is not a test.

The symmetric caution applies. Two failed tests are stronger evidence than one, but they still do not show compression has resumed, any more than four rises showed it had ended. The last three readings, $0.028367, $0.028072 and $0.027827, span 1.9 percent between them, and the rise being given back was only about 3 percent to begin with. The most defensible description today is a measure drifting back down over two readings after a four-reading rise. What was wrong was the confidence, not the arithmetic.

Why the seventh reading and not the fifth. Readings 5 and 6 both showed this, and we did not publish them, because they were taken 3.5 hours apart on a rolling 30-day window. Almost none of that window turns over in 3.5 hours, so they were largely the same measurement twice, and treating two agreeing values as corroboration would have been the same class of mistake that caused our npm retraction. Reading 7 was deliberately spaced 12 hours after reading 6. It agreed, so all three are published together here.

And one correction to the earlier reading of this chart. An earlier version of this section noted volume sliding about 4.8 percent and declined to call it a trend. That caution was right twice over: the slide stopped at $9,173, $9,169 and $9,178, and it then turned back up through $9,392, $9,411, $9,577, $9,704, $9,989, $10,215 and $10,386 - past where it began - before easing to $10,322 and $10,309. Anyone extrapolating the early slope would have been wrong, and anyone calling a floor from those three flat readings would also have been wrong, in the other direction. Worth stating plainly: this is a rolling 30-day window, so the early decline is still inside it and its exit alone lifts the figure. The counted listing total also still wobbles, by about a percent between most pulls, by 3.4 percent between readings 9 and 10, by 4.6 percent between readings 14 and 15 (15,376 to 14,669) and by 1.1 percent between readings 15 and 16 (to 14,502), for reasons on the API side. The call and volume totals do not move with it, which is why it is read as a filter artefact and not as supply moving in either direction.

CC BY 4.0. Use it, quote it, argue with it. The file carries the method, the field definitions and an explicit list of what we do not know.