The whole board captured again at hour 164, the same way and against the same file as the day-four capture the case study is built on. 1,775 rows against 1,188.
- $27,310 that was not on the board three days ago is now inside the top thirteen, and one of it came within $1,000 of the top row.
- Six of the day-four top ten have not paid another cent and are still up there. The new money took the seats beside them, not the ones under them.
- A listing's own cost per click falls while it sits still. The board's median went the other way, $1.59 to $1.94.
- Visitors are down 86% on their run rate and the money is coming in 72% slower. Both fell; traffic fell harder.
- The top ten took in $4,687 more, not less. Their share dropped only because everything under them grew faster.
Both captures ship as JSON: day four, 1,188 rows and day seven, 1,775 rows.
Ten days ago a developer in Germany put a leaderboard online where your rank is whatever you paid. I captured the whole board on day four, all 1,188 rows, and wrote up where the money came from. The last line of that document was a promise: this board gets read again on day 7, 14, 30 and 90, same method, same file.
This is the day-seven reading, taken at hour 164 on 26 August. The board is at 1,775 rows and $224,473 in live bids. Everything below is the difference between two files.
The top of the board was contested, and almost nobody had to fight for it
I expected to find the top ten defending itself, because that is what the mechanic is for. There is no refund and no expiry, so money spent on a rank you no longer hold is gone, and the cheapest way to protect it is to spend more.
What actually happened is stranger. $27,310 arrived at the top of the board in three days, and it mostly did not displace anyone.
| Arrived in the top thirteen | Rank | Bid |
|---|---|---|
| tutti.so | #2 | $16,000 |
| zerorank.ai | #8 | $4,000 |
| pecan.ai | #9 | $3,750 |
| modulate.ai | #11 | $3,560 |
tutti.so came within $1,000 of the top row. That is what see.io's extra $2,000 was for, and it is the only defence anyone mounted all week. Of the other nine incumbents, six are sitting at exactly the bid they held three days ago and are still inside the top thirteen. Two moved by $10 and $5, which is a rounding error rather than a fight. One, requesty.ai, is out of the top thirteen entirely.
A board built to make you keep deciding whether to defend your money charged $27,310 for seats that mostly nobody had to be pushed out of.
A click gets cheaper for you and more expensive for everyone else
This is the finding that needs two readings to exist, and I had it backwards in my first pass at this piece.
| Listing | Cost per click, hour 89 | Hour 164 |
|---|---|---|
| see.io | $1.30 | $0.48 |
| joni.ai | $0.95 | $0.72 |
| outrank.so | $0.98 | $0.78 |
| orynth.dev | $0.98 | $0.88 |
| crowdreply.io | $1.97 | $1.66 |
| trycomp.ai | $0.79 | $0.73 |
| fuellog.ai | $1.94 | $1.21 |
| context.dev | $8.33 | $2.25 |
Every survivor fell. see.io's fell 63% while its bid rose, because its clicks more than tripled and its bid only went up 13%.
But the board's median went the other way. Across the top 250, where the click counter still behaves like a measurement, the median cost of a click went from $1.59 to $1.94, up 22%.
Both things are true and they are not in tension. A listing that sits still accumulates clicks against a fixed number, so its own cost per click decays. The board's median rises anyway, because rows like the four in the table above keep refilling the expensive end of the range faster than the sitting listings can drag it down.
Which means the $1.59 in my own write-up was not a rate. It was a photograph of one afternoon. If you are pricing a listing against your own acquisition cost, the number to compare against is the one on the day you buy, not the one in anybody's case study.
Both curves fell, and traffic fell harder
Against the run rate to hour 89, visitors are down 86% and the sum of live bids is growing 72% slower, $569 an hour against $2,043.
I want to be careful here, because the easy version of this sentence is wrong. The money did follow the traffic down. It just did not fall as far, and the gap between the two falls is the whole story: attention left faster than spending did.
The board still added 587 rows in those 75 hours, one every eight minutes. And the top ten's money went up by $4,687. Their share of the board fell from 56.1% to 47.6%, but only because everything underneath them grew faster. Nobody at the top withdrew.
The copies, and one correction to my own document
The clone registry counted 254 boards on day four. It counts 414 now.
Against the nine boards that registry publishes totals for, the original holds 84.4% of the money. My case study put the figure at 78%, but that was measured against a whole-genre total, and these are not the same denominator. I am not going to claim the share went up on the strength of two numbers that were built differently. It is a figure to carry forward and measure the same way next time.
One correction I owe the document. It printed flopay.com at rank 8 with 21 clicks and said, in the chapter listing what could not be verified, that a two-digit count in the top ten between neighbours reporting thousands cannot be right. It now reads 1,011. The caveat was correct and the row is at least in a believable range now, though nothing in either file proves the counter was fixed rather than differently wrong.
What I will measure on day 14
Whether the seats keep filling without anyone being pushed out of them. If they do, the interesting question stops being how high the wave went and becomes why a market that sells position does not have to charge anyone to keep it.
The full capture ships above. Day 14 is 2 September.
