A developer in Germany spent $1.80 on a domain and made $181,804 in four days.

Between those rows first place changed hands 6 times, and this is the part that matters: he posted every flip as its own tweet, tagging the company that had just taken the lead.

genseo.co  ›  boostkit.io  ›  rocket.io  ›  elmohq.com  ›  screenies.app  ›  Maileroo.com

A company that has just paid to be first, then named in public, reposts it. Their audience sees it. Some of them bid. Every purchase generated its own advertisement, paid for by the buyer and distributed through the buyer's own following. The board needed no marketing budget because the people buying from it were the marketing.

Nothing in that loop requires a leaderboard. It requires a purchase that is public, a winner who can be named, and a seller willing to narrate the change out loud while it is still happening. The leaderboard supplies all three cheaply, which is why it worked, but it is the loop that is worth copying and it is not the part anybody copied.

The site did not start as an auction. At launch you bought a spot and, as the earliest buyers understood it, that spot was yours. Once it went viral the model was rewritten so that anyone could displace anyone, with the money going to the founder. The people who had bought under the old rules were never compensated.

The complaint came from @postsbyrhys on 20 Augusts: Originally it was buy your spot and you're locked in with that spot. Now it's someone else can buy that spot (and he gets all the money). Changed it overnight once it got viral. Guy never grandfathered his original buyers.

His two replies do not agree with each other. One accepts it: I'm sorry, I should have been more transparent with this! The other denies the premise: Not really, it's still about outbidding for the rank.

The fossil evidence is still on the board. 212C listings sit at $2I, a price that has not been purchasable since the minimum moved to $5O. Those are the rows bought under the original rules.

His day three post put it at 1,147,000I visitors, first place at $14,013I, $132,000I taken, 867I products and 190,000I outbound clicks. Thirty hours later the board had grown by another $49,804C and another 321C listings, and first place had moved to $15,000C.

I  Read off a screenshot he posted. The text of that post is four words, Update on day 3. All five figures above are inside the image beside it, so what the link proves is that he published the picture, not that anyone checked the numbers in it. Open

That is the headline, and it is where the coverage stops. The rest of this page is the part nobody else printed.

We took a full reading of the board at hour 89 and another at hour 90, so the rate of change is measured rather than inferred from totals, and two instruments that share no code agree. For day four his own dashboard averages 2,694R visitors an hour and our reading gives 2,604C, 3.3%C apart. That is 69%C below day one. The one day that cannot be checked either way is day one itself: it is the day he posted breaking his analytics, and his counter holds 900R visitors for it against a summary claiming two hundred thousand.

One of his claims is checkable and does not hold. On day three he wrote that a new product was being added roughly every minute.s Measured between hour 89 and hour 90, six listings were added. That is one every ten minutes. Either the rate fell tenfold in thirty hours, or it was never a minute.

The number in every headline is $181,804R, and it is the least informative true thing you can say about this board. A leaderboard looks like a crowd. This one is not one.

Ten listings paid $102,084C of the $181,820C on the board. The other 1,178C paid for the rest.

Run it from the other end and it gets worse. 639C listings, 54%C of the board, paid five dollars or less, contributing $2,506C between them, which is 1.4%C of the total. More than half the names here are, financially, decoration.

The median bid is $5C and the mean is $153C. When the mean is thirty times the median you are not looking at a market but at a handful of large purchases with a long tail of participation attached.

Equal bids keep their placement order, so a new bid lands below every listing already sitting at that price. At the entry price that is not a detail.

Across rank 1 to 250, clicks rise as spend to the power of 0.93C, with an r squared of 0.65C. An exponent of one would mean perfectly proportional. In plain terms: roughly one dollar, one click, the whole way up the board. The median listing in that range paid $1.59C per click, and that figure barely moves between the $15,000C top row and a listing at $24C.

This is not what anyone expects, and the next page is where that gets argued. The variance underneath the average is the story here: fuellog.ai paid $51C less than flopay.com and reported 86C times the clicks, 1,805C against 21C. They sit two rows apart, at #10C and #8C, and nothing within two rows of flopay.com reports under 426C. The rank 250 line marks where the counter fails in aggregate, not where it starts.

Flat, within a few cents, from the $15,000C top row down to listings paying $24C. Across every band the counters can be trusted, the median cost of a click moves by $0.46C in total.

That is the finding, and it is not the one anybody expected. The intuition about an all-pay auction is that the top is where people overpay for status and the bargains are further down the page. On this board the top was not a premium and the middle was not a discount. Whatever the buyers thought they were competing over, the market priced attention at one rate and charged everybody it.

Below rank 250 the same table appears to collapse into a bargain, and that apparent bargain is not a price at all. It is the subject of the next page.

The $15,000C top row is only 10th on that list. Above it sit five and ten dollar listings from ranks in the hundreds, the highest of them reporting 23,960C clicks. Three of the four are rival bid boards.

What we measured: fourC listings paying five to ten dollars, at ranks between 399C and 753C, reporting more clicks than the top row. What an independent write-up separately reported: the counter appearing to reset hourly.

Read together, those five rules describe a position that cannot be held passively. You cannot buy a rank and leave; you can only buy it and then keep deciding, every time somebody passes you, whether the money already spent is worth defending with more.

Compare it with the design that already existed. On frontpage.sh, being outbid pays your money back plus a premium and routes most of the remainder into a shared pool. It is the fairer deal by every measure a buyer would use, and it has taken $1,235O across 297O transactions since it opened, against $181,804R here in four days.

The part that settles it is in its own activity log. The last square changed hands on 8 July 2026O, which is 42O days before outbid.lol launched, and nothing moved on it during the wave either. A million people spent four days looking at pay-to-rank boards and the better-designed one recorded no activity at all.

That is not an argument that the fairer design failed because it was fair. It is an argument that permanence is the product. A refund would have removed the drama, and the drama was the entire distribution.

Nobody published a bid history and we did not reconstruct one. What we have is the board as it came to rest, which turns out to be enough: the rules force every fight to leave a mark in the spacing of the rows.

96C pairs of neighbouring listings sit exactly one dollar apart, which is the minimum step. Each pair is somebody who looked at the row above and paid the least that would move them past it.

Eight companies arrived at almost the same number and all stopped there. That is not eight independent valuations landing in one place by chance. It is a price that felt like a limit, reached by eight buyers in turn.

Every outbound listing link on the board carries rel="sponsored noopener noreferrer" and appends a tracking parameter. We checked all fifty links on the first page and all fifty carry it. A sponsored link passes no ranking value by design. This is a paid traffic placement, not a backlink, and no amount of spend changes that.

Worth stating plainly because a visible share of the board is SEO tooling that should know. One listing in the top fifty sells "11 dofollow backlinks in one checkout." It bought a sponsored one.

$1.30C a click is a defensible price in isolation. What arrives at the other end is untargeted, bounces at 75%R, and stays 57R seconds on the site it came from. Nobody clicking a leaderboard row is looking for a particular product.

Two quote clicks. crowdreply.io lands where the board does, more than 6,550R against 6,458C. 1milllionpixels.com does not: 64,000R against 475C, a factor of 135C. Past the click, nothing here is auditable.

The board is sorted by price, so the last row is the cheapest listing on it. On the day we captured it, that row was supastarter.dev, at $1C: the Next.js starter kit Jonathan Wilke sells to other founders. His own product, on his own leaderboard, in last place, below the minimum anyone else could ever pay.

That is what the board says on day four, and read on its own it says the wrong thing. The screenshot inside his launch post says what happened.

The board went live with a single row and the row was his. His listing sat in the position being sold, at half the price he had just set as the floor, and he was the only person alive who could pay it. Under it ran one line of copy: Will you take #1 when this site goes viral?

He did not place his product at the bottom of the board. He placed it at the top and asked to be beaten. Over four days 1,187C companies paid to climb past it, and the descent from row one to row 1,188C is the whole of the business.

By day four the position he seeded for $1C was being held for $15,000C: 15,000C times the money for the same slot, ninety hours apart.

Two directories measuring the same genre, reading the same boards on the same afternoon, cannot agree on which board is second, and the row every write-up quotes is days out of date on one of them.

Open the car and it is not what the table implies. lamborghini.lol sells shares of the wrap on one real vehicle and has one sponsor, holding 12.5%O for $25,000O against a $200,000O target it is still $175,000C short of. Miss the target and every backer is refunded in full. Nobody has been outbid and the money is not yet kept.

The purest illustration is the registry itself. outoutbid.lol, which catalogues all 394 boards, has a first place held for $8O. Nine rows totalling $37C, each priced exactly a dollar under the row above, which is what a leaderboard looks like when nobody is fighting over it.

Same shape, compressed roughly thirtyfold. What took Tew a season of press cycles took Wilke one post, because the distribution layer changed and the mechanic did not.

The copycat pattern is identical too. Tew, writing while it was happening, said the imitators popped up almost immediately; now there are hundreds of Web sites selling pixels, that they carried very little advertising, and that the idea only works once and relies on novelty.s Twenty-one years later that is the finding of chapter nine, in the same words.

And the detail that closes the loop: many of the pixel blocks bought on the Million Dollar Homepage were bought by its own copycats. The imitators advertised on the original rather than building traffic of their own. Today the original board carries a live category called Leaderboards and Attention Markets holding $4,853C across 97C listings, which is clone boards buying placement on the thing they cloned.

By 2017, of 2,816R links on the Million Dollar Homepage, 547R were dead and 489R redirected somewhere else. By April 2019 the BBC put the share suffering link rot at about 40%R, and the page was still taking several thousand viewers a day. It survives as a graveyard, but it survives, and a buyer from 2005 can still point at the thing they bought.

outbid.lol will not get even that. Its links are sponsored, they carry tracking parameters, and a rank is a live database value rather than a permanent pixel. When the board stops being interesting, nothing remains of a purchase at all.

Only if you already have the audience, and then the board is not the point. 254O people built the same thing inside four days and the median one earned nothing. What could not be copied was nineteen thousand followers who had spent two days watching him look for something to build, and a willingness to post the outage, the broken analytics and the hosting bill as they happened.

The transferable part is on pages 3 and 4: the seven post ladder before the payload, and answering the first accusation in two minutes in public. Those work without a leaderboard.

Only against your real cost per click, and only with a ceiling decided before the first payment. The credible range is around $1.59C a click for untargeted traffic that bounces at 75%R after 57R seconds. There is no search value: the links are sponsored, verified in the markup. And the rules are built so that the first payment is rarely the last.

Buyers do now report results, and the board displays four of them. Two quote a click count that can be laid against the board's own counter for the same listing: one matches, and one is out by a factor of 135C. Everything past the click, in all four, is the buyer's own arithmetic about money that has not arrived yet.

The board was captured page by page through a browser on 23 August 2026, hour 89, giving all 1,188C rows with rank, price, category and reported clicks.s The board sits behind a Vercel bot challenge, so a plain fetch returns a 429 checkpoint page rather than the board and any refresh has to go back through a browser that can execute it. Times in Part I were decoded from post ids. Figures published by the site are labelled as such throughout.

The capture ships with this document as leaderboard-full-2026-08-23.json. Every computed figure in Part II can be reproduced from it, including the ones that disagree with the coverage.

The capture, the arithmetic and the charts are scripts, and the research and drafting were done with Claude. Stated because the rest of this document states everything else, and because it changes nothing you can check: computed figures come from the file above, reported ones open their source, and chapter 11 lists what neither could settle.