It is a blob of machine text, a few dozen lines, and it flashes past faster than a hummingbird's wingbeat. There are somewhere north of fifty million of them per second crossing the open internet at any given moment, which means that in the time it takes to read this sentence, more auctions have concluded than there are people in Ohio.

Nobody reads them. They are not written to be read. They are written to be processed and discarded, and their entire value as evidence comes from the fact that no one ever expects one to be held up to the light. A bid request is the industry's junk mail, its shredder output, its background radiation. Which is exactly why it is the most honest document ad tech produces. Nobody lies in a place they assume nobody looks.

In August 2025, somebody held one up to the light.

The line item showed an SSP taking twenty dollars out of a thirty dollar CPM. Sixty percent. One hop. One impression. The kind of number that in any other industry produces a subpoena, a resignation, and a very quiet Friday press release, and which in this one produced absolutely nothing at all.

And here is the useful thing about this business: it forgets, but the record doesn't.

Ad tech operates on the working assumption of a two-week memory. A number gets said on a stage. A take rate gets quoted in a funding announcement. A commitment gets posted to LinkedIn with a line break after every four words, the way people write when they want to sound solemn and have nothing solemn to say. Eighteen months later the thing is provably false, and the explanation arrives, and it is always the same explanation. It was aspirational. It reflected where the market was at the time. It has been taken out of context, a phrase which in the entire recorded history of advertising has never once been followed by the actual context.

The deck is archived. The panel is on video. The blog post deleted in March is not deleted, it is merely moved. The earnings call that contradicts what publishers were told is a transcript with a date on it, sitting on an investor relations page, indexed, free, waiting.

None of that is a threat. It's a filing system, and it runs in both directions. Say true things and it becomes an asset that will sit there in three years proving somebody was early. The archive has no opinion. It is the only genuinely neutral party in this entire market.

Now. Back to the twenty dollars.

Mike O'Sullivan posted two things to LinkedIn about a fortnight apart that August.

He founded Sincera, the company that measures the quality of publisher supply and grades what a bid request is actually carrying.

The Trade Desk bought it, and O'Sullivan now runs that work from inside the buyer, reporting directly to chief executive Jeff Green, while opening a slice of the data to the industry as “OpenSincera”

In other words: the supply-quality auditor for the largest independent DSP on the open internet.

Hold onto that, it matters later.

Post one was the bid request. Twenty out of thirty. Sixty percent, in public, with a screenshot.

Post two was a question about transaction IDs, a plumbing identifier whose entire purpose is to let a buyer notice that it is being offered the same ad slot eleven times by eleven different exchanges. O'Sullivan asked what the principled arguments against them were. He said he was struggling to come up with any. It is arguably the most polite sentence ever entered into that website. There is no accusation in it. There is barely a verb in it.

One of those posts broke containment. It was not the one about the money.

The transaction ID post went thermonuclear. Comment threads. Group chats. A podcast debate. An emergency statement from the IAB Tech Lab declaring the resulting code change a violation of the OpenRTB spec. An actual commit pushed into Prebid at the end of August that broke the identifier across exchanges. Then two months of everyone demanding to know who authorized it, which nobody satisfyingly answered. Then a clarification. Trade coverage throughout. The industry's full immune response, deployed at maximum volume, over a de-duplication key.

The sixty percent take rate got nothing. O'Sullivan said as much himself, on the podcast where all of this surfaced: he put that bid request out, and nothing was discussed about it.

Nothing was discussed about it.

Twenty dollars out of thirty, and the market's considered response was to change the subject to auction hygiene and stay there for eight weeks. This is the behavior of a man who opens a catastrophic bank statement and responds by reorganizing the cutlery drawer. It is technically productive. The forks are now sorted. He is still broke.

Several installments of this column have now put the middle of the supply chain at roughly half of every dollar. A former PubMatic executive read that math and wrote in to say it was generous. And the question underneath the entire exercise has been wrong from the beginning. The question has been: why does nobody catch this?

Everybody catches it. Nobody wants to be the one holding the ruler.

There is a version of this story in which the publisher is the victim.

Fee stack takes half. Publisher survives on the remainder. Publisher grows desperate. Desperation produces auto-refresh, twelve slots above the fold, a cookie banner with the ergonomics of a hostage negotiation, and a video player that follows the reader down the page like a Victorian ghost. Sad. Sympathetic. Enormously flattering to whoever is telling it, because it casts the writer as the only person in the building who noticed.

That version does not survive ten minutes with the people who actually operate the machine.

Chris Kane runs Jounce Media, which exists to trace where money goes inside supply paths. This means he has read more bid requests than is medically advisable and carries the affect of a man who long ago made his peace with what he found.

On The ADOTAT Show, he described the sell-side incentive without blinking:

"The financial incentive for media companies on the sell side of the open internet isn't to create the very best supply. It's to create the maximum volume of supply that is deemed to be sufficiently good."

Sit with the construction.

Not good.

Deemed to be sufficiently good.

Somebody else is doing the deeming.

The publisher is producing to a specification it did not write, the way a factory produces to a tolerance, and the tolerance was set by the person buying. Then he finished the thought:

"The quality of open internet supply has sort of degraded down to the lowest quality that buyers will accept."

Degraded down to. Not fallen to.

Degradation implies a process with a direction and a floor, and the floor is not a physical limit, it is a preference.

It is wherever the buyer stops flinching. Then, before any of this could be turned into an accusation, Kane closed the door on it:

"That's not publishers doing anything wrong. It's publishers being rational operators. They're behaving based on the financial rewards, the financial incentives that the buy side has created."

Fine. Rational actors. Incentives. Everyone's playing the board. It's a familiar speech and it normally functions as an exit ramp, the thing a guest says right before the segment ends and nobody has to feel bad.

Then he produced the number, and the number is the entire story.

"Any publisher would probably be foolish to have a 50/50 ads to content ratio, because they will eventually get flagged and blocked by some big buyers. But they'd also be very foolish to have a 29% ad to content ratio. You want to be exactly at 30%."

Exactly at thirty.

Not twenty-nine, which is charity. Not thirty-one, which is suicide. Thirty, which is the maximum tolerated dose, arrived at independently by thousands of unrelated businesses, like a room full of strangers who each privately calculated how much they could drink before somebody said something.

Desperation is sloppy. This is not sloppy. There is a spreadsheet behind that number. There is a dashboard. There is a quarterly review. Somewhere there is an employee whose actual job is keeping a website pinned at precisely as bad as it is permitted to be, and that employee is good at it, and hits their targets, and receives a bonus, and is not a villain but a professional — which is worse by a considerable distance, because a villain can be fired and a professional gets promoted.

How we reported this: this installment is built on direct email correspondence with four sources between August 14 and August 21, 2026, all of it in ADOTAT's possession, plus arithmetic published in the August 18 and August 20 issues. One source, a former senior executive at PubMatic, is described by role rather than name at ADOTAT's discretion; his correspondence is on file. Fee figures throughout are estimates and are labeled as such, and the composite stack is a range rather than a measurement, because no party in the chain publishes auditable take-rate data, which is the subject of the piece. ADOTAT has no financial relationship with any company named here, takes no paid placement in editorial, and is funded by subscriptions. The media audit consultancy quoted anonymously is one ADOTAT is separately reporting on, and its CEO had an interview scheduled with ADOTAT at the time of publication. Confidential tips: [email protected]