Seven errors in a book about errors

I wrote a book with a thesis that now looks a little ambitious. The thesis is that the numbers in business advice mostly come from nowhere — that if you follow a confident statistic home, you’ll find a firm that published the opposite, or a sample size that changes shape between two paragraphs, or a man in a vest.

Then I ran a full fact-check on my own book. Seven errors. Six of them in the source list, which is the single worst place in a book to keep your mistakes, like storing your fire extinguisher inside the fire.

Here they are.

The one I got exactly backwards

The book cites Substack Pro as an example of paying suppliers who bring their own audience. I described those deals as advances recouped against Substack’s 85 percent first-year revenue share.

That’s the opposite of what Substack Pro is. Recoupment describes the earlier programme Substack ran and then killed, specifically because writers ended up owing money back. Pro exists to remove that: Substack pays a non-repayable sum guaranteeing a writer’s first year, takes 85 percent of that year’s subscription revenue instead of its usual 10, and reverts to the standard split afterward. The company absorbs the risk. The writer owes nothing.

I had Hamish McKenzie’s post open. It draws the distinction explicitly, because the distinction is the entire point of the programme. I read it and wrote down the reverse.

The irritating part is that the correct version makes my argument stronger. Pro is an expensive instrument precisely because nobody claws it back. I had a better example than the one I published and I fumbled it into a worse one.

The two where I named the wrong person

Airbnb’s founders photographing their early listings is one of the most-cited moves in startup folklore, and I cited it too. I credited Nathan Blecharczyk, sourced to a 2014 conversation where he says “we.”

Paul Graham names them in the essay I was already citing, two paragraphs from the line I quoted: Brian Chesky and Joe Gebbia, taking “professional” photos — Graham’s scare quotes, not mine — of their first hosts’ apartments. Blecharczyk was the engineer. He didn’t go. I turned a collective pronoun into a named attribution, which is a small thing and also the exact move I spent a chapter complaining about.

The second one is subtler and worse. The book describes OpenTable pulling back from fifty cities to four, then quotes “its own chief executive” on why restaurants adopted the product. Two different chief executives, six years apart. Thomas Layton ran the retrenchment starting in 2001. Jeff Jordan, whose quote I used, didn’t arrive until 2007. I ran them together into one sentence and manufactured a continuity that never existed.

The one where I hedged the true thing and asserted the shaky one

Same OpenTable passage. I wrote that the fifty-cities story “comes from the standard business-school case; verify against the 2009 S-1 before repeating it.”

Good instinct. I should have taken my own advice, because I’ve now read the S-1 and it doesn’t support the story at all. It presents four metropolitan areas as the original focus and never mentions a retreat, a burn rate, or fifty of anything. So I hedged toward a document that says something else, which is worse than not hedging.

The retreat itself appears to be real — the numbers check out against independent accounts, and Layton’s arrival in 2001 is well documented. It’s just sourced to secondary retellings, which is what the book now says, in those words.

The one where I put a range in a dead man’s mouth

Not dead. Andrew Chen is fine. But the book had him being “explicit that the smallest stable network varies from two people to thousands depending on the category,” and he never wrote that range. The varies by category half is solidly his and it’s the load-bearing half. The numbers on either end are mine, attributed to him, in a chapter whose argument is that people put numbers in other people’s mouths.

I have read that sentence perhaps forty times. It has the shape of a citation. It has a name in it and everything.

The one where the arithmetic ate a variable

The book’s Step 2 has two constraints and argues you take the larger, not the product. Then it demonstrates what the old, wrong formula would have given you:

Multiplying the constraints gives (10 ÷ 2) × 1 × 10 = 50 contractors.

That expression isn’t the product of the two constraints. It’s the product of two quantities the book never defines, because the target fill rate quietly fell out of it somewhere. The actual constraints in that example are 3.5 and 12. Their product is 42. So “38 people they did not need” is thirty, and the overstatement is three and a half times, not four.

I built a whole example around catching someone else’s bad multiplication and got my own multiplication wrong inside it. There’s a version of this where I’d defend it as a demonstration.

While I was in there: the book says moving from 70 to 95 percent fill “roughly triples” the pool you have to hold. It’s 2.49. In a chapter accusing a rival formula of overstating by four, rounding 2.49 up to three is the same crime in a smaller jacket. And it says “six steps” five separate times about a sequence that starts at Step 0 and ends at Step 6, which is seven, as anyone who counted discovered before I did.

The one that cited a claim I never made

The source list confirmed the macaroon-shop origin of DoorDash against an interview. Solid source, real quote, no macaroon shop anywhere in the book. I sourced a fact I hadn’t published, which is a new kind of thrifty.

Related and more substantive: the book credits Stanley Tang’s 2013 blog post for the PDF menus, the Google Doc, Square, and Find My Friends. Those details are all Tang’s, and they’re all from a lecture he gave a year later. The post doesn’t contain them. I also printed “eight local restaurants,” a number that circulates widely and traces to nobody, so it’s “a handful” now. And Andy Fang says the first order came in 45 minutes where Tang says half an hour — the accounts diverge, and a book like this one has no business quietly picking a side without saying so.

What this cost

An edition. Everything above is corrected in 2.1, which is on Gumroad and Amazon now. If you bought Edition 2, the corrections are this page, and the substantive ones are the Substack reversal and the Step 2 arithmetic. Everything else is sourcing hygiene.

It cost some pride, but pride was never the expensive part. The expensive part is that for a few weeks the strongest thing I could say about the book — check every number in here, I did — was not true. That’s the only claim the thing has. It doesn’t have a research department or a logo on a chart. It has one promise, and I broke it in seven places before anyone else got a chance to.

The rule I should have had

Every error above has the same shape: I had the correct source in hand and summarized it wrong. Not one of them came from a bad source. They came from me reading a good one at speed and writing down the version I expected to find.

So the rule is dumber than I’d like it to be. It isn’t find better sources. I had the best sources available in every single case — Graham’s essay, McKenzie’s post, an S-1, Tang’s own account. The rule is: after you write the sentence, open the source again and read only the part your sentence claims. Not the article. The sentence. Out loud, if you’re alone, which you will be.

The man in the vest doesn’t do this. That’s the thing I keep coming back to. He’ll never publish a list like this, because his product isn’t the number, it’s the certainty, and certainty doesn’t survive an errata page. I’d rather have the errata page. It’s the only part of this whole business that’s actually checkable.

The corrected book is here. If you find number eight, tell me and it goes on this page with your name on it, or without, your call. I’d genuinely rather hear it from you than from the vest.