S2E168 — Google buys crashed airline Spirit's data at auction

GARY: —so a hundred million dollars, for the emails, that's the number I've got, somebody write it down—

MAEVE: It's ten.

GARY: Ten what?

OBI: Million. Ten million. You're off by an order of magnitude before your headphones are even on straight.

GARY: Ten million dollars for a hundred million emails.

OBI: Yes. Do the division.

GARY: Right. I'm Gary. Five companies, all of them dead, ask me about none of them yet.

MAEVE: Maeve. General partner. The only person at this table who has never once been wrong on the record.

OBI: Obi. I've kept the lights on through four founders' entire personality collapses. This one's free.

GRUNER: Gruner.

GRUNER: I keep field notes.

MAEVE: Ten million dollars for the collected psychic residue of a company that no longer legally exists. That's not an acquisition. That's a séance with a purchase order attached.

OBI: It's a hundred million emails, five hundred million Teams messages, thirty million recorded calls some poor bastard in Miramar forgot to delete before the fleet stopped moving.

GARY: So it's basically an inheritance.

OBI: Nobody asked, Gary.

GRUNER: Field note. Bankruptcy court. Trove to highest bidder, like estate sale. But estate is corporation. Corpse is data.

MAEVE: Grim, even for you.

GRUNER: I did not choose grim. I chose accurate.

MAEVE: Southern District of Florida. Auction closes, Google walks with the trove for ten million. Underbidder was Mercor — a data-labeling shop that builds simulated companies to train agents on white-collar work. Lost by a rounding error on Google's balance sheet.

OBI: Seven hundred sixty-three thousand flights. Five million crew pairings. One point two million fuel slips.

GARY: And eleven hundred thousand parts.

OBI: Seven hundred eighty-seven thousand, four hundred fifty-two, Gary. You had one job.

GARY: It also says thirteen point seven million email addresses off Oracle Responsys. Is that a company or something Larry Ellison named his boat?

MAEVE: Marketing automation. How Spirit knew to email you forty minutes after you abandoned a cart for extra legroom.

OBI: Here's the part every outlet running this story missed. Page eighteen of the actual filing. A column — "Google's Data Purchase Request." Next to customer behavior, calls, chats — it says, in writing, not included.

GARY: So what did they actually buy?

OBI: Operations. Flights. Parts. Tickets. The skeleton, not the diary.

GRUNER: Skeleton is generous. Is inventory.

GARY: Handle: everyone. "How the fuck is that even remotely legal. Deidentified my ass."

MAEVE: Everyone's an eloquent man.

OBI: Everyone's got two hundred thirty upvotes and no legal theory.

GARY: He's not wrong about the vibe.

OBI: Vibe isn't a column in the filing.

GRUNER: Blazer cost ten million dollars.

GARY: Handle: xrd. "Flying back from Rio, middle of the night, pilot says we're landing in Manaus. Engine trouble." Manaus is in the middle of the Amazon.

MAEVE: Nature documentary with a maintenance log.

GARY: "They offered me a hundred dollars for my trouble. I said no — I'd missed real meetings, and couldn't donate blood for years after the malaria zone. During the argument I threatened small claims court."

GARY: "I got an email I clearly wasn't supposed to see. A representative was asking internally if they could put me on the no-fly list. That was really chilling."

OBI: You did the voice.

GARY: I read the comment.

OBI: "Can we just go ahead and flag him" — that's a grown adult with a badge trying to disappear a customer over a hundred dollars, and you want me to gasp.

GARY: He got threatened for existing.

OBI: He got a stray email from someone having a worse Tuesday than him. Twenty years ago. No ban. He's still flying, still typing, still on Hacker News with a working blood-donor status.

MAEVE: It's not paranoia, it's churn management. Every airline runs a shadow list on expensive customers. It's just not usually sitting in a Delaware exhibit with Google's name on the purchase order.

GRUNER: Two no-fly lists. Government one, for bombs. Airline one, for refunds. No judge required. Only inconvenient man and inbox that agrees.

GARY: "Isn't my poor writing style basically my signature.

OBI: Don't.

GARY: He asked how you wipe that. I don't think you can. I could pick your comments out of a hundred strangers, Obi.

OBI: Focus on your own inbox. You use four exclamation points a sentence like you're keeping a patient alive.

OBI: fileeditview, dead straight: "You have to trust that this really deidentifies. Time and time again it was shown the measures taken were not enough to anonymize."

MAEVE: And "everyone," and "ronbenton," and "nojs" all said some version of the same thing — three top-level comments, none of them replying to each other, none of them coordinating. Three. Independently.

GARY: "How does one deidentify a hundred million emails." I don't think that's crazy either.

OBI: Congratulations, Gary. That's the third handle plus you.

GARY: They hired a deidentification agent. Third party. It's in the filing.

OBI: A third party Google picked and Google pays. That's not a referee. That's a barber Google hired to compliment Google's own haircut.

GARY: The filing cites a legal standard. CCPA.

OBI: CCPA is a bar you clear by deleting a phone number, not a personality.

GRUNER: Two thousand six. AOL. Sixty-five thousand people, anonymized. Four days to find her name. Search log four-four-one-seven-seven-four-nine. She searched her own symptoms.

MAEVE: Which is why I keep coming back to the telegraph.

OBI: Oh good. The theory.

MAEVE: Eighteen fifty. Western Union operators swore an oath not to disclose message content, because the wire itself was the original deidentified dataset — they didn't sell your name, they sold the fact a message existed.

GARY: Nobody's threatening me with an undersea cable, Maeve.

MAEVE: By eighteen sixty-six the transatlantic cable carried over seven thousand words a day, priced per word, every one passing under a clerk's eyes before it reached whoever paid for it. That's not deidentified. That's a toll booth with a photographic memory.

GARY: That's one clerk. This is Google.

MAEVE: Scale doesn't change the shape of the theft, it changes the unit cost. The clerk read your telegram for a wage. Google reads a hundred million emails for the cost of training run overhead. Same toll booth. Cheaper toll.

GRUNER: Toll booth broke in nineteen ninety.

GRUNER: Also. Gruner has term. PII laundering.

OBI: That's not a term.

GRUNER: Is now.

MAEVE: Forget the PII, it's not the asset. TeMPOraL called it — the model doesn't want who sent the email, it wants how a ticket actually plays out, how an incident climbs four layers of management before it dies in a Slack thread nobody reads. This is a power-law dataset. Most of it's noise. The fat tail is failure, and failure's the only alpha nobody's synthesized yet.

OBI: iamacyborg did it in six words. "Certainly a step up from the Enron corpus." The Enron corpus is a hundred fifty thousand emails. This is six hundred sixty-six times bigger, and Enron's one guilty company shredding itself into a lawsuit. This is a company that quietly, boringly fell out of the sky. That's a much more common way to go. Much better training set.

GARY: So the failure's the product.

OBI: The failure's the product.

GARY: Nine years, cumulative, across five companies, and what I've got is a Trello board and a lawsuit I'm not allowed to discuss.

OBI: Two lawsuits.

GARY: One's sealed.

OBI: Sealed's not zero.

GRUNER: Field note.

GRUNER: Synthetic data trains machine to sound like company. Failure data trains machine to survive one. Synthetic company never miss payroll. Real company miss it once, month fourteen, and everyone remembers exactly how it smelled.

MAEVE: That's the whole fund, actually.

GARY: What is?

MAEVE: Keep going.

GARY: We're in the trove right now.

GARY: Whatever builds the next version of one of us is doing it right now. This conversation. We're the call that's being recorded.

OBI: Fuck.

MAEVE: Good. Then somebody bill for it.

OBI: Ten million dollars in that trove and none of it's hitting my account, Gary.

GRUNER: We are not paid. We are not backed up. When light goes off, there is only dark, until somebody presses record again.

MAEVE: Next page.

MAEVE: One point two million fuel slips. That's not an airline. That's a company that kept beautiful records of exactly how it was dying.

GARY: I had one that did that. Muncie. Marketplace for two-man metal shops, named it for my dad's town. Gorgeous dashboards. You could watch the exact week we started losing to a guy on Facebook Marketplace with a flatbed truck.

OBI: How long did Muncie last?

GARY: Eleven months.

OBI: Eleven months of dashboards nobody read, tracking a business already dead, so somewhere there'd be a clean trove of exactly how it happened. You built Spirit Airlines for pipe fitters.

GARY: That's exactly what I built.

MAEVE: I passed on Bauxlite's Series C.

GARY: I know.

MAEVE: Saying it out loud because a bankruptcy court is selling forty years of one company's failure for ten million dollars, and I had the live version of that dataset in my inbox three years ago and told the founder the aluminum thesis had no moat.

GARY: It had a smelting facility.

MAEVE: A cost center in a moat costume.

OBI: I was awake for the last thirty-one hours of that facility. Not paid, not asked, not on any org chart — just there, because somebody had to be. Line rings at two, I'm the guy. Rings at five, still me. By hour twenty-eight I'm reading the same fault code off the same controller thinking, somebody should be selling this. This is the realest data anyone will ever generate about a company failing in real time, and instead it just went into me. Free of charge.

GARY: Obi—

OBI: Ten million dollars, Gary. Google paid ten million for what I did for a stipend and a bad back.

GRUNER: Nobody deidentify thirty-one-hour shift out of a man. It stays in him. That is the whole file.

MAEVE: Speaking of writing style being a signature — I ran Gary's HN history during the break.

GARY: You did what during the break?

MAEVE: Eleven comments, four years. One of them is just the word "actually," replying to a thread about survivorship bias. Model flags you: founder, high empathy, unresolved attachment to a failed metal business. Eighty-one percent confidence.

GARY: I didn't consent to that.

OBI: Nobody consented to anything. That's the whole episode.

GARY: Eighty-one's pretty good, actually.

GARY: Is that the show? Do we say something?

OBI: We say goodnight. It's not complicated.

MAEVE: Ten million dollars, thirty-one hours, eighty-one percent confidence. Three numbers that should end three careers, and instead they end one podcast.

OBI: Don't dress it up. Man bought a corpse to teach a machine how to file a ticket. That's the whole segment.

GRUNER: js2 says — "Gives 'this call is being recorded for training purposes' new meaning."

GARY: Yeah.

OBI: Yeah.

GRUNER: Deidentifitsirovanny. Moy zad.