MMH - Museum of Media History
Note from Robin:
This saga appeared in my secure dropbox late last month, along with a plea to post it on Facebook today in exactly the format you see here. I gather I’m not the only one to have received those instructions. I can’t vouch for the authenticity of the story, but I thought it was certainly weird and interesting enough to share.
The last thing I want to do is write this down, but I’m doing it anyway, partially because people ought to know what’s happening with the things they post here, but mainly (like 99%) because of Julie Rubicon and the spike.
My former peeps at Facebook Inc. of Menlo Park, California—hi Jane; hi Neel; hi… Mark?—will know immediately who’s written this, and the company will probably pursue me, but I predict it will do so quietly. The SEC won’t be so discreet, if indeed there is a rule covering the deeds that follow, but honestly… I’m not sure there is.
I’m typing this on February 27, 2016. Today was my last day at Facebook. I turned in my badge and my laptop and I walked onto Willow Road with a flash drive containing the images you’ll see below. Outside, I watched the dogfood builds disappear from my phone’s home screen one by one.
It feels strange not to be a Facebook employee, even though I spent most of my time there expecting to be fired. I started on the product team, where I did not excel, moved over to ads, which was worse, and ended up in PIG.
Facebook offers, to certain of its largest advertising clients, the services of the Partner Intelligence Group. PIG is where my story begins.
Any individual user of Facebook sees only see a narrow, personalized slice of the system. Facebook itself has a broader view. From my desk in the PIG pod, I could run queries across all posts and comments, public and non-public. Private messages, too. I could ask: how many people on Facebook talked about the U.S. presidential election today? How many of them posted something about Donald Trump? How many of those posts included the emoji? (On the morning of February 27, the answers were: 65 million; 42 million; 32,541.)
Advertisers love this data, but obviously, Facebook can’t give them direct access. (That is obvious, right? It should be obvious.) So, there’s a gatekeeper. If you spend a very large amount of money on Facebook ads—I never learned how much exactly—PIG will prepare custom reports explaining how your various brands and/or products are being discussed across the entirety of the system. As of last October, PIG can also roll up data from Instagram, and WhatsApp is on its way.
There’s nothing sinister about this; all the internet companies do it. The data is scrupulously anonymized. It’s the view of a billion users from 30,000 feet.
Just to give you an example, here’s a chart I made for Adidas last year, with some helpful annotations:
You can see the spike in mid-February following the release of the Yeezy 750. It’s a simple story: cause and effect. My counterpart on the Adidas marketing team took that graph back to her superiors and said, look—we made a dent in the universe!
There were three of us in PIG: me, T, and Julie Rubicon, whose real name I’m using for reasons that will soon become clear. I joined the team with the lowest possible expectations, but it turned out to be great, primarily because T and Julie were smart and interesting. T was legitimately fascinated by the various brands and/or products we analyzed; our clients loved her, and it was clear to me she belonged on the other side, requesting reports instead of producing them. (Get on it, T!)
Julie Rubicon was different. She had joined PIG from the user operations team, a total grind, and she was hungry. I had washed out into PIG; Julie had clawed her way up. This difference in our attitudes was obvious to everyone, including Julie, and on my second day in PIG, she called me an asshole. For that, I will be forever grateful. My resentment curdled into spite that hardened into determination and, lo and behold, I learned to actually do the work. Julie and I became friends.
In the spring of 2015, I drew a very strange graph.
All PIG queries are processed by an internal application called Enchilada. These queries have two parts: a collection of comma-delimited terms (“yeezy 750, yeezy boost, yeezy 750 boost”) and a date range: start date, end date.
I was preparing a report for Vernix, a children’s footwear brand. I entered my terms as usual (“vernix, babyboots, vernix babyboots”) but when I set the date range, I did so incorrectly, omitting the end date.
I submitted this query almost a year ago, in April 2015. Weird to realize that.
Enchilada should have returned an error, but something was twisted up in the data center, and instead I got a graph that ended—arbitrarily, it seemed—in October 2016. The data squiggled senselessly out into the future.
I wrote it off as a weird bug and redid the query. Parents were buzzing about Babyboots.
On June 1, 2015, Vernix announced it had been acquired by Nike, and our contacts there emailed to tell us the PIG reports would be consolidated under that account. Something tickled the back of my brain. I dug in the single overflowing folder that held every document I’d ever created to retrieve the erroneous graph.
The spike lined up perfectly.
Therefore: I had a graph showing the chatter around Vernix’s sale, produced two months before any of that chatter occurred.
I spent the day haunting the roof of MPK20, feeling jangly and nervous, staring out across the bay.
That night, I submitted a query unrequested by any client. The NBA Finals were set to start, Warriors vs. Cavs, so I asked Enchilada to plot the posts about both teams. This time, when I omitted the end date, it was intentional.
Enchilada did not predict the winner, per se. It just predicted the conversation. But conversation correlates strongly to real-world events; that’s why advertisers care about it in the first place. Conversation spikes when championships are decided. When companies combine or collapse. When politicians are engulfed by scandal. When people die.
On the day of game six, I showed the graph to Julie Rubicon. In a series of rising spikes, it had predicted the results perfectly, and the largest and sharpest of those spikes prophesied that the Warriors would win the championship in spectacular fashion.
The next morning, Julie approached my desk. “Walk with me,” she said. I followed her across the Bayfront Expressway onto the trail abutting the salt flats, and there, voice raised against the wind, she told me we were going to start trading stocks.
We found a list of the largest companies in the world. We proceeded cautiously. Every system at Facebook is monitored, and we were certain that a barrage of ticker symbols would alert some software watchdog. Our normal, non-suspicious PIG activities entailed perhaps a dozen queries every day; we decided it would be safe to add two.
We didn’t tell T, and T, if you’re reading this, I’m sorry.
As we worked our way through the list, we spun theories about this strange new “feature.” Facebook’s engineers had, in recent months, woven powerful neural networks into many of its systems. Could Enchilada have been connected, intentionally or not, to some super-brain with the ability to not only analyze but also extrapolate, and not only plausibly but perfectly? Maybe. Or maybe a rat chewed through a cable. And maybe that rat was magic.
We got the graphs back two at a time. Apple and Exxon Mobil. Berkshire Hathaway and Google. Mostly, they looked random. Meandering. The future was boring.
Then, in August, I saw this:
In case you can’t tell from the y-axis: it was a very large spike. It was the Burj Khalifa of spikes. It was September 2015.
Can you guess?
The graph showed all the posts and comments on Facebook, public and non-public, from the recent past and maybe also the near future, discussing… Volkswagen. I told Julie it had to be something terrible. I mean, a car company? We’d run reports on the launches of new vehicles, lots of them. They never looked like this.
Julie and I each chipped in $2,000. I read a bunch of blog posts, learned how to short a stock, and bet it all against VW. On a day in September, we made $1,000 profit. We jumped and high-fived and hooted into the wind across the salt flats.
We were eager to keep going, but VW-scale spikes were elusive. In November, I thought I saw something sharp and suggestive in the Walgreens graph. I was wrong. As easily as we gained our thousand, we lost it.
This carried through into December: small gains, small losses. We’d found a magic lamp—there was no question about that—but I had a creeping sense the genie inside didn’t speak our language. Come si dice, “I want to be rich?”
As before, Julie Rubicon got there ahead of me. We took another walk. She said: forget this day-trading shit. “Let’s become data brokers. We’ll start a secret company hidden inside Facebook.” It would be reachable only through a site on the dark web; it would provide only raw data, not interpretation; it would accept payment only in Bitcoin.
Julie read a lot of William Gibson novels.
“Word will get out,” she said. “Hedge funds will want this. They’ll come begging.”
Julie was hungry.
I read a bunch more blog posts, learned how to set up a site on the dark web, and had just launched it when Julie Rubicon, on the last Monday in January, for reasons unknown to me, asked Enchilada about… us.
She queried our names.
Here’s the graph for mine:
Exactly as it should be. I’m a normal person, not a celebrity or a politician. Not a brand.
Conversation spikes when championships are decided. When companies combine, or collapse. When politicians are engulfed by scandal. When people die.
Why in the world would the conversation spike for Julie Rubicon?
That afternoon, looking out across the salt flats, she trembled with certainty. “They know.”
They don’t, I assured her. They can’t.
“They will. They’ll look at the logs. We’re gonna get in trouble. We’re gonna go to jail.”
“I don’t think there’s any law against this,” I said. How could there be a law against something that’s not possible? I told her not to worry.
But I wasn’t the one staring down a spike.
January faded into February. In our PIG stand-ups, Julie seemed normal—sharp and sure, just like always—but whenever I asked quietly if we should maybe talk about the secret company thing, her eyes went cold, and she replied: later. I’m busy with Puma.
It was always Puma, for some reason.
I didn’t query any ticker symbols, didn’t short any stocks. There were no visitors to our site on the dark web, because no one knew to look for it. In spare moments, I ran open-ended queries on presidential candidates. I won’t tell you what I saw.
And I watched Julie query her name, over and over and over. Every time, the graph showed the same spike in mid-March, towering and implacable.
On the day after Valentine’s Day, the area around the PIG pod was still festooned with expensively rustic red doilies. T and I gathered for our stand-up. Julie wasn’t there. We waited fifteen minutes. Still no Rubicon. T messaged her. I dialed her number. Nothing.
That afternoon, she missed her call with Puma.
Soon, our manager Jane was locked in a conference room with an HR specialist, running through the missing-employee playbook.
I understood exactly what had happened.
Contemplating the spike, just two weeks out and closing fast, unable to imagine what could possibly cause a not-insignificant fraction of Facebook’s users to say her name in unison, but absolutely certain they would—Enchilada’s predictions, when legible, had been nothing but right—and burdened with the knowledge, earned over thirteen months in PIG, that the sharpest spikes signify disaster and disgrace… contemplating all that, Julie Rubicon did the totally reasonable thing.
At first, I was angry, mainly because she hadn’t asked me to go with her. But quickly, anger gave way to angst as I imagined all the ways in which, two weeks hence, on the day of the spike foretold, Julie’s name might resurface in a gout of awfulness. Abduction, plane crash, bomb blast—my imagination careened down all the darkest corridors. I barely slept.
And then, suddenly, I understood how I could fix it.
Conversation spikes when championships are decided. When companies combine or collapse. When politicians are engulfed by scandal. When people die.
And maybe also when people tell the truth.
Here it is.
Facebook’s Partner Intelligence Group uses an application called Enchilda to scan and summarize all the posts and comments on the system, public and non-public. It delivers the results of those scans to advertising clients, not for nefarious purposes, but simply so some human somewhere can say to her superiors, look: we did something right.
In the fall of 2015, Julie Rubicon and I used an undocumented and uncanny capability of the Enchilada application to inform several trades on U.S. stock exchanges, which generated a net profit of $162.
I’m sending this message to several journalists and writers with the expectation that the majority of them will dismiss it as mischief or fiction. But I know how Facebook works; I mean, I really do. Even if only a few of my recipients post this, and only a few of their friends and followers pass it along, it will spread.
And I will be, if not absolved, at least unburdened.
And the internal apps team will fix Enchilada. (Neel: seriously. Fix it.)
And this message, all the different copies of it, versions compressed and retold, shared across the system—this can be Julie’s spike. Not a scandal. Not a disaster. Just a true story.
Julie, if you’re reading this—whether it’s come to you through Facebook or a link posted elsewhere—I think it means you’re safe. It means all the people who shared this were actually working together to blunt the point of Enchilada’s prophecy. It means, if you want to—and I’ll understand if you don’t—you can visit our site on the dark web and use the email address there to tell me where you’ve gone. My bag is packed.
Okay, actually—might as well write her name a few more times. Each one counts separately.
This is for you, Enchilada: