Last Thursday I published a spreadsheet of everything my coworkers had told me privately about one another. It included who thought our manager was sleeping with a contractor, who expected layoffs, who had applied elsewhere, and which two people had called the new office a fire trap. I shared it in the company channel at 9:06 a.m. with a note that the evidence was uneven. At 9:13, the manager called an all-hands meeting, which several people have interpreted as evidence that my intervention was premature. I think the eight-minute response time is actually one of its stronger results.
The problem was common knowledge. It isn't enough for Priya to think the project is behind schedule and for Marco to think the same thing; each needs to know that the other knows, and know that the other knows she knows, before either is willing to say it in front of the director. A private conversation cannot reliably create this recursive state. An announcement to the whole office can. I had spent six weeks collecting claims that people were plainly willing to make to me but unwilling to put into the shared channel. The gap suggested a substantial coordination loss.
I used columns for Claim, Original Speaker, Confidence, Date Heard, Other People Implicated, and Whether the Speaker Asked Me to Keep It Private. The last column was necessary because a promise of confidentiality is evidence about the expected social cost of disclosure. Removing that information would have biased the analysis in favor of easy cases. I left the contractor's name out of the first version, though her department and the distinctive green helmet she wears while biking were enough for several people to identify her. This gave us a useful natural experiment in how little identifying information a rumor requires.
Marco says I promised not to repeat what he told me. I remember promising I wouldn't tell the director he had said it. The distinction mattered when the director asked where the layoff claim came from. There were two independent sources by then: Marco, and Priya repeating what Marco had told her. I listed both. Priya objects that I quoted her precisely enough that she could be identified from her punctuation. I have offered to standardize everyone's punctuation in the next release, although she says she would prefer that there not be a next release.
One critique is that I conflated rumor with fact. This is why I included confidence estimates. The manager-contractor relationship was at 35 percent, and the fire-trap claim at 70 percent, conditional on the hallway exit being blocked as described. I am under no illusion that assigning numbers makes these claims true. It allows readers to weigh them and correct them. The contractor did correct the record: she and the manager are cousins. That update took less than an hour. Without the spreadsheet, some people might have continued to believe the relationship rumor for months, with no centralized place for the correction.
Another critique is that publication changed the office. The director has asked employees to stop confiding in me; two people have withdrawn job applications because they no longer want their names associated with them; the contractor works from home. These are interventions on the thing I was trying to observe, and I have marked the data after Thursday as post-treatment. I don't think it follows that the pre-treatment data was unimportant. Before I published it, everybody was making decisions under a different picture of what everybody else knew.
We have a meeting with HR on Tuesday. I asked if I could circulate a short anonymous survey beforehand to determine how many participants believe the others will speak candidly. HR said no and asked me to bring my laptop. This is the first time in months that I've been confident everybody knows what the meeting is about.