The Short Essay Version

By Andrew Conru.

On January 13, 2021, six days after the Capitol breach, I posted the same question to two websites I had built. The phrasing was, roughly, “What did Trump do, if anything, to incite violence?” I posted it as a sincere “I am trying to figure this out myself” question, which is the posture I had taken on every contested topic since 2017. One of the sites was called agnostic.com. The other was called IDW.community. They ran on the same MySQL backend, the same PHP code, the same humanism seed essays, and the same moderator (me, sleeping six hours a night, taking it personally).

When I went back four years later and asked an AI to score the response threads on civility, ad hominem density, and willingness to engage the actual question rather than the asker, agnostic landed at 2.5 out of 10. IDW landed at 7.5. Same prompt. Same paragraph. Two weeks later I asked “How is Biden doing?” The agnostic thread scored 7.5. The IDW thread scored 3.5.

I wanted, when I sat down to write about this, a clean verdict. The data did not give me one.

A brief about-me, because the data is going to ask you to trust something. I grew up on a 10-acre farm in northern Indiana. My parents were union steelworkers. My mom’s rule was “if you’re going to do something, don’t do it half-assed.” I started writing code at age twelve, in 1982. In 1994 I built the world’s first online dating site. In 1996 I founded Friend Finder Networks, which carried over 700 million registered members across its lifetime. I sold it in 2007 and spent the next several years trying to figure out who I was when I was not building anything, which is harder than it sounds.

Then in 2017 I bought a domain called agnostic.com and went back to building rooms for grown-up conversation.

agnostic.com went live on June 19, 2017. The frame was humanism without God, a meeting hall for people who had walked out of a religion and did not have a place to sit down. IDW.community went live on Valentine’s Day 2019. The frame was open minds welcomed, a Mill-flavored room for people whose questions had stopped being welcome on other platforms. Same engine. Same prose, lightly re-framed at the front door. 36 long-form humanism essays at agnostic became 47 at IDW with the framing tweaked. The only variable I did not control was who came in.

At peak, agnostic held 209,110 user records and 782,123 questions. IDW held 145,827 user records and 431,482 questions. Combined I held the engine, the seed essays, the moderator’s chair, and the abuse logs. Five years later, the data is what the data is.

The matched cross-posts (49 scoreable pairs, drawn from a wider pool of 52 near-duplicates) tell a time arc, not a point estimate.

Pre-2020, on platform housekeeping prompts (button fixes, layout questions, search improvements), agnostic averaged 8.94 out of 10 on civility across 16 pairs. IDW averaged 8.61. Both maxed out. Summer 2020, with COVID and BLM and unrest in the headlines, agnostic warmed to 9.36 across 7 pairs. IDW held at 7.57. Sitting in 2020, I read this as confirmation of the engineering hypothesis I had started in 2017, which was that you could build civility into the rails of a platform and the conversation would come along for the ride.

Then the 2020 election walked in. Across 17 matched pairs from September through December, agnostic dropped to 5.89. IDW dropped to 5.48. Both rooms crashed in parallel. The next 5 pairs spanning January 6 through the inauguration: agnostic 4.04, IDW 6.26. A 2.22-point gap, in the only direction most readers would expect, for fourteen days. The next 4 pairs, with Biden a week into office: agnostic 5.75, IDW 5.25. The average flipped back.

The averages hide the texture. Look at single-question pairs.

January 13, 2021. “What did Trump do, if anything, to incite violence?” agnostic 2.5. IDW 7.5.

January 27, 2021. “How is Biden doing?” agnostic 7.5. IDW 3.5.

February 9, 2021. “What dangerous, wrong, or false things do Conservatives believe?” agnostic 2.5, with 17 separate dehumanization events on the top 25 responses (the highest single-pair count in the corpus). IDW 5.5.

November 9, 2020. The week the Pfizer vaccine emergency-authorization news landed. “Would you take it?” agnostic 8.5, with the science-trusting crowd handling it calmly. IDW 4.0, with the freedom-frame crowd erupting.

Each tribe’s purity reflex was outgroup-triggered, not constant. The room that erupted at one prompt held the next one cool. The room that held the first prompt cool erupted at the next.

The other thing the data taught me was who the rooms actually were.

I had spent seven years assuming, when a thread blew up, that a community of two hundred thousand humanists, or one of a hundred and forty-five thousand heterodox readers, was reacting. That is not what was happening. When I joined the matched-pair response threads back to behavioral clusters of the active user base, two clusters per side generated 92% of every reply on contested prompts. On agnostic: a cluster the model named Thoughtful Elder Skeptics (60.7%) and one named Bruised Seekers (31.8%). About 1,170 specific accounts out of an active universe of 17,062. 6.9%. On IDW: Heterodox Traditionalist Skeptics (75.3%) and Cultural Grievance Veterans (16.4%). About 1,020 of 7,522 active. 13.6%.

The largest cluster on agnostic was something the model named Seeking Connection Seekers, 22% of the active universe. Reading their bios, that was people who had deconverted from a Christian household and now lived in a county where they had nobody to talk to about it. They wrote about isolation. They mentioned their pets. They asked for advice about Thanksgiving with the family. On political pairs they contributed 0.3% of responses. They tuned out. They posted in the relationship threads, the holiday threads, the memorial threads when somebody’s dad died. The “community melting down,” on most days, was about fifty specific accounts. The other ninety-five percent were doing what they had come for.

The cleanest single piece of evidence is the cross-site members. 15,957 people made accounts on both rooms. About half on the same day. 291 of them posted on both sides at meaningful volume. The same human being, between 2017 and 2022, filed 23.3 times more abuse reports on agnostic than on IDW. Received 13.8 times more. On a sub-sample of 108 of those dual-active users, with an AI scoring their tone separately on each side, the same login was 1.46 civility points lower, 2.04 warmth points lower, 2.36 self-disclosure points lower, and 1.89 argumentativeness points higher on the IDW side. 85% of the sample were chameleons whose dominant register meaningfully shifted between rooms. Same person. Same engine. Two doors. Two voices.

I owe you one more pair from the matched corpus. December 18, 2020, three weeks before the Capitol breach, I posted to both rooms the same question: “Should some research topics be banned or results censored?” agnostic civility 6.5. IDW civility 7.0. Both rooms engaged it. About 18 agnostic responders, mostly Bruised Seekers and Thoughtful Elder Skeptics, working through their views, mostly arguing the same direction, that you do not ban scientific questions, you do them carefully and you publish the findings. About 26 IDW responders, mostly Heterodox Traditionalist Skeptics, arguing the same way, with a few dissenters who got heard. Nobody got reported. Nobody got blocked. The threads are still in the database. The rooms I had built were able, on a Friday afternoon in December 2020, to hold the actual hard question.

So which side was meaner online?

Neither, for almost three years. Both, for about fifteen months. Each one, more than the other, depending on which prompt was on the table at which week.

The headline ratios on agnostic, the ones I used to quote at investors and at parties (17 times more blocks, 29 times more formal abuse reports than IDW, system level), are real. They shrink to 4.7 times more blocks and 9.6 times more user reports per active-user-month, once you control for engagement. Most of what is left is one cluster pair on agnostic (313 elder skeptics and 859 bruised post-religious people) reciprocally combatting each other while the rest of the community read past the fight. IDW does not have that single dominant pair. It has a sparser, thinner pattern, with a smaller power-reporter tail and a more diffuse combative tone in the daily threads.

What the matched-pair data taught me, once I let it, is that each tribe was civil everywhere except the few topics it would not let you question. Inside the protected zone, where the prompt named something the room loved to think about, civility was portable, durable, and beautiful to watch. Outside that zone, where the prompt named one of the room's untouchable topics, civility flipped into its opposite use. It became the polite vocabulary of expulsion. The same person, in the same chair, in the same room, could run both modes within ten seconds of each other. I know this is true at the individual level because the cross-site members did exactly that, on the same engine, with the same essays at the front door, between two browser tabs.

I am not equipped, after eight years of running both rooms, to tell you which national tribe has the worse manners. The honest claim the data forces is structural. Both rooms melted down. Each one melted down, predictably, at the prompt that named its outgroup. The medium amplified both. The moderator did not slow either of them down. If a reader wants a verdict on which side is meaner, they have to bring it in with them, because the data does not.

I am writing this in 2026 from Seattle.

I closed the question of whether civility could be engineered into a medium itself a few years ago. The answer, near as I can tell, is yes for housekeeping prompts and no for outgroup-triggering ones, and the line between those two has been moving faster than any dashboard I built could keep up with. So I started building rooms with walls instead of comment threads. There is now a 16,000 square foot art salon at 110 Union Street, downtown Seattle, across from the Seattle Art Museum. Most Wednesdays, between 4 and 7 pm, I am there. There is coffee. There is a painting on the back wall. There is no thread. The thesis I am running now is not about engineering civility into the medium. It is about whether a room with a beautiful object in it and a regular night and a few people who keep coming back can do something that a comment thread, on a worse year, could not.

If you live somewhere I can drive to, come by. If you do not, find a regular night and a beautiful thing and three to seven people who will keep showing up, and pour the coffee yourself for the first six months, which is also when you will think it is not working.

That is the experiment now. The previous one is in the data, dated, signed.

Andrew Conru founded Friend Finder Networks in 1996 and runs the Conru Art Foundation in Seattle. The Art Love Salon is at 110 Union Street, across from the Seattle Art Museum, open Wednesdays 4 to 7 pm. This essay draws on data and material from a longer book in progress, available in several versions at conru.com/tribes/.