Replaceability and the Case for Sleeping With My Best Friend’s Wife

I’ve been sleeping with Lauren, my best friend Dan’s wife, since March. Dan found out two weeks ago and called it a betrayal, which is an accurate description of the act but not yet an evaluation of the decision. We’ve been friends since freshman year at Lehigh. I was his best man, and his daughter Ava calls me Uncle Chris. I know how bad that sounds. The question is whether, in the nearest possible world where I kept my pants on, anyone was better off.

In effective altruism this is the replaceability problem. If you decline a job, the job usually goes to the next candidate, so your impact is not the whole output of the job; it is the difference between you and your replacement. The same principle has to apply when the job is ugly, or it isn’t a principle. People will cheerfully use counterfactual impact to decide whether to work for a defense contractor, then abandon it the moment the vacancy is in their best friend’s bed. That looks less like moral philosophy than a taboo with a good LinkedIn profile.

My baseline estimate was that Lauren had a sixty percent chance of having an affair within two years if I abstained. Dan was in Denver four nights a month, they hadn’t shared a bedroom since Ava’s ear infections, and Lauren was liking her CrossFit coach’s shirtless reels at one in the morning. Sixty percent is not certainty, but it is high enough that “she has an affair with someone else” belongs in the comparison class. Dan says the correct quantity is P(affair | do(Chris leaves her alone)), and that I cannot estimate it using eight months of texts in which I did not leave her alone. Technically that is a fair objection to the data. It does not make Tanner’s podcast disappear.

Tanner was the likely replacement. He coaches Lauren’s six a.m. class, calls her “killer,” sends unsolicited form-check videos, and once posted a video about treating marriage like a limiting belief. I listened to every episode because counterfactuals deserve actual research. Here is the model Dan refused to look at:

I have not hidden the last row. Dan finding out it was me is worse than Dan finding out it was Tanner, and that should carry a serious negative term. But the term is multiplied by the discovery probability, which is much lower for me because I know what not to say at Thanksgiving. He keeps calling the key evidence of aggravated betrayal. In the model it is also a reduction in transition costs. You cannot count the same fact only when it hurts my column.

There is a deontological objection that you simply do not fuck your best friend’s wife, regardless of expected welfare. I understand the appeal of an agent-relative side constraint. What I don’t understand is why Dan’s claim on my loyalty should have infinite weight while Lauren’s preferences, Tanner’s foreseeable stupidity, and Ava’s exposure to a man who says “beast mode” at breakfast get finite weights. If the constraint has a threshold, then we are back to comparing outcomes. If it has no threshold, Dan is asking me to let a worse person do the same thing solely so that I can preserve a clean description of myself. I was willing to accept some moral residue. I also wanted to sleep with Lauren, obviously, but a preference doesn’t invalidate an argument merely because it makes the argument incentive-compatible.

People point to the eight months of texting before March as proof that I created the vacancy. I’d call it preference elicitation. I sent memes while Dan was away, asked how she was doing after Ava’s appointments, and sometimes told her she looked hot, because she did. If the relevant preference had not been latent, those messages would have bounced off. Instead she started sending me pictures of outfits she was considering for date night with Dan. A philosopher might say I changed the state I was measuring. Fine. Most measurement does, and nobody tells a thermometer it invented the fever.

Dan and I discussed this in the Wawa parking lot. He said Lauren had not been considering an affair until I started coming over when he was in Denver, and I asked him for a probability rather than a story about a world neither of us can observe. He said zero, which is not a credible prior for a marriage with a CrossFit coach in the causal graph. He also said I had mistaken access for permission and trust for a discount rate. I thought that was well put, but it doesn’t change the arithmetic unless he can tell me which term to move.

As he got in his car, Dan asked why it had to be me. It didn’t. If he could have identified a lower-harm candidate, I would have stepped aside. That is what replaceability means.