My daughter Tess is ten. Since September a classmate named Madison Kreider has made her cry on about four school days out of five. My wife's response has been to call Madison "a little monster," email the teacher twice, and tell Tess that none of it is true. I understand the impulse. But "none of it is true" is a confident claim about seven separate propositions, and nobody in my house had evaluated any of them.
To steelman a position is to build its strongest version before you answer it, and the practice matters most when the position belongs to someone you dislike. It is easy to do for a genius philosopher. It is harder when the other party is a fifth grader who has called your child "soup girl" since the second week of school. That is the case the method was made for.
I started with primary sources. Tess's tablet syncs the fifth-grade group chat to our family account, and I read four months of it over a weekend. With the emoji removed, Madison's case comes to seven claims.
| Madison's wording | Strongest version | Finding |
|---|---|---|
| "your bangs look like a butt" | The haircut does not suit Tess's face | Supported. I measured. |
| "soup girl" | Tess's thermos lunch has a noticeable smell | Supported on lentil days |
| "nobody wanted you on their team" | Tess is chosen late in gym | Supported: 17th of 19, by her own account |
| "you laugh like a fart" | Tess's laugh is atypical within her peer group | Inconclusive. Small sample. |
| "your recorder sounds like diarrhea" | The recorder solo contained errors | Supported. I count nine on the video. |
| "tryhard" | Tess's effort is visible and exceeds her results | Partly supported |
| "everyone hates you" | Tess is universally disliked | Not supported. I would put it nearer 60 percent. |
Five of the seven survive in some form. I want to be careful about what that means. Madison's delivery is poor, and she is obviously enjoying herself. But a claim does not become false because the person making it is having a good time.
To check that I had really understood her, I ran an ideological Turing test, which asks whether you can state an opponent's view so well that an observer cannot tell you from the real thing. I wrote three messages in Madison's voice, mixed them with three of her own, and showed all six to Tess. She could not pick mine out. She left the kitchen before I could give her the score.
The next step was to help the two of them find their crux, the underlying point on which their disagreement turns. I sent the table to both girls and, since the chat was already open, to the group. By Monday Madison had begun citing it. One message read, "even her DAD says the bangs." Another was a screenshot of row five with eleven poop emojis beneath it. My wife regards this as my document being used against our daughter. I would describe it as Madison revising her arguments in response to better evidence, which is more than most adults manage.
Erin says a father's job is to be on his daughter's side. I am on her side. I want Tess to have an accurate model of her bangs.
The school counselor, Ms. Abbate, asked us in on Wednesday. I brought copies. She said the meeting was about Madison's behavior. I said I thought we would get further at the object level, and asked whether anyone present wished to defend the solo. Madison's mother said she had found the table "really helpful." Ms. Abbate asked me to put the copies away.
Tess's own contribution has been limited. In the car she said, "Dad, please stop helping," which I have logged as feedback on format.
She now eats lunch in Ms. Abbate's office, and she has not spoken to me since Thursday. I have been trying to steelman this. The strongest version of her position, as far as I can reconstruct it, is that I am her father. I do not yet see what that has to do with the bangs.
Sending the table to the group chat is the part I can't get past methodologically. You published preliminary findings to the population under study. Everything that happened in that chat after Monday is contaminated by your own intervention, including the eleven poop emojis.
It accelerated the search for a crux. Madison engaged with the evidence faster than I expected.
You've evaluated whether individual insults contain accurate information. The question Tess brought you was whether her father would help stop a classmate making her cry. Where does that enter the table?
The table establishes which claims we can responsibly challenge. Asking Madison to withdraw a supported proposition would damage our credibility on the unsupported ones.
The ITT result doesn't mean much with Tess as the judge. She has every reason not to read Madison's messages closely, and a judge who won't engage with the material can't distinguish anything. You'd want a neutral judge, another fifth grader who isn't in the chat.
Steelmanning is something you do to arguments. Most of the table is insults with a factual residue. If you were going to steelman anything you'd steelman Madison's position, which is roughly "Tess is beneath me," and you didn't touch that one.
How is Tess doing now?
She's back at the lunch table. She cut her bangs herself over the weekend, so row one no longer applies.
Nine errors isn't interpretable without a comparison recording from the rest of the class. Please don't ask Tess to obtain this comparison herself.
I can give you the comparison. I teach elementary music. Nine errors in a fifth-grade recorder solo is a perfectly good performance. A lot of the class would have had twenty, and most of them wouldn't have volunteered for a solo in the first place.
Thank you, that's useful. I've revised row five to "Partly supported, within the normal range."
I did something similar with the criticisms my son's teammates had of his goalkeeping last spring and found four of six were supported. Kids are often more accurate than adults give them credit for. It's the delivery that's the problem.
How did you get "nearer 60 percent"? If it's from the group chat, that's a convenience sample drawn from a chat Madison more or less runs.
From the chat, with a downward adjustment for Madison's influence. I agree the sample isn't ideal.
The strongest version of "soup girl" can't just be that soup has a smell. The epithet also carries a claim about whether a thermos lunch warrants social exclusion. You've removed that claim during reconstruction.
madison cooked