steelman

Most critique arrives weaker than it should be. A reviewer is being polite, a counter-argument is a strawman that was easy to knock down, or you have already rehearsed the rebuttal. None of that tells you whether your position holds. The steelman skill constructs the strongest version of the opposing case and then locates where your reasoning is actually load-bearing.

What it solves

A position that has only faced weak opposition is untested, not validated. The failure mode is specific and common: the “alternatives considered” that were never serious, the objection you answered because it was the easy one, the decision that felt obvious because nobody argued the other side well. Steel-manning inverts the usual move — instead of hunting for flaws in the opposition, it grants the opposition every reasonable assumption and asks what survives.

The output is not a verdict on who is right. It is a sharper version of the disagreement, which is what lets you answer the part that matters instead of the part that was easiest to answer.

How it works

Six steps, all read-only:

1. Extract the real position — what is being defended is rarely what was stated. “We should use Kafka” usually defends “our consumers need replay”. Steel-manning the wrong position makes everything downstream noise.
2. Find the genuine opposition — the strong contrary, not the convenient one. The strawman is the default failure of this exercise.
3. Build the maximal case — best evidence, best framing, the most competent critic who would plausibly exist. No hedging and no pre-rebutting; a steel man that telegraphs its own weakness is a strawman wearing better clothes.
4. Separate what bites from what is rhetoric — not every strong point is lethal. A steel man that lands everything is flattering the opposition, not testing it.
5. The survival test — what would you have to believe for your position to still stand? This is where the real weakness surfaces, usually as an assumption that was never examined because it never had to be stated.
6. Two buckets — what demands an answer, and what can be conceded without costing the thesis. Conceding the second kind is what makes the remaining position defensible rather than stubborn.

When to use

Reach for it when there is a position being defended and you want it tested: before committing to a technical direction, when a decision feels too easy, or when the opposition so far has been weak. doc-adr invokes the same protocol before an alternative that was a live option gets rejected, so the ADR records why the decision was close rather than a Pros: line nobody can audit later.

Do not reach for it to weigh several options — that is council, which runs five divergent lenses over one decision. And note the boundary that defines this skill: council’s Contrarian attacks an idea, while steelman builds the other side’s case up. Losing that distinction collapses one into the other.