Steelman / Charitable Interpretation
TL;DR
Steelmanning is the discipline of engaging the strongest version of a position rather than the weakest — the opposite of a strawman. As a graph it first reconstructs the opposing case at its best, in its own best light, then evaluates the original position against that, and only then rules. Reach for it whenever a conclusion’s credibility depends on having taken the other side seriously: a denial, a rejection, a risk dismissed. If the graph can beat the best version, the conclusion is strong; if it cannot, the conclusion was wrong.
Winning against an argument nobody made
A proposal is rejected, and the rejection reads convincingly — until you notice it is arguing against a weak version of the proposal that no one actually put forward. The real case was stronger, and it was never engaged. The decision feels justified because it defeated something; it just defeated the wrong thing. This is the strawman, and it is the most common way confident reasoning goes quietly wrong.
Steelmanning inverts the move. Before the graph rules against a position, it is required to rebuild that position at its strongest — to state the opposing case as well as its best advocate would, and then engage that. A conclusion that survives the strongest version of the opposition is worth trusting. One that only survives a weak version has proven nothing.
An unguided model reaches for the easiest counter. The steelman graph is required to build the hardest one first — and to say so, so a reviewer can see the opposition was met at full strength.
How the graph works, step by step
Generate the position. The graph states the conclusion it is inclined toward — the denial, the approval, the recommendation — as the thing that will have to earn its place.
Generate the strongest opposing case. This is the move that names the strategy. The graph reconstructs the opposite position at its best: not the version easiest to beat, but the one its most capable advocate would put forward. A dedicated node, so the strongest counter cannot be quietly skipped.
Evaluate against the best version. The original position is tested against that strengthened opposition, not against a convenient weak one. This is where a shaky conclusion is exposed — it survived the strawman but not the steelman — and where a sound one earns real credibility.
Arbiter and ruling gate. The arbiter rules on the position as tested against the best opposition, and the gate records both: the conclusion, and the strongest case it had to overcome. A reviewer sees not just what was decided but what it beat.
Where it fits — use cases
1. Denials and rejections that must hold up
When a decision goes against someone, the fastest way for it to fail on appeal is to have argued against a weak version of their case. Steelmanning produces the denial together with the strongest version of the claimant’s position and shows it was answered — which is exactly what turns “we disagree” into “we considered your best argument and here is why it does not carry.”
2. Risk and assumption review
Dismissing a risk is easy when you frame it weakly. Steelmanning forces the graph to state the risk at its most credible before deciding it is acceptable — catching the “we thought about that” failures where a real concern was waved off in a form no one seriously held.
- Steelman vs. strawman
- Engaging the strongest version of the opposing case, not the weakest — the difference between a conclusion that proved something and one that only appeared to.
- Strongest-opposing as a node
- The best counter-argument is built by a dedicated step, so it cannot be quietly skipped in favor of an easy one.
- Survives-the-best test
- A conclusion earns credibility only by holding against the strengthened opposition; the trace records both the ruling and what it overcame.