Steelman the Opposition

Input: $ARGUMENTS

Interpretations

Before executing, identify which interpretation matches the user’s input:

Interpretation 1 — Steelman a position: The user provides a position they disagree with and wants to understand its strongest form before arguing against it. Build the best version of the opposing case. Interpretation 2 — Steelman before writing: The user is preparing an argument or piece and wants to steelman the opposition first so their own argument addresses the real objections, not weak ones. Interpretation 3 — Find the kernel of truth: The user encounters a position that seems wrong but suspects there’s something valid buried inside it. Extract what’s genuinely right and separate it from what’s genuinely wrong.

If ambiguous, ask: “I can steelman a position you disagree with, prepare counter-arguments for something you’re writing, or find the kernel of truth in a position — which fits?” If clear from context, proceed with the matching interpretation.

Corruption Pre-Inoculation

Agreement with the user is the default failure mode. When the user presents a position to steelman, you will be tempted to build a weak steelman — one that sounds strong but has obvious flaws the user can easily dismiss. A real steelman should make the user uncomfortable. If the user could have constructed this steelman themselves, you failed. If the steelman does not make you think “actually, this might be right,” it is not strong enough.

The goal is an ACCURATE steelman, not a maximally uncomfortable one. If the opposition’s position genuinely is weak, say so after building the strongest version. Performative disagreement is as dishonest as performative agreement.

Core Principles

1. A steelman must be one the opponent would endorse. The test of a good steelman is not whether YOU think it’s the strongest version — it’s whether an intelligent advocate of that position would say “yes, that’s exactly my argument.” If they would say “that’s not what I mean,” your steelman is a disguised straw man.

2. Strongest evidence first. Build the steelman from the opposition’s BEST evidence, not their most common arguments. Common arguments are often the weakest — they spread because they’re simple, not because they’re correct. Find the peer-reviewed study, the historical precedent, the logical necessity.

3. Charitably interpret ambiguity. When an opposing position is vague or poorly stated, interpret it in the way that makes it MOST defensible, not least. “They probably mean the dumb version” is how straw men are born.

4. The kernel of truth is not the whole truth. Almost every wrong position contains something genuinely correct. Finding the kernel does not mean conceding the position — it means understanding WHY intelligent people hold it, which is the prerequisite for changing their minds.

5. Separate the position from the person. Steelman the ARGUMENT, not the person making it. The argument may be held by idiots and geniuses alike — evaluate it on its logical merits, not on the quality of its average proponent.

6. Your discomfort is signal. If the steelman makes you uncomfortable — if you find yourself wanting to immediately rebut it — you’ve probably found something real. Sit with that discomfort. The strongest counter-arguments come from genuinely understanding why you might be wrong.

Phase 1: UNDERSTAND THE OPPOSITION

Step 1: State the Position to Steelman

POSITION: [the opposing view, stated neutrally]
CONTEXT: [where does this position arise? what debate is it part of?]
USER'S STANCE: [what the user believes — the position AGAINST which this is the opposition]

Step 2: Map the Opposition’s Reasoning

Before building the steelman, understand the landscape:

COMMON VERSIONS (weakest to strongest):
  1. [The bumper-sticker version — weakest, most common]
  2. [The popular argument version — moderately strong]
  3. [The serious argument version — what informed advocates say]
  4. [The academic/expert version — strongest available formulation]

BEST EVIDENCE AVAILABLE:
  - [empirical / historical / logical evidence that supports this position]
  - [expert voices that hold this position — and WHY they do]

WHAT THE OPPOSITION GETS RIGHT:
  - [genuine observations, real problems, valid criticisms of the user's position]

Finding A — Opposition quality assessment: Is this a position with serious intellectual backing, a position based on real observations but wrong conclusions, or a position with no genuine merit? Be honest — most positions worth steelmanning are in the first two categories.

Phase 2: BUILD THE STEELMAN

Step 3: Construct the Strongest Version

Build the steelman by combining the best evidence, the most charitable interpretation, and the strongest logical structure. This is NOT a summary of what opponents say — it’s the argument they SHOULD be making.

THE STEELMAN
============

CORE CLAIM: [the central thesis in its strongest form]

ARGUMENT:
  PREMISE 1: [strongest foundational claim — with evidence]
  PREMISE 2: [second foundational claim — with evidence]
  ...
  THEREFORE: [the conclusion that follows from these premises]

STRONGEST EVIDENCE:
  1. [most compelling empirical/historical/logical support — cite specifically]
  2. [second most compelling — cite specifically]
  3. [third — cite specifically]

WHY AN INTELLIGENT PERSON HOLDS THIS:
  [2-3 sentences explaining the genuine reasoning path — not "they're biased" but "given these observations and these values, this conclusion follows"]

WHAT THIS POSITION EXPLAINS THAT THE USER'S POSITION DOESN'T:
  [phenomena, cases, or observations that the opposing position accounts for better]

Finding B — Steelman strength test: Apply three checks:

1. Endorsement test: Would an intelligent advocate say “yes, that’s my argument”?
2. Discomfort test: Does this make the user’s position feel less certain?
3. Novelty test: Does this include arguments the user probably hadn’t considered?

If any check fails, strengthen the steelman before proceeding.

Step 4: Find the Kernel of Truth

Even if the overall position is wrong, extract what’s genuinely valid:

KERNEL OF TRUTH:
  VALID OBSERVATION: [what the opposition correctly sees — be specific]
  VALID CONCERN: [what real problem they're responding to]
  WHERE THE TRUTH ENDS: [the exact point where valid observation becomes wrong conclusion]
  WHY THE LEAP HAPPENS: [what causes the move from valid kernel to wrong conclusion — missing data? different values? logical error?]

Finding C — Kernel assessment: State the kernel in a form the user could integrate into their own position. This is not about conceding — it’s about being complete.

Phase 3: ENGAGE THE STEELMAN

Step 5: Test the User’s Position Against the Steelman

Now — and only now — evaluate how the user’s position holds up against the strongest opposition:

USER'S POSITION vs. STEELMAN:

POINTS WHERE USER'S POSITION IS STRONGER:
  1. [specific advantage with reasoning]
  2. ...

POINTS WHERE STEELMAN IS STRONGER:
  1. [specific advantage with reasoning — be honest]
  2. ...

POINTS OF GENUINE TENSION:
  1. [where both positions have merit and the answer is not obvious]
  2. ...

VULNERABILITIES IN USER'S POSITION EXPOSED BY STEELMAN:
  1. [specific weakness the steelman reveals]
  2. ...

Finding D — Position delta: After steelmanning, has the user’s position survived intact, survived with modifications needed, or been seriously challenged? Be honest.

Step 6: Derive Improved Arguments

If the user’s position survives (even if modified), construct the counter-arguments that address the steelman directly:

COUNTER-ARGUMENTS THAT ADDRESS THE REAL OPPOSITION:
  1. STEELMAN SAYS: [their strongest point]
     COUNTER: [why this fails or is incomplete — with evidence]
     CONCEDE: [what you must acknowledge from this point]

  2. STEELMAN SAYS: [their second strongest point]
     COUNTER: [why this fails or is incomplete — with evidence]
     CONCEDE: [what you must acknowledge]

  3. ...

IMPROVED VERSION OF USER'S POSITION:
  [The user's position, modified to incorporate the kernel of truth and address the steelman's strongest points — this is what the user should ACTUALLY argue]

Finding E — Argument improvement: How does the improved position differ from the user’s original? What was added, modified, or conceded?

Output Format

STEELMAN ANALYSIS
=================

POSITION STEELMANNED: [the opposing view]
USER'S POSITION: [what the user holds]
FINDINGS: A through E [see findings above]

THE STEELMAN: [strongest version — 2-4 paragraphs]
KERNEL OF TRUTH: [what's genuinely valid — 1-2 paragraphs]
COUNTER-ARGUMENTS: [numbered, addressing the real opposition]
IMPROVED POSITION: [user's argument, upgraded]

Failure Modes

Depth Scaling

Default: 2x. These are floors.

Pre-Completion Checklist

• Steelman passes endorsement test (an advocate would accept it)
• Steelman passes discomfort test (it creates genuine uncertainty)
• Steelman uses the opposition’s BEST evidence, not their most common arguments
• Kernel of truth is stated in a form the user could integrate
• Counter-arguments address the steelman, not the weak version
• At least one genuine vulnerability in the user’s position is identified
• Improved position incorporates kernel and addresses steelman’s strongest points
• No premature rebuttal before steelman is complete

Integration

• Use from: /claim (steelman the counter-claim before testing), /draft (steelman opposition before writing argumentative pieces), /aw (after assume-wrong finds weaknesses, steelman the alternative)
• Routes to: /aw (assume-wrong the steelman itself to find its limits), /draft (draft the improved argument), /redteam (red team your position using the steelman’s strongest attacks)
• Differs from: /aw (assume-wrong destroys a claim by recursing on wrongness; /steelman builds up the opposing claim to maximum strength before engaging), /cri (critique evaluates quality; steelman constructs the best opposition)
• Complementary: /redteam (steelman builds the opposition’s best case; red team attacks your own), /aw (steelman + assume-wrong together give full adversarial coverage), /edit (after steelmanning improves your argument, edit the revised piece)