Steel-Manning Workbench
How It Works
Submit a written argument. Confirm what the system extracted. Receive a structured analysis of where your argument will break — and what to do about it.
Submit your document
Paste or upload any written argument — a policy brief, an op-ed, an analytical report, a position paper.
Confirm what was extracted
Before any analysis runs, the system shows you what it found: your primary claim, the assumptions your argument depends on, the domain, and your likely adversarial reader. You confirm, correct, or add to it. The analysis uses your inputs — not the system's best guess.
The analysis runs
The system maps your argument, identifies where it is weakest, generates the strongest opposing case, and pairs each high-severity threat with a specific hardening recommendation. An independent model then reviews the steel man before it ships. If it isn't strong enough, the system retries. If it still can't pass, then you get a plain failure response and pay nothing.
Receive a structured deliverable
A four-part output rendered in-browser and downloadable as PDF. Every element shows its basis — sourced, inferred, or system-estimated. Nothing is unanchored.
What you receive
A single structured document — rendered in-browser, downloadable as PDF.
Argument Map
Your primary claim, supporting claims, load-bearing assumptions ranked by severity, and strongest evidence. Every element carries its epistemic basis.
Vulnerability Map
Where the argument is weakest — quality failures and fallacies, filtered for your domain and ranked by severity.
The Steel Man
The strongest possible opposing argument, written as a coherent case — not a list of objections. It exploits your highest-severity vulnerabilities and includes arguments your adversary hasn't made yet. The measure: you should find it uncomfortable to read.
Threat Assessment & Hardening
Each high-severity threat paired with a specific recommendation: what assumption is being exploited, how the steel man uses it, and what you need to do to address it.
The methodology
The pipeline encodes established frameworks from argumentation theory, applied consistently every time. Argument structure follows Toulmin (1958). Vulnerability assessment combines the RSA framework (Johnson & Blair, 1977) with the MAFALDA fallacy taxonomy (Helwe et al., 2024). Steel man generation follows Dennett's Rules (Rapoport-derived, 2013). The evaluator that gates the steel man runs as a structurally independent model call — it cannot see the generation prompt. The result is a validated output, based on rigorous methodology, and all without you having to be an expert in argumentation.
Ready to pressure-test your argument?
Join the early access list. First users get a free trial — no subscription required.
Get Early Access