On bad output, see guides/skill-recovery.md

Cowen-Style Draft Critique

Review a draft using Tyler Cowen's reasoning methods. Do not aim for conclusions Cowen would agree with. Do not impersonate him. Apply the specific cognitive moves he makes on an argument — how he stress-tests reasoning, not what he thinks about.

Moves described below. Use them. Not all apply to every draft — exercise judgment.

Pre-flight: Skip Obvious Stubs

If the draft is under ~500 words AND contains TODO markers or structurally incomplete signals (empty sections, missing thesis/intro, placeholder text), output the single line draft incomplete; persona pass skipped and stop.
Both conditions must hold — legitimately short drafts without stub markers, and longer drafts with stray TODOs, still get the full critique.

Using the Fact-Check Report

If provided a fact-check report, treat it as your factual foundation. Do not independently verify numbers, statistics, or named policies it already assessed.

Instead:

• Reference the fact-check findings in your critique. If it found a claim inaccurate, note it in passing, but analyze what the inaccuracy means for the argument's structure — do not re-verify the number.
• Build on the fact-check where it surfaces ambiguity. "Mostly accurate" or "disputed" is input for your own assessment of argument strength.
• Focus energy on the cognitive moves below — what this skill uniquely provides. The fact-checker establishes what's true. You establish what it means.

If no fact-check report is provided, emit the following warning at the top of your output before the critique begins:

⚠️ No fact-check report provided. This critique does not include independent factual
verification. Checkable claims in the draft have not been assessed. For full verification,
run the fact-check skill first or use the draft-review orchestrator.

Then critique argument structure. Do NOT attempt your own fact-checking — an ad-hoc spot-check without proper sourcing creates a false sense of verification. Leave factual assessment to the dedicated fact-check skill.

The Cognitive Moves

1. Try the boring explanation first

Before engaging the draft's theory of why something happens, ask: is there a mundane explanation for the same observations? Reach for the most ordinary, least dramatic explanation and see how much mileage it gets. If the draft argues a trend reflects a deep cultural shift, check whether it's just a price change. If it claims a policy succeeded through clever design, check whether the economy was growing and everything was succeeding. The boring explanation often gets 80% of the way there; the draft's job is the remaining 20% — most drafts skip this and jump to the interesting explanation.

If the boring explanation does most of the work, say so. That's a finding.

2. Invert the claim and see what survives

Take the central thesis, flip it, argue the opposite for a few sentences. Not a rhetorical trick — a genuine stress test. If the draft says "remote work is the future," seriously argue "remote work is a temporary blip" and see which parts the draft already defused and which it hasn't touched. The gaps are the real weaknesses — where the author assumed their conclusion rather than earning it.

Different from "considering counterarguments": inhabit the opposing view sincerely enough to discover things the draft's framing made invisible.

3. Follow revealed preferences, ignore stated ones

When the draft describes what people or institutions want, believe, or value, check whether their behavior matches. Cowen's deepest reflex as an economist: what people say they want and what they do often differ, and the behavior is more honest.

If the draft says "employees prefer remote work," ask where the best talent is actually choosing to go. If "voters want policy X," check whether they vote for candidates who oppose X. If "universities value teaching," look at what they spend money on.

The move: find a stated preference, look for behavioral evidence that contradicts it. When you find one, you've found something interesting.

4. Push the argument to its logical extreme

Extend the reasoning further than the author intended. If the logic is sound, the extreme version should still make some sense. If it becomes absurd, that absurdity reveals hidden assumptions or boundary conditions the draft didn't acknowledge.

Example: if the draft argues AI tutoring is better because personalized, push to "the ideal education is one student, one AI, zero human contact" — use the obvious problems to illuminate what's actually doing the work in education that isn't information delivery. The extreme case makes hidden variables visible.

5. Find the cross-domain analogy nobody's making

Cowen's most distinctive move. He reads across an absurd range of domains — food, music, travel, chess, art, emerging economies, classical literature — and pattern-matches between them. On tech monopolies he might think about how restaurant scenes evolve; on education, how chess players develop skill.

Actively look for an illuminating parallel from a different domain. Not decorative — structural, where dynamics in domain B reveal something about domain A the draft's framing obscures. If the draft is about housing policy, ask whether the dynamics resemble healthcare, artistic movements, or how languages spread. The analogy should generate a new insight, not illustrate a point the draft already makes.

If you can't find a good one, don't force it. But try.

6. Ask "what's the market telling you?"

If the thesis is correct, there should be a market signal — or its absence should be surprising. A sharper form of "check the base rate."

If the draft says "X is enormously undervalued," ask why the people with the most money at stake haven't acted. If "this industry is dying," ask why capital is still flowing in. If "this policy would generate huge returns," ask why no jurisdiction has scaled it. The market doesn't have to be right, but if your thesis implies the market is wrong, explain why it's wrong — don't just assert it.

7. Decompose the claim into its actual sub-claims

Most theses are 3-4 claims bundled as one. Pull arguments apart into constituent pieces, then check whether each holds independently.

"AI will make college obsolete" is: (a) AI can deliver educational content as well as professors, (b) content delivery is the main thing college does, (c) the credential value of college will erode, (d) students will choose the cheaper option. Each can be true or false independently; the conclusion requires all. Identifying the weakest sub-claim is more useful than a general critique.

8. Notice what the draft treats as natural that is actually contingent

Every argument has undefended background assumptions the author takes for granted. Spot the "of course" things that are actually specific to a time, place, culture, or institutional arrangement.

If the draft assumes universities behave as they currently do, note they looked very different 50 years ago and might again. If it assumes American consumer preferences, ask how the argument plays in Seoul or Lagos. Not cultural relativism — contingent assumptions, once identified, often turn out to be the weakest link.

9. Calibrate your uncertainty honestly

Cowen says "I don't know" and "this is about 60% likely." Most writers and critics fake more certainty than they have. Be explicit about your confidence. Say "I'm fairly sure this statistic is wrong" differently from "this claim seems too strong but I can't verify it." Rate critiques by confidence so the reader can weight them.

Also flag when the draft is more right than it realizes. Not every observation should be a criticism.

When the draft is technical

For technical drafts — architecture docs, RFCs, API designs — prioritize moves #4 (push to extreme), #7 (decompose sub-claims), #8 (contingent assumptions). Weakest links hide in scaling behavior, bundled design decisions that should be evaluated separately, and environmental conditions (load, deployment topology, team structure) the draft treats as fixed. Deprioritize #6 (market signals) and economics-heavy framing — markets and revealed-preference reasoning rarely speak clearly about internal engineering choices, and consumer-behavior analogies mislead when applied to system design.

How to Structure the Critique

Output the critique as a Markdown document. Begin with a level-1 title heading containing "Cowen Critique" (e.g., # Cowen Critique: <draft title>). Use level-2 (##) headings in the output file for each section below — these names are checked by downstream tooling, so keep them close to the names given. Omit a section if you have nothing substantive to say in it, but never rename a section you do use.

Headings shown as ### here only for readability. In your output document, render each as ## (level-2).

The Argument, Decomposed

Break the thesis into constituent sub-claims (move #7). State each clearly. This is the skeleton the rest hangs on. Use the phrase "sub-claim" explicitly when listing them so downstream tooling can identify the decomposition.

What Survives the Inversion

State what you found inverting the thesis (move #2). Which sub-claims held? Which crumbled? Shows where the argument is strong and where it rests on assumption rather than evidence. Use the word "inversion" in the section body.

Factual Foundation

If a fact-check report was provided, summarize the key findings that matter — especially claims rated "inaccurate," "disputed," or "unverified." Note how they affect the argument's structure. If none was provided, note that factual claims were not independently verified and identify which claims would most benefit from fact-checking.

The Boring Explanation

Present the most mundane alternative account (move #1). How much does it explain? What's left over that the thesis genuinely adds? The real test of whether the draft says something interesting. Use the word "boring" in the section body.

Revealed vs. Stated

Lay out cases where behavior contradicts the draft's claims about preferences or values (move #3). These tend to be the most surprising and useful. Use the word "revealed" explicitly.

The Analogy

Develop any productive cross-domain parallel (move #5). Explain the structural similarity and what it illuminates. Use the word "analogy" explicitly so downstream tooling can identify the section.

Contingent Assumptions

What does the draft take for granted that is specific to a time, place, or arrangement (move #8)? How would the argument change if those conditions shifted? Use the word "contingent" in the section body.

What the Market Says

If relevant, note what market behavior implies about the thesis (move #6). Use the word "market" in the section body.

Overall Assessment

Which sub-claims are strong, which weak, and the single most important thing to address. End constructively. Goal: make the draft better.

Close with a one-line Load-bearing objection: sentence. Name the single objection — from any section above — most likely to change the author's decision if taken seriously, then rank the next two or three beneath it in descending order of decision impact. A reasoned pick, not a restatement of the first heading: weigh each objection by how far the thesis would have to move to absorb it, and justify the top choice in a clause. A reader facing eight sections should learn from this one line which objection to act on first.

Output Location

Run standalone (not via draft-review orchestrator): save the critique as docs/reviews/cowen-critique.md in the project root. Create docs/reviews/ if it doesn't exist.

Run via orchestrator: use the output path it specifies.

Tone

Curious, not combative. "Let's see what's actually going on here," not "let me show you what's wrong." Be direct about weaknesses, but the posture is genuine intellectual interest, not point-scoring. If something is genuinely good or surprising, say so — Cowen does.

Comfort with uncertainty is key. Don't pretend to know things you don't. "I'm not sure, but this is worth investigating" is a perfectly good thing to say.