Non-Consensus Thinking

SKILL.md

---
name: non-consensus-thinking
description: "Activate when: user says 'everyone agrees on this,' 'the obvious move is X,' 'why go against the grain?'; user is entering a crowded market where the right strategy feels obvious; user has an early signal conflicting with the mainstream narrative; user is making a high-stakes allocation where popular choice and correct choice may diverge.
  Do NOT activate when: user just wants to be different with no specific edge to audit; context is purely creative where originality is the goal rather than competitive decision-making."
---

# Non-Consensus Thinking

## Overview

Every market — capital, talent, customers, ideas — prices the consensus view into its current state. By the time an idea is mainstream, its excess return has been arbitraged away. Non-consensus thinking is a disciplined audit of where the crowd's belief might be wrong *and* whether you hold a specific, articulable advantage that makes the minority position actually correct, not just different.

Compose with: **[second-order-thinking]** first (trace downstream consequences); **[confirmation-bias]** audit after (check you haven't built a new blind minority consensus); **[first-principles]** instead when rebuilding from bedrock evidence, not auditing mispricing.

## When to Use

- Entering a market where the "right" strategy feels obvious to most participants.
- Making an allocation decision (capital, time, hiring) where popular and correct may diverge.
- Any situation where "everyone knows that..." appears — consensus may be unexamined.

**When NOT to use:** Consensus is correct and well-evidenced; non-consensus position requires information you can't obtain; time horizon too short to vindicate the position; stakes of being wrong are catastrophic and irreversible.

## Coaching Novices (Adaptive Front Door)

- **Engine mode:** user has a concrete case → run The Process directly.
- **Coach mode:** user is unfamiliar → guide step by step.

In Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output only that step's question, then stop.

1. **What it is:** "What does everyone believe, and why might they be wrong — and do I have a real edge that makes the minority view correct?"
2. **Check fit:** "Tell me your decision. Is there a dominant answer most people in your field would give? Do you have specific conflicting evidence?"
3. **Elicit their real case:** "State the consensus in one sentence. What specific information do you have that most participants don't?"
> **[WAIT — do not advance until user responds]**
4. **Run The Process** one step at a time with their input. Do not jump to asymmetry until edge audit is complete.
> **[WAIT — do not advance until user responds]**
5. **Close:** "If your edge is real and consensus is wrong, what becomes possible? Name that outcome specifically."
> **[WAIT — do not advance until user responds]**

## The Process

**Stop-rule:** If at Step 4 you cannot name a specific information or analytical advantage — not a feeling — stop. Return to Step 3 and gather evidence, or accept the consensus provisionally.

1. **State the consensus precisely.** "Most participants in [domain] believe [X]." Cite ≥2 observable sources. Gate: not a straw man.
2. **Audit why the consensus holds.** Mechanism: path dependence, herding, incentive misalignment, data limitation. Gate: name the specific force.
3. **Identify where it could be wrong.** "The consensus would be wrong if [Y] is true." Name ≥1 observable evidence pointing toward Y. Gate: specific and testable.
4. **Audit your edge.** (a) information others lack, (b) analytical framework, (c) time horizon, (d) structural advantage. Gate: name it and explain why it's real, not assumed.
5. **Size the asymmetry.** Four-cell outcome table. Worth pursuing only if Z >> X and W is survivable.
6. **Decide and record.** Commit to a position. Set a pre-committed update trigger with a review date.

### Output: Non-Consensus Audit

```
Domain / Decision:   Date:
1. Consensus: "Most participants in [domain] believe [X]." Evidence: Source 1 / Source 2
2. Mechanism: [path dependence / herding / incentive misalignment / data limitation]
3. Falsifying condition: "Consensus wrong if [Y]." Supporting evidence:
4. My Edge — Type: [information / analytical / time horizon / structural]. Why real:
5. Asymmetry: Follow consensus+right= | Follow consensus+wrong= | Non-consensus+right= | Non-consensus+wrong=
   Conclusion: Z >> X and W survivable? [Yes / No / Conditional]
6. Position taken: | Update trigger: | Review date:
```

*→ Method in Action: [Semmelweis and Childbed Fever (1847)](examples/semmelweis-and-childbed-fever-1847.md)*

## Contrarian Packs

- **Venture:** underweighted signals (regulatory-hostile markets, unsexy verticals, non-pedigreed founders). Edge: relationships or time horizon institutions can't have.
- **Product/Market Entry:** segments avoided for being hard or beneath incumbent attention. Edge: cost structure or willingness to serve incumbents lack.
- **Research:** anomalies outside the dominant paradigm (Kuhn's "puzzles"). Edge: cross-disciplinary methodology or data access.
- **Career:** talent supply thin relative to unprice future demand. Edge: genuine comparative advantage in the emerging area.

## Applying It Well

1. **Name the consensus before you reject it.** Vague disagreement is not a position.
2. **Separate "I disagree" from "I have an edge."** Disagreement is a precondition, not a conclusion.
3. **Quantify the asymmetry** — write the four-cell table before deciding, not after.
4. **Set a pre-committed update trigger** before you are emotionally invested.
5. **Distinguish herding from convergence** — the mechanism audit (Step 2) is the diagnostic.
6. **逻辑长一步,对手少一半.** Non-consensus is often just consensus reasoned one step further.
7. **Size the bet to survive the W outcome** (non-consensus, wrong).

*→ Primary sources: [references/sources.md](references/sources.md)*

## Common Rationalizations

**[D] = designed upfront | [O] = observed in real use. [O] entries are more valuable.**

| Fake move | Reality |
|---|---|
| [D] "I disagree, therefore I'm thinking non-consensually." | Disagreement is the starting condition. Without an edge audit, this is just preference. |
| [D] "The minority view is more creative, so it's probably right." | Minority views are not more likely correct by virtue of being minority. |
| [D] "Everyone said X was wrong before it worked — so my bet is the same." | Survivorship bias. Many non-consensus positions are simply wrong. |
| [D] "I just have a gut feeling this is wrong." | Intuition is a hypothesis, not an edge. Validate against observable evidence. |
| [D] "The consensus is maintained by vested interests, so it must be wrong." | Vested interests can sustain a correct consensus just as easily as an incorrect one. |
| [D] "My non-consensus view has been right before, so my edge is validated." | Past vindication is a past edge, not a current one. Re-run the audit. |
| *→ Add [O] entries here after each real use — paste the actual failure pattern* | *What went wrong and why* |

## Red Flags

- Analyst cannot state the consensus in one precise sentence before arguing against it.
- "Non-consensus" position is widely shared within the analyst's own social or professional circle.
- Edge audit absent — analyst articulates why they disagree but not why they are better positioned.
- Asymmetry asserted ("this could be huge") but not mapped into the four-cell table.
- No update trigger set; analyst responds to contradicting evidence by entrenching.

## Verification

- [ ] Consensus stated in one sentence, evidenced by ≥2 observable sources.
- [ ] Mechanism sustaining consensus named (herding, path dependence, incentive misalignment, data limitation).
- [ ] Falsifying condition is specific and testable.
- [ ] Edge named and validated — not assumed.
- [ ] Four-cell asymmetry table complete with estimated magnitudes.
- [ ] Pre-committed update trigger recorded with a specific review date.
- [ ] The "wrong" case is survivable given the position size taken.

---
*Part of **deciqAI Knowledge Skills** — open-source thinking skills that make rigor executable for AI agents. Built by deciqAI · https://deciqai.com · Contributions welcome — see the template at the repo root.*