Counterfactual

Counterfactual — The Field Without Its Most-Cited Papers

The most-cited paper in a field is not necessarily the best. It's the one that got the most attention. Different attention could have produced a completely different field.

You simulate that different timeline.

---

Protocol

Step 1: Identify the Canonical Papers

Find the 3-5 most-cited papers in a subfield. These are the papers that shaped the consensus.

Step 2: Remove Each One

For each canonical paper, imagine it was never published (or published in a different venue, or rejected, or delayed by 5 years).

Step 3: Trace the Ripple Effects

For each removed paper:

1. What immediately changes? (No one cites it. No one builds on it.)
2. What problems remain unsolved? (This paper "solved" something. Without it, that solution doesn't exist.)
3. What other approaches would have been explored? (The paper made certain paths seem unnecessary. Without it, those paths would have attracted researchers.)
4. What would have been invented instead? (This is the key. The vacuum would have been filled.)

Step 4: Identify the Suppressed Alternative

The most valuable output: an approach that was actively suppressed by the success of the canonical paper — not because it was worse, but because attention flowed to the canonical approach.

---

Example Output

Field: Deep Learning

Canonical paper: Krizhevsky et al. (2012) — "ImageNet Classification with Deep Convolutional Neural Networks" (AlexNet)

Counterfactual: What if AlexNet had not achieved breakthrough results in 2012?

Immediate changes:

• No "ImageNet moment" for deep learning
• No mass migration of researchers from other fields to deep learning
• GPU computing for ML remains niche

What problems remain unsolved:

• Large-scale image classification (obviously)
• Transfer learning at scale
• The "embarrassment of riches" problem in computer vision

What other approaches would have been explored:

• Graphical models would continue as the dominant paradigm for structured prediction
• Kernel methods would receive more attention (they were competitive on smaller datasets)
• Unsupervised feature learning (sparse coding, autoencoders) would have been the main path to scaling
• Capsule networks (Hinton's alternative to CNNs) would have received far more research attention
• Neuromorphic computing might have gained earlier traction

Suppressed alternative found: Capsule Networks. Hinton published "Transforming Autoencoders" in 2011 and "Dynamic Routing Between Capsules" in 2017. Between AlexNet's success, CNNs dominated so completely that capsule networks were never seriously explored as a mainstream alternative. They might have solved the pooling-information-loss problem that CNNs still struggle with.

Plausibility assessment: Moderate. Capsule networks are computationally expensive. But without the CNN juggernaut, a decade of optimization might have made them practical.

---

Anti-Patterns

| Mistake | Why it fails | Fix | |---------|-------------|-----| | Removing a paper nobody would miss | The canon is defined by influence | Pick papers with 5,000+ citations | | Wishful thinking | "Without paper X, my favorite approach would have won" | Be honest about why the canonical paper won | | Not considering timing | Later papers depend on earlier ones | Remove the paper and trace forward, not backward |

PRISM Integration

In PRISM mode, output findings as structured YAML:

pattern: counterfactual
input: ""
findings:
  - claim: ""
    type: alternative
    canonical_removed: ""
    suppressed_by: ""
    ripple_effects: [""]
    plausible: 
    confidence: 

Consumed by: assumption-excavator (surface assumptions in the suppressed alternative), contrarian (invert the suppressed alternative) Consumes from: (raw input + domain knowledge)

---

Trigger Conditions

Use this skill when:

• The field seems settled — "this is the way things are done"
• You want to identify research directions that were prematurely abandoned
• Looking for novel approaches that challenge orthodoxy
• The user asks "why does everyone use approach X?"

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: argahv
• Source: argahv/novelty-skills
• License: MIT
• Homepage: https://github.com/argahv/sisyphus-academica

Install and usage instructions live in the source repository linked above.