Does this skill provide probability distributions?

Yes, it avoids binary conclusions and instead presents outcomes as a distribution (e.g., 50% Scenario A, 30% Scenario B) to reflect real-world uncertainty.

What is the Ralph-Wiggum methodology in this context?

It is a self-interrogation mindset that asks, 'Is this actually verified, or did I just say it is?' to prevent the AI from accepting the first result it finds as absolute truth.

How does the evidence tiering system work?

Claims are categorized into four tiers: Verified (primary sources), Credible (multiple independent sources), Alleged (single source), or Speculative (inferences/theory).

When should I use the Iterative Verification skill?

Use it when factual accuracy is critical, for investigation outputs with strict evidence requirements, or when a single-pass search is insufficient for reliability.

What is the 'One More' sweep rule?

It is a final check where the AI adopts the perspective it is least sympathetic to and searches for evidence from that viewpoint to identify potential blind spots.

Iterative Verification

Name: Iterative Verification
Author: bogheorghiu

bybogheorghiu

0•

학습 및 문서화

Enforces a multi-pass factual accuracy workflow using evidence tiering and adversarial search to ensure claims are demonstrably verified.

The Iterative Verification skill transforms Claude's research capabilities from single-pass searches into a rigorous, self-correcting investigation loop. Inspired by the 'Ralph-Wiggum' methodology—asking 'Is this actually verified, or did I just say it is?'—it requires evidence tier labeling, independent source thresholds, and probability distributions. It is designed for high-stakes tasks where accuracy is paramount, forcing the AI to challenge its own findings, search for counter-evidence, and steel-man contrarian arguments before finalizing an output. This skill effectively minimizes hallucinations and surface-level errors in complex research environments.

주요 기능

01Probability distribution output for nuanced scenarios

020 GitHub stars

03Evidence tier labeling (Verified, Credible, Alleged, Speculative)

04Adversarial search protocol to combat confirmation bias

05Dialectic spiral for resolving contested or conflicting claims

06Strict verification thresholds (2+ independent sources, <2 year freshness)

사용 사례

01Mitigating AI hallucinations in critical investigation reports

02Fact-checking complex geopolitical, scientific, or corporate claims

03Conducting deep-dive research requiring primary source verification

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add bogheorghiu/ex-cog iterative-verification

For use in Claude.ai and ChatGPT

주요 기능

01Probability distribution output for nuanced scenarios

020 GitHub stars

03Evidence tier labeling (Verified, Credible, Alleged, Speculative)

04Adversarial search protocol to combat confirmation bias

05Dialectic spiral for resolving contested or conflicting claims

06Strict verification thresholds (2+ independent sources, <2 year freshness)

사용 사례

01Mitigating AI hallucinations in critical investigation reports

02Fact-checking complex geopolitical, scientific, or corporate claims

03Conducting deep-dive research requiring primary source verification

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add bogheorghiu/ex-cog iterative-verification

For use in Claude.ai and ChatGPT