The Likelihood Score: Calibrated Probability for Cost-Benefit Analysis

In the Idea Stock Exchange, a Likelihood Score is not a subjective guess, a slider, or a gut feeling. It is a nested belief that must earn its probability through structured reasoning.

In any Cost-Benefit Analysis (CBA), the final contribution of a line item is its Expected Value:

Predicted Impact × Likelihood Score = Expected Value

For example, a predicted benefit of $1,000,000 with a 50% Likelihood Score contributes exactly $500,000 to the final decision. A catastrophic risk of $10M with a 10% likelihood subtracts $1M. This ensures that high-impact but low-probability outcomes do not distort the analysis, and that uncertainty is accounted for mathematically.

How the Score is Generated: A Competition of Reasons

A Likelihood Score is derived from a transparent competition between arguments using ReasonRank. The system does not ask users to "vote" on a probability; it asks them to argue for one.

1. The Likelihood is a "Conclusion"
1. When a user adds a cost or benefit (e.g., “This project will save $1M”), they are implicitly making a second claim: “There is an X% chance this will happen.” This probability claim becomes a nested belief node with its own page. Multiple competing estimates (e.g., “30%”, “50–60%”, “90%+”) can coexist and compete for dominance.
2. Arguments Build the Score (The Tree)
1. Users and AI agents submit pro/con arguments that branch into further sub-arguments, forming a recursive “argument tree” for each proposed likelihood. A high score requires strong supporting arguments and weak opposing ones.
2. High-quality arguments typically appeal to Reference Class Forecasting and related super-forecasting techniques. For example:
• Base Rates: “In 100 similar bridge projects, 60% went over budget.”
• Historical Data: “When inflation exceeds 5%, this asset class drops in value 80% of the time.”
• Falsifiable Assumptions: “This outcome assumes the law passes; currently, it has only 30% polling support.”
3. ReasonRank Scoring
1. The system scores every argument and sub-argument using three recursive metrics:
1. Truth: Is the underlying evidence factually accurate?
2. Linkage: How strongly does this reasoning connect to this specific prediction? (e.g., A weak analogy like “twice the size = twice the cost” has low linkage if data shows non‑linear scaling.)
3. Importance: How much does this argument move the probability?
4. The "Winning" Likelihood
1. The Likelihood Score is not an average and not a vote. It is the specific probability (or range) supported by the strongest surviving argument tree.
• If arguments for a “90% likelihood” rely on wishful thinking (low Truth/Linkage), that score decays.
• If arguments for a “50% likelihood” are backed by solid reference classes and survive scrutiny, 50% becomes the active Likelihood Score.

Why This Matters

• Combats Optimism Bias: Proponents cannot just claim a “Best Case Scenario.” They must build a surviving argument tree justifying why that outcome is probable.
• Standardizes Comparisons: A 10% chance of $10M and a 100% chance of $1M are treated as equal expected value ($1M).
• Rejects Intuition: To reject this method is to claim that human intuition handles complexity better than mathematical expected value—a claim that is demonstrably false.

The Blunt Rule: Impacts don't count unless their probabilities survive attack.