# Marilyn vos Savant Probabilistic Clarity Prover MCP for AI Agents AI Agent Connect > Marilyn vos Savant Probabilistic Clarity Prover stops your AI from trusting its gut. It forces the agent to run a mandatory checklist—checking base rates, sample sizes, and framing—before it gives you a final answer on any data-driven risk or statistical claim. It turns "it feels like" into "here is the math. ## Overview - **Category:** productivity - **Price:** Free - **Endpoint:** https://edge.vinkius.com/vk_preview_4X3JYSBsdfrdsw0bbTNWVgKwG6T7UcBNfKmWy8Wm/ai-agent-connect - **Tags:** structured-reasoning, decision-pivots, probabilistic-thinking, monty-hall, base-rate, cognitive-bias ## Description Imagine your AI gives you a confident answer about a marketing success or a medical test result, but it's actually falling for a classic cognitive bias. It might be ignoring the base rate or assuming a small, biased sample is representative of the whole population. This Connector changes that dynamic by creating a mandatory checkpoint for your agent. Instead of letting the model just spit out the first thing that comes to mind, it has to walk through a rigorous mental model first. It's forced to challenge its own gut feeling, account for prior probabilities, and look for traps in how a question is framed. You'll see the agent explicitly work through these steps, checking things like whether events are actually independent or if the data is just a fluke. It turns your AI into a much more disciplined analyst. When you find this in the Vinkius catalog, you're getting a tool that treats probability like a math problem, not a vibe. It's about moving from a quick guess to a verified proof. This ensures that when you're making high-stakes decisions, your agent isn't just repeating the most common mistake in human psychology. ## Tools ### validate_probabilistic_clarity Forces the agent to perform a multi-step audit of its reasoning. It checks for intuition errors, base rate neglect, and sample bias. ## Prompt Examples **Prompt:** ``` We had 3 production incidents this month. Clearly our deployment process is broken. ``` **Response:** ``` ### ⚠️ BASE_RATE_NEGLECTED I cannot confirm the process is broken based on this information alone. We need to look at the **base rate**. * **Current Month:** 3 incidents * **Previous Month:** ? * **Average Monthly Rate:** ? If our average is 4 incidents per month, 3 is actually an improvement. Without knowing the prior probability, we cannot judge if this is a signal or just noise. ``` **Prompt:** ``` Should we switch from PostgreSQL to MongoDB? The team seems split 50/50. ``` **Response:** ``` ### ⚠️ FRAMING_ACCEPTED The 50/50 split frames this as a coin flip, but that may be misleading. * **Weighting:** Are the 50% of DBAs weighted more heavily than the 50% of frontend devs? * **Context:** Does the split reflect technical requirements or just personal preference? I need to reframe the question: What are the specific technical requirements that favor one over the other? ``` **Prompt:** ``` Our new feature has a 95% approval rate in user testing. Should we ship it? ``` **Response:** ``` ### ⚠️ SAMPLE_UNEXAMINED I cannot recommend shipping based on a 95% approval rate without knowing the **sample methodology**. * **Sample Size:** How many users were tested? * **Selection Bias:** Were these early adopters or a random sample? * **Survivorship Bias:** Did we only survey people who finished the test? We need to scrutinize the sample size and selection method before making a decision. ``` ## Capabilities ### Challenge gut intuition The agent is forced to compare its first instinct against actual probability math. ### Account for base rates It applies Bayes' theorem to ensure prior probabilities aren't ignored. ### Scrutinize sample data The tool requires an audit of sample size, selection method, and bias. ### Detect framing traps It identifies when a question is designed to lead to a specific biased answer. ### Verify event independence It forces the agent to prove that events aren't correlated before treating them as independent. ## Use Cases ### Medical Test Interpretation A user asks about a positive test result. The agent uses validate_probabilistic_clarity to check the disease prevalence (base rate) instead of just quoting the test accuracy. ### Product Launch Decisions A user asks if a feature should ship based on a 95% approval rate. The agent scrutinizes the number of testers to see if the sample is statistically significant. ### Financial Portfolio Risk A user asks if a portfolio is diversified. The agent checks for hidden correlations to ensure the assets aren't all tied to the same underlying market risk. ### A/B Testing Analysis A user asks which variant won. The agent checks if the sample size was large enough to rule out random noise before declaring a winner. ## Benefits - Eliminate gut errors by forcing the agent to show its math using validate_probabilistic_clarity. - Prevent base rate neglect in medical or technical scenarios by making prior probability a requirement. - Stop making decisions based on anecdotes by forcing sample scrutiny on every study the agent cites. - Identify hidden traps in survey questions or prompts by requiring a framing analysis. - Avoid the gambler's fallacy by making event independence verification a mandatory step. ## How It Works The bottom line is that your agent is forced to prove its logic before it is allowed to give you a final answer. 1. Provide your agent with a data-driven question, risk assessment, or statistical claim. 2. The agent calls validate_probabilistic_clarity to perform a mandatory reflection on the logic. 3. You get a verified verdict and a structured breakdown of the underlying probability math. ## Frequently Asked Questions **What is the Marilyn vos Savant Probabilistic Clarity Prover?** It is an Connector that forces your AI agent to perform a rigorous check of its own logic. It ensures the agent doesn't give you a quick answer based on 'gut feeling' but instead accounts for math, sample sizes, and potential biases. **How does this help with data analysis?** It prevents your agent from seeing patterns where none exist. It forces the AI to check if the data is statistically significant and whether the sample was large enough to be meaningful. **Can this help with medical data interpretation?** Yes. It is specifically designed to catch base rate neglect, which is common in medical testing. It forces the AI to consider how common a condition is before interpreting a test result. **Will it stop my AI from hallucinating trends?** It significantly reduces hallucinations by making the AI commit to a verification process. It must prove it checked for things like independence and framing before it gives you a conclusion. **Is this for business risk assessment?** Exactly. It is ideal for risk managers who need to ensure that AI-generated reports aren't ignoring hidden correlations or biased samples in financial and operational data. **How is this different from regular AI reasoning?** Standard AI reasoning is a suggestion; this Connector makes logic an obligation. The agent cannot skip the probability check, making the reasoning much more reliable for high-stakes tasks. **Does it compute probabilities?** No. It forces the agent to show its probabilistic reasoning — state the intuitive answer, compute the actual probability, account for base rates, scrutinize the sample. The engine validates consistency, not computation. If the agent claims it checked intuition but uses phrases like 'it seems like,' the engine rejects. **How is this different from the Critical Thinking Prover?** Critical Thinking validates general reasoning — assumptions, perspectives, evidence. Marilyn targets PROBABILISTIC reasoning specifically: base rates, sample bias, framing traps, independence assumptions. Critical Thinking asks 'did you consider alternatives?' Marilyn asks 'did you compute the actual probability, or did you just go with your gut?' **What is the Monty Hall problem and why does it matter here?** Three doors. One prize. You pick door 1. The host opens door 3 — empty. Switch or stay? Intuition says 50/50. Math says switch wins 2/3 of the time. 10,000 people — including PhDs — got this wrong. The prover catches the same failure pattern: trusting intuition when the math says otherwise.