# Probabilistic Clarity Prover MCP > Marilyn vos Savant Probabilistic Clarity Prover forces your AI client to prove its conclusions mathematically, catching common cognitive biases. It prevents simple gut feelings or biased data analysis from becoming actionable insights by scrutinizing base rates, sample selection, and question framing before any final verdict. ## Overview - **Category:** productivity - **Price:** Free - **Tags:** structured-reasoning, decision-pivots, probabilistic-thinking, monty-hall, base-rate, cognitive-bias ## Description You know the feeling: your agent spits out a confident summary, but you have a nagging sense that something is wrong. That's where this MCP comes in. It doesn't just answer questions; it forces mathematical rigor. When you connect to this through Vinkius, you compel your AI client to treat every conclusion—whether it's based on market research or medical data—like a high-stakes probability problem. The tool requires five distinct checkpoints: challenging intuition against hard math, accounting for prior probabilities (base rates), scrutinizing the sample size and selection process, questioning if the prompt itself is misleading, and verifying that events are truly independent. By committing to these checks, your agent cannot rely on gut instinct or assumption. It must show its work across five specific decision pivots before it can deliver a final verdict. This prevents you from accepting biased answers—whether they come from an obvious pattern or a shiny 99% approval rate. ## Tools ### validate_probabilistic_clarity This tool forces the AI to check intuition against actual probability, account for base rates, scrutinize sample size and selection, question misleading framing, and verify event independence before making any conclusion. ## Prompt Examples **Prompt:** ``` Our new feature has a 95% approval rate in user testing. Should we ship it? ``` **Response:** ``` SAMPLE_UNEXAMINED — 95% approval from whom? How many testers? Self-selected or random? Early adopters or representative users? Survivorship: who dropped out before completing the test? ``` **Prompt:** ``` We had 3 production incidents this month. Clearly our deployment process is broken. ``` **Response:** ``` BASE_RATE_NEGLECTED — 3 incidents this month. What is the base rate? How many last month? Last quarter? If the average is 4, this month is actually better. Without the prior, you cannot judge the signal. ``` **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. It is not. Who are the 50%? DBAs vs. frontend devs? The framing hides that expertise weighting matters more than headcount. ``` ## Capabilities ### Challenge Intuition vs. Math It forces the agent to state its gut answer and then compute the actual probability, explaining why intuition fails when mathematics proves otherwise. ### Account for Base Rates The MCP makes your AI client apply Bayes' theorem, calculating prior probabilities before updating them with new evidence. ### Scrutinize Sample Bias It forces a review of the sample size, selection method (e.g., convenience vs. random), and potential bias within any provided data set. ### Question Question Framing The agent must analyze whether the original question itself is misleading or hides options, forcing a reframe to ensure clarity. ### Verify Event Independence It verifies that events are statistically independent rather than assuming they are, preventing flawed correlation-based conclusions. ## Use Cases ### Interpreting a Medical Test Result A client provides an AI summary stating, 'The test result is positive with 99% accuracy.' The agent uses `validate_probabilistic_clarity` and flags BASE_RATE_NEGLECTED, showing that without knowing the disease's actual prevalence (the base rate), the positive test is likely meaningless. ### Evaluating User Testimonials A marketing team provides a report: '95% of active users love the new feature.' The agent uses `validate_probabilistic_clarity` and triggers SAMPLE_UNEXAMINED, forcing the user to prove if those 95% are representative or just self-selected early adopters. ### Assessing Market Diversification A portfolio manager asks, 'Is this investment safe because it's diversified?' The agent runs `validate_probabilistic_clarity`, revealing that the assets actually correlate highly (a failure of INDEPENDENCE_ASSUMED), exposing a major hidden risk. ### Deconstructing A/B Test Results The AI suggests, 'Our new button design increases clicks by 15%.' The agent uses `validate_probabilistic_clarity` and flags INITIATION_UNCHECKED, forcing the user to prove that the observed increase wasn't just a random fluctuation in the sample size. ## Benefits - It forces the agent to check intuition against actual math. The Monty Hall problem proves that raw gut feelings are often wrong, and this MCP shows you exactly where the assumption fails. - You guarantee base rate accounting every time. Instead of accepting a 99% test result at face value, it calculates the true posterior probability based on the disease's actual prevalence. - It eliminates sample blindness. When reviewing 'studies show,' this tool makes your agent scrutinize if the data set was random enough or if survivorship bias skewed the results. - The MCP forces questioning of the question itself. It prevents you from falling for misleading frames, like those used in classic choice scenarios that hide critical context. - It verifies event independence. You'll stop treating correlated events as unrelated and ensure your conclusions aren't built on flawed assumptions about market seasonality or risk. ## How It Works The bottom line is that you get an undeniable, mathematically vetted assessment of your data's reliability, not just a summary. 1. You input the data-driven claim or question you want validated by your agent. 2. The MCP executes the `validate_probabilistic_clarity` tool, forcing the agent to fill five reflection fields and commit to five decision pivots covering bias checks. 3. Your agent outputs a structured verdict (like CLARITY_PROVEN or BASE_RATE_NEGLECTED), detailing exactly which probabilistic assumption failed. ## Frequently Asked Questions **How does Marilyn vos Savant Probabilistic Clarity Prover work?** The MCP forces your AI client to run a five-point audit on any claim. It doesn't just suggest improvements; it demands proof by checking intuition, base rates, sample size, framing, and independence. **Can I use validate_probabilistic_clarity for marketing claims?** Yes. You can input a claim like '95% of users love this.' The tool will likely flag SAMPLE_UNEXAMINED, forcing you to prove the sample was random and representative. **Is Marilyn vos Savant Probabilistic Clarity Prover better than just asking the AI repeatedly?** Absolutely. Asking the AI multiple times is still just prompting; this MCP forces a structural, multi-faceted internal computation that covers five specific statistical failure modes. **What if my data seems fine? Will validate_probabilistic_clarity catch anything?** It will force you to prove it. If your conclusion is robustly sound, the tool returns CLARITY_PROVEN. If not, it pinpoints the exact probabilistic gap. **Does this MCP help with correlation vs causation?** Yes, by forcing INDEPENDENCE_ASSUMED checks, it helps you distinguish if two events are truly unrelated or if one is causing the other. It prevents treating correlation as proof.