Research
From AI Guessing to Data Science: How Clerk Calculates Odds
March 10, 2026 · 5 min read
Instead of predicting from vibes, Clerk computes "odds" — statistically-backed predictions with confidence scores derived from collected data.
Most AI research ends with a confident guess. Clerk ends with an odds calculation.
After 100+ workers return structured data, the Analyst phase loads everything into DuckDB — an in-process analytical engine that can crunch thousands of JSON files in seconds.
From that data, Clerk computes correlation matrices (how tightly sub-niches move together), trend vectors (which directions are increasing or decreasing), and deduplication statistics (how many claims were genuinely unique).
The result is an odds calculation: a prediction like "Layer 2 adoption will exceed Layer 1 by 2028", paired with a confidence score of 0.82 and the basis — "based on 2017 whitepaper trends vs 2024 implementation data".
That basis is the difference. Every odds statement points back to the data that produced it, which is why Clerk research is auditable, reproducible, and genuinely quantitative.
It is not AI guessing. It is data science.