AI Recommendation Stability Calculator

Paste one row per repeat run. The calculator compares the sets of recommended brands, not their order. It runs entirely in your browser; nothing is uploaded.

Synthetic example: the prefilled rows are invented test data for demonstrating the calculation. They are not a benchmark of any AI engine or market.

Run data (CSV: run_id,brands; separate brands with |)

Example: run-1,Atlas AI|BrightDesk. A header row is optional. Empty brand sets are allowed but are reported.

Bootstrap resamples (deterministic seed: 20260922)

Results

Interpretation: exact-set agreement is the share of run pairs with identical brand sets. Pairwise Jaccard is intersection ÷ union, averaged across run pairs. Coverage is the share of runs containing each brand. The interval is a deterministic percentile bootstrap of the pairwise Jaccard mean; it does not make clustered or convenience samples representative.