Caribbean AI Association
Back to the archive
Boot Loops

How the review loops work

Follow a claim from its source to the chart a reader sees. Each loop can send the work back for correction.

  1. Check the evidence

    Read the original table. Record the date, unit and population. Check the wording against what the source measured.

  2. Repeat the calculation

    A separate reviewer repeats the calculation. A difference sends the reviewer back to the data, definitions or code.

  3. Check the publication

    Compare the figure with the table. Check each language edition, open the PDF and test navigation and downloads.

Featured study

The Jamaica source lists 250 training pairs in its dataset. Its largest balanced test condition uses 83 pairs per class across three classes. The reviewer checked both tables. The paper now reports 249 pairs for that condition and explains the difference.

How many Jamaican Patois examples close the gap between two language models?