# Data dictionary

Snapshot date for all files: 2026-09-26. All counts refer to the 11,890 AI tools active on that date. Verdicts: `block`, `controls` (allow with controls), `allow`.

- `sector_key`, `sector`: one of 15 sector profiles. `strictness`: 2 = regulated or minors, 1 = standard, 0 = permissive.
- `category`, `subcategory`: names from the AI tool taxonomy (18 categories, 165 subcategories), see https://doi.org/10.21227/hrxe-fa21.
- `category_default`: the baseline verdict a profile gives a category before subcategory rules and tool attributes.
- `tools`: number of tools with that primary category (in the subcategory file, `tools_labelled` counts tools labelled with that subcategory).
- `block`, `controls`, `allow` and `*_share`: final verdict counts and shares after all rules.
- `sector-pairwise-disagreement.csv`: `tools_with_different_verdict` counts tools whose verdict differs between the two sectors.
- `verdict-agreement-across-sectors.csv`: `distinct_verdicts_across_15_sectors` is 1 when a tool gets the same verdict everywhere.
- `k12-subcategory-rules.csv`: `k12_rule` is the K-12 baseline for the subcategory; `departs_from_category_default` is true when it differs from the category default; `final_*` are final verdicts for tools labelled with the subcategory; `blocked_share` = final_block / tools with a verdict; `rationale` explains the rule (empty where the paper gives none).
- `k12-decision-layers.csv`: `baseline_kept` vs `changed_by_tool_attributes` splits the tools decided at each layer.