Compound Labs open datasets: 18 measurements of public AI tooling

18 datasets, 145,791 rows. Each one is a measurement of public things: public repositories, public component registries, public pricing pages, public app store listings, public agent configuration files. Each is computed by a Compound Labs index product that publishes the same figures on its own site, and each is refreshed monthly.

Every dataset ships as a CSV and a JSON file of the same rows, named with the date of the cut they hold. The two largest ship gzipped. The column list for every dataset is an allowlist written one column at a time, so a column added to a source table appears in no export until somebody adds it to that list on purpose, and a second check refuses any column name carrying a private shape before a byte is written. Tables carrying subscriber email addresses, author handles at scale, or submitter fields on a pending submission are named in manifest.json under never_published and are not in this record in any form.

Each table is read 1000 rows at a time against the server's own exact count, and a dataset whose fetched row count does not equal that count is not written. The measurement method the whole set shares is published at https://toolproof.thecompound.tech/methodology.

This record holds two cut dates, 2026-09-04 and 2026-09-10. Each file states its own in its filename, and manifest.json states it per dataset alongside the row count, the column count, the source table and a citation line.

The datasets

• Toolproof: the nine indexes and what each one currently measures. 9 rows, 16 columns. One row is one index. Measured by Toolproof, https://toolproof.thecompound.tech. Cut of 2026-09-04.
• SkillWorks: Claude Code artefacts by category and kind. 74 rows, 11 columns. One row is one category and artefact kind. Measured by SkillWorks, https://skillworks.thecompound.tech. Cut of 2026-09-10.
• StillShipping: maintenance verdict for every tracked agent tool. 340 rows, 39 columns. One row is one tool. Measured by StillShipping, https://stillshipping.thecompound.tech. Cut of 2026-09-04.
• StillShipping: the daily verdict history. 10,142 rows, 14 columns. One row is one tool on one day. Measured by StillShipping, https://stillshipping.thecompound.tech. Cut of 2026-09-04.
• ToolDrift: the AI coding tools under watch. 36 rows, 34 columns. One row is one tool. Measured by ToolDrift, https://tooldrift.thecompound.tech. Cut of 2026-09-04.
• ToolDrift: OpenRouter model usage rankings, captured daily. 44,369 rows, 10 columns. One row is one model in one ranking window on one capture day. Measured by ToolDrift, https://tooldrift.thecompound.tech. Cut of 2026-09-04.
• ToolDrift: OpenRouter app usage rankings, captured daily. 641 rows, 9 columns. One row is one app in one ranking window on one capture day. Measured by ToolDrift, https://tooldrift.thecompound.tech. Cut of 2026-09-04.
• KitGrade: SaaS starter kits and what is measurably in the box. 36 rows, 43 columns. One row is one starter kit. Measured by KitGrade, https://kitgrade.thecompound.tech. Cut of 2026-09-04.
• KitGrade: the component scores behind every kit grade. 36 rows, 12 columns. One row is one kit. Measured by KitGrade, https://kitgrade.thecompound.tech. Cut of 2026-09-04.
• StoreReady: AI app builders and whether their output ships. 14 rows, 18 columns. One row is one app builder. Measured by StoreReady, https://storeready.thecompound.tech. Cut of 2026-09-04.
• StoreReady: the cited evidence behind every verdict. 47 rows, 10 columns. One row is one piece of evidence. Measured by StoreReady, https://storeready.thecompound.tech. Cut of 2026-09-04.
• BlockDex: every public shadcn registry. 1,123 rows, 26 columns. One row is one registry. Measured by BlockDex, https://blockdex.thecompound.tech. Cut of 2026-09-04.
• BlockDex: every component, block and theme inside those registries. 88,612 rows, 27 columns. One row is one registry item. Measured by BlockDex, https://blockdex.thecompound.tech. Cut of 2026-09-04.
• StackTab: the developer services under price watch. 29 rows, 8 columns. One row is one service. Measured by StackTab, https://stacktab.thecompound.tech. Cut of 2026-09-04.
• StackTab: every plan, its price, and the page the price was read from. 64 rows, 15 columns. One row is one plan. Measured by StackTab, https://stacktab.thecompound.tech. Cut of 2026-09-04.
• RuleStack: the agent config formats and what each one supports. 8 rows, 24 columns. One row is one config format. Measured by RuleStack, https://rulestack.thecompound.tech. Cut of 2026-09-04.
• RuleStack: how much each config format is actually used, day by day. 168 rows, 12 columns. One row is one format on one day. Measured by RuleStack, https://rulestack.thecompound.tech. Cut of 2026-09-04.
• ShipWall: launched products and the badge check behind each one. 43 rows, 23 columns. One row is one launched product. Measured by ShipWall, https://shipwall.thecompound.tech. Cut of 2026-09-04.

Files in this record

Two data files per dataset, plus manifest.json (the machine readable index: every dataset's title, unit, cut, row count, column count, source table, file sizes and citation), README.md (the same index for a reader), CITATION.cff and LICENSE. Each dataset directory in the GitHub repository also carries its own data card and a schema.json stating every column and its type.

Repository: https://github.com/kyisaiah47/compound-datasets