# How DevImprint measures AI-assisted commits

Canonical: https://devimprint.com/methodology/

Methodology version: 2026-09-13.

## Detection

Counts refer to observed tool attribution in public repository history. Associations are evidence, not verified authorship. Exact matching rules, attribution identifiers and detection examples are not published, to limit manipulation. See [observed tool activity](https://devimprint.com/tools/).

## Ranking

The trailing 30-day window starts 2026-08-15. People, tools and unresolved identities have separate category ranks. All-time counts do not determine current rank. The published JSON preserves global rank as well. 1000 published rows out of 222270 total ranked identities. 846 people, 19 tools/agents, 135 unresolved. Published rows are a bounded subset, not a complete export.

## Limitations

Private repositories, incomplete scan coverage and differences in public attribution limit what is observable. False positives and missed activity are possible; counts are not verified totals or guaranteed lower bounds. The current day may be incomplete. Counts do not measure productivity, skill, effort or code quality.

## Provenance

DevImprint recorded 28,137,270 detectable AI-assisted commits for all ranked identities during 2026-08-15–2026-09-13 (UTC). Population: complete ranked population. Snapshot: 2026-09-13T03:27:04.226857+00:00. 49.6% of known repositories scanned; provisional. Methodology: 2026-09-13. Source: https://devimprint.com/. Aggregate snapshot: https://devimprint.com/snapshots/da8f0f66a1eb4f0fa64db42f3b3762a405e05dca6f9f6286ae653e74b07055b7/

## Corrections and removal

[Request a correction or removal](https://devimprint.com/data-request/). Personal records are regenerated on publication; only aggregate records are archived. Full explanation: https://devimprint.com/methodology/.