an instrument, run like one

gaige is a calibration and drift instrument for AI-text measurement: a Python package, a command line, and this site, which publishes the receipts the tool emits. the pitch fits in one sentence: whatever detector you already run, gaige reports what it actually does on your material, with error bars and a fingerprint attached. the certificate carries the full scope; the package page walks the install.

gaige is run as a deliberately small, deliberately closed bench. issues and receipt-backed reproductions are welcome; pull requests are not accepted. that is chain of custody rather than unfriendliness: every number this project publishes must reproduce from a command somebody here actually ran, and a closed bench keeps that promise cheap to keep.

the work comes out of practice: security engineering in environments where the network is a privilege, doctoral-level research in AI measurement and detection calibration, and years of defending numbers to reviewers who were paid to doubt them. names and résumés stay off this site on purpose. here, receipts outrank credentials, and every receipt reproduces without knowing who ran it first. that is policy, stated on the certificate too, and it holds until the receipts themselves give a reason to change it.

the tool is dogfooded on this site: public copy gets scored by the same instruments the package wraps before it ships, and the pages carry their own build receipts in the footer. when a promise here changes, the privacy page changes in the same deploy.