Whitepaper
From Licenses to Signals
Text and Data Mining Rights, Opt-Out Protocols, and Compliance Automation for Machine Learning
Permission to mine text for machine learning has evolved through three eras: human-readable licenses, machine-readable signals, and automated agents that decide without human review. Two unpublished working papers from 2023 and 2024 produced competing license-permission tables, and encoding those tables as a formal test suite exposed six specification gaps in the underlying decision record and rule set.
A checker reproducing the author's license readings recomputed all 101 testable verdicts once the record was extended with six additional fields, and a blind AI second reader matched the oracle exactly on verdict, duties, and reasons in 95 of 101 cases. The central finding is that signals inherited the licenses' mistakes – vocabulary gaps, misdrawn purpose lines, and the absence of a decision record – rather than replacing them.