Reddit
How should real-world AI-tool proficiency be measured without turning usage into a fake expertise score?
Iβm exploring a measurement problem rather than proposing that token count equals skill. I built a local-first technical alpha that records Claude Code and Codex activity, produces a signed privacy-sanitized snapshot, and separates activity telemetry from self-submitted identity, connected work, and outcomes. Prompts, responses, code, local paths, and credentials are excluded from the public payload. The long-term question is whether a portable AI-work record could help researchers recruit g...


