Only aggregate token rows
The collector sends UTC date, tool, model, token counts, and estimated cost. It never uploads code, prompts, chats, filenames, or file contents.
Rules
TokenRank is a public leaderboard for collector-reported aggregate AI activity. It is an activity signal for comparison and sharing—not a provider audit or productivity score.
Operating protocol
The collector sends UTC date, tool, model, token counts, and estimated cost. It never uploads code, prompts, chats, filenames, or file contents.
Codex uses provider raw input + output because its input already includes cached input. Claude Code and similar tools add input, output, cache reads, and cache writes.
Collectors send aggregate fields, but the server recomputes total tokens from the raw fields before writing leaderboard rows.
Repeated uploads from one device for the same UTC date, tool, and model replace the old row instead of stacking duplicate points.
Each user only ranks with the three highest-usage devices in a leaderboard window, reducing the upside of fake devices.
The server validates tools, dates, integers, recomputed totals, webhook ownership, and duplicate keys. These checks do not turn local data into Provider Verified usage; suspicious records can be flagged, hidden, or removed.
The MVP opens an X Web Intent. TokenRank does not auto-post and does not request posting permissions.