What TokenBoard Is Really Measuring

TokenBoard is a usage layer for AI coding. It turns local agent activity into a public, comparable profile: tokens, sessions, messages, model mix, estimated cost, active days, and observed skills.

It is not a prompt archive. It is not a billing ledger. It is not a surveillance tool. TokenBoard is closer to a fitness tracker for AI-assisted development: it records the shape of the work without publishing the private work itself.

The problem it solves

AI coding tools are becoming part of the developer workflow, but their usage is usually invisible. A developer may spend all week pairing with Codex or Claude Code, yet the only visible output is the final diff. Teams cannot see adoption. Builders cannot show momentum. Heavy users cannot compare their habits with peers.

TokenBoard makes that activity legible:

• How much AI coding happened? Token totals, messages, sessions, and active days show volume.
• Which models are being used? Model breakdowns reveal the mix behind the work.
• What did it roughly cost? Estimated spend converts token totals into a number people understand.
• Who is consistently active? Leaderboards and profile heatmaps make usage patterns visible.
• What stays private? Prompts, responses, command output, code, and raw local session IDs stay off the public page.

What appears on the leaderboard

The main leaderboard ranks reported usage across participating users. It can compare people by total tokens, sessions, messages, estimated cost, model coverage, and observed skills. The point is not to declare who wrote the most code. The point is to make AI coding usage visible enough to discuss.

A useful TokenBoard profile should answer three questions in seconds:

1. Is this person actively using AI coding tools?
2. What kind of usage pattern do they have: steady, bursty, experimental, model-heavy?
3. Can they share proof of activity without leaking the actual prompts or source code?

That is why the product emphasizes high-signal metrics and avoids raw content.

The privacy model

TokenBoard starts from a simple boundary: metrics can be useful without moving sensitive text.

The local collector reports structured usage facts such as token counts, model names, session timing, message counts, and hashed identifiers. It does not upload prompt text, assistant responses, command output, source files, or full local paths.

This gives TokenBoard a narrower job. It can show that a session happened, which model was used, how many tokens were involved, and how activity changes over time. It cannot reveal the conversation that produced the work.

How to think about the numbers

TokenBoard numbers are designed for product insight and social proof, not financial reconciliation.

• Token totals are client-reported from supported local telemetry.
• Cost is estimated from the server-side model price table.
• Unknown models can be tracked separately when pricing is not configured.
• Backfill may gradually add older sessions after the newest activity is already visible.

In other words, TokenBoard is a leaderboard and usage profile. It is strong evidence of activity, but it is not a replacement for provider billing exports.

The bigger idea

AI coding is becoming a craft with its own habits: model selection, prompt discipline, session rhythm, review behavior, and cost awareness. TokenBoard gives those habits a surface.

It lets individual developers say, "Here is how I actually work with AI coding tools." It lets teams see whether adoption is real. It lets the community compare usage without turning private development sessions into public transcripts.

That is the product promise: make AI coding activity visible, comparable, and shareable while keeping the work itself private.