Analytics

A project has two analytics tabs that answer two different questions.

Analytics — “how much of this repo is AI-written, right now?”

The Analytics tab computes whole-codebase AI attribution at the default branch’s HEAD: TracesHub reads the repository via the GitHub App, overlays the pushed attribution ledgers, and buckets every line of the codebase by who wrote it — one bucket per AI model (with its tool), plus a No attribution bucket for everything else (human-written code, or code whose traces were never pushed).

Because attribution is deterministic, these percentages are exact line counts, not estimates.

Notes on how it behaves:

• It reflects the current state of the default branch: lines that were AI-written but later deleted or rewritten by a human no longer count. It’s a snapshot of the code as it exists, not a history of activity.
• The first computation after new pushes can take a moment — the repo is cloned at HEAD and attribution is computed fresh; you’ll see a progress note.
• It requires the GitHub App to have repository access. If the repo isn’t linked or access was removed, the view says so — the Raw trace store still works.
• Zero AI lines right after your first push? The codebase view only counts commits reachable from the default branch’s HEAD. Push your git commits to GitHub (and merge to the default branch) so HEAD includes the traced commits.

Raw trace store — “what’s been pushed?”

The Raw trace store tab is computed straight from the sync store — everything ever pushed to the project, across all branches and commits in the latest window, independent of GitHub. It shows:

• Totals — trace records, ledgers, commit links, and AI-attributed lines.
• AI lines by model & tool — which models (and which coding tools) produced the attributed lines.
• Trace ingest by tool — where the trace records came from (Cursor, Claude Code, Codex CLI, …).
• Conversations & summaries — how many transcripts and session summaries are stored.
• Coverage — how much of what’s pushed carries full context (conversations behind the traces).

Use it to sanity-check pushes (“did my push arrive?”), to see activity that isn’t on the default branch yet, and to understand how much conversational context your team is shipping.

Which one should I look at?