Analytics

The Analytics page aggregates every record across all Lanes. It's where you go to answer cross-orchestrator questions: how much are you spending, how fast runs go, what's stuck.

Activity

• Active right now — non-terminal records.
• Total runs — every record ever.
• Last 7 days / Last 24 hours — recent volume.

Cost & tokens

• Total spend — sum of cost.totalCostUsd across all records.
• Avg cost / ticket — mean cost across records that have any cost reported.
• Cache hit rate — cache_read / (cache_read + input + cache_creation) × 100. The single best signal of whether prompt caching is working; a high value means the cache is doing most of the heavy lifting.
• Avg turns / ticket — mean number of agent ↔ tool round-trips per ticket. A "turn" is one cycle of Claude speaks → tool result returns → repeat; more turns means more agent work / more iteration.

The 30-day spend area chart plots daily total cost.

Quality

• Success rate — DONE / (DONE + FAILED + CANCELLED).
• Failure rate / Cancellation rate — same denominator.
• Avg attempts — mean attemptNumber across records. Above 1 means tickets are being rerun.

Latency

• Avg / p50 / p95 duration — for terminal records, wall-clock from pickup to terminal state.
• Longest active — the oldest currently-running record (useful for spotting hangs).

Breakdowns

• Runs by kind (donut) — split of records by kind (tickets / indexer).
• Throughput (30 days) (bar) — daily run count.
• Top spend by Lane — five Lanes accounting for most lifetime cost.
• Stuck records — any non-terminal record older than 30 minutes in its current state.

Coverage

• Blueprint coverage — % of Lanes with a Blueprint module attached.
• PRs opened (total) — count of records that produced at least one PR.

Where the numbers come from

Cost and token telemetry come from the Anthropic API's stream-json result event. The runner extracts them after each Claude run and posts them back to the orchestrator, which stores them on the record. Analytics figures are computed server-side over your full run history — they are true totals, not a recent-window sample.

Tokens are the primary usage measure. Token counts (input, output, cache read, cache write) are captured reliably in both auth modes and drive the Total tokens tile and any monthly token budget on your account. Dollars are secondary: the Metered spend tile sums API-key runs only. Subscription (OAuth) runs report an API-equivalent cost estimate on each record — useful for judging a run's weight — but it isn't a bill and is excluded from spend totals.