홈
콘솔의 첫 페이지: 얼마나 들어오는지, 무엇이 움직였는지, 무엇이 사람을 필요로 하는지. 모든 숫자는 최근 14일을 그 이전 14일과 비교해 읽고, Needs you 패널은 실패했거나 대기 중인 워크플로 실행을 주의가 필요한 플래그와 나란히 모읍니다.
The metrics band
Two totals for the last fourteen days: events received and the people they came from. A project with nothing in the window shows an explanation rather than zeros, and a failed load says it failed rather than reading as a quiet account.
Needs you
Everything a person has to act on, in one list: workflow runs that failed, runs waiting on a signal, flags nothing has evaluated lately and flags past the lifespan their intent declares. A failed run wears the error tone; the other three are calls on the reader rather than faults. Each row carries its own way out: open the run or the flag it names, fire the workflow again, send the signal a waiting run expects, or archive a flag that is done. The count beside the title is the whole list, so a scrolling list still says how much it is holding.
Top movers
The events whose counts moved most against the preceding window of equal length: the last fourteen days against the fourteen before them. Each row shows the current count and a signed delta, so a spike and a collapse are equally visible.
Event volume
Ingest volume across the window as a chart, with empty days drawn as empty rather than skipped: a project that went quiet for a week looks quiet, not compressed.
Anomalies
One event at a time, chosen in the panel's picker and opening on the project's busiest: a bar per day across the fourteen-day window, so the baseline is visible rather than asserted. A day whose count left that baseline is tinted and dated, and a row beneath marks it ▲ or ▼ and names the count against the baseline it left. A fortnight that held its line still draws the band and says so.
Where this meets the API
The concept behind the numbers is the Analytics pillar. On the wire, every panel is a read: the metrics band, movers and volume are runQuery aggregations, the anomalies strip pairs runQuery with aiInsights (a z-score over the same daily counts, no model call), and Needs you joins flags with workflow runs (GraphQL reference). The events being counted arrive through the ingest edge (Ingestion), and Needs you stays current between reloads by listening to flag and workflow run changes (Realtime events).