Canonical spends view based on normalized metering events. Chat is aggregated per session while distinguishing visible Q/A from internal runtime events; NRU is grouped by HEA + job_id, falling back to run_id when needed, and broken down by Foundry phase.
Using deduped normalized metering rows from /api/insight?action=metering&granularity=event. Visible questions use distinct turn_id when present; runtime events remain step-level.
No normalized events in this window
Try widening the date range, clearing filters, or sending a few test conversations/refreshes.
Chat Sessions
—
Σ NCU: —
Visible Q/A
—
avg/session: —
Runtime Events
—
avg/question: —
NCU / Question
—
vs BP: —
NCU / Runtime Event
—
internal step cost
Session NCU p50 / p95
—
session-level shape
Refresh Runs
—
Σ NRU: —
NRU / Refresh Run
—
vs BP: —
Median / p95
—
shape check
NRU by Foundry Phase
Discovery
—
Enrichment
—
Training
—
Localization
—
NCU per Chat Session — Histogram
NRU per Refresh Job — Histogram
Per-Session NCU (aggregated from normalized events)
Session ID
HEA
Questions
Runtime events
NCU (sum)
NCU/question
NCU/runtime event
First seen
Last seen
Per-Refresh NRU (grouped by HEA + job_id, fallback run_id)