See the third contact before your customer gives up.

Support agents handle every ticket as if it were the first. Returning users with two prior escalations get the same welcome template as a free-tier sign-up. Sentiment declines across multiple sessions; nothing in the pipeline looks at the trajectory.

Per-call traces in Langfuse show that an agent ran. They don't show which users are about to churn, or which prompt version made it worse.

Signals from the registry fire on every ticket: user_frustration (re-prompting, negative sentiment), session_abandon (premature exit), low_quality_response (failed first-contact resolution), cost_escalation (runaway token spend on complex tickets).

Each signal stacks per user. The /users dashboard groups people into the standard buckets, happy, neutral, at-risk, churning , and shows you who's drifting in which direction.

Power Score = CSAT × Cadence. The users page surfaces the accounts contributing most signal volume relative to the cohort baseline. Cross-reference with the /versions page to see whether the drift started after a specific deploy.

Custom signals you can promote: e.g. “reopened within 48h on same issue category”, “thread reaches 3rd turn without resolution token”, “escalation never triggered for paid tier user”.

Point Flowlines at Langfuse, OTEL, or JSON drop. First connect backfills 30 days. Workspace preset support turns on the right signal blocks by default, frustration, CSAT proxy, first-contact resolution, escalation detection.