Sessions today
fleet-wide · +14% vs prior day
Median response
target ≤ 3.0s · p95 4.8s
Validator pass
0 hallucinations · fleet · 24h
Citation rate
every response source-cited
Interventions
8 escalated · 15 paused for crew
Knowledge gaps
↓ from 12 yesterday
End-to-end latency, p95 tail, validator pass rate across all 5 ships. Drill in for per-ship breakdown and the slowest sessions today.
- SLG-0429 Alaska embark — highest volume (561 sessions), 2.5s median
- SLG-0430 Transatlantic — fastest (2.1s median, 99.9% pass)
- P95 latency 4.8s — within target, no tail issues
Volume and routing breakdown by intent category. Structured queries (PostgreSQL) handle exact-time questions; semantic (Pinecone) handles narrative; hybrid fuses both.
- Menu lookup — 184 sessions · semantic · 100% pass
- Port-of-call info — 142 sessions · structured · 100% pass
- Show times & venues — 118 sessions · hybrid · 99% pass
Every response goes through a dual-validator quality gate. Most pass. Some get hedged for precision; some get rewritten for relevance; a few get paused entirely for human review.
- Validator pass — 1,838 today (99.6%)
- Precision flags — 12 (AI hedged, no exact source match)
- Critical pause — 5 (safety, allergens, PII)
Every response resolves to a specific source record (Pinecone vector, PostgreSQL row, or both). Source-centric view here. Conversation-centric view at Assistant → Citations.
- cd_venue_hours.lido-buffet — 412 citations today
- specialty-italian.allergen-sheet.v3 — 218 citations
- cd_events.debark-schedule.v2 — 89 citations