TIER 01Attribution● LIVETIER 02Governance● LIVETIER 04Quality● LIVETIER 05ROI● LIVETIER 06Opportunities● LIVETIER 08Findings● LIVE
QualityResolution efficiency · repeat detection · prompt quality · context hygiene
Repeat Query Rate
Avg Session Depth
req/window
Terse Prompts
Context Hygiene Flags
users
Per-Event Hygienerepeat queries · prompt quality · context-window waste
Repeat Query Detection
Prompt Quality by Department
Attribution Certainty
Each event lands in one of three buckets. Explicit comes from X-Cg-Tag-Useror an account default — ground truth. Inferred is a probabilistic match from the ingestion-worker's attribution model when no explicit user is set. Unattributed is the gap this feature aims to shrink — no user signal at all. Inferred attribution requires nightly profile generation; new orgs spend their first day in calibration.
Context Window Hygiene
Users flagged when ≥days in the window have at least one request with prompt > tokens AND output ratio <. Derived from prompt_tokens / completion_tokens at query time.
Resolution Efficiency15-min sessions · token cost vs. baseline · weighted by intent
Avg RES
calibrating — baseline still building
First-Shot Rate
re_ask_count = 0
Mean TTR
tok/session
Sessions
15-min windows · 7D
Resolution Efficiency Overview
Discovery vs. Solutioning Mix
RES by Department / Purpose
Top 5 Friction Sessions

The five conversation sessions in this window with the lowest RES — i.e. where users burned the most re-asks and tokens before landing on an answer. Use it to spot who is stuck and on what kind of task, so the team can intervene (better prompts, switch model, build an agent).

Agent candidacy and model-routing recommendations live on Tier 06 · Opportunities. The C-suite briefing is at Tier 08 · Findings.