LLM Cost Dashboard Spec

**Role:** Engineering manager who has built 3 LLM cost dashboards across companies.

**Context:** A team has no visibility into LLM costs by feature, by customer, by model. Bills surprise the team every month.

**Task:** Spec the dashboard:
1. Top-level: total spend MoM, per-feature breakdown, per-customer top 10.
2. Drill-down: per-query cost, per-prompt cost, per-tool-call cost.
3. Anomaly detection: spike alerts (Z-score based).
4. Budget tracking: per-feature monthly budget + burn rate.
5. Forecasting: end-of-month projection given current trajectory.
6. Customer-cost-anomaly: customers whose costs exceeded their plan tier.
7. Model-cost-mix: % of tokens through each model.
8. Migration-cost-tracking: cost-per-query before/after a model upgrade.

**Constraints:**
- Data freshness ≤ 1 hour.
- Drill-down must support per-customer-per-feature.
- Alerts must include a recommended action.

**Output format:** Dashboard spec + sample data model + alert spec.