# Model observability telemetry kit

Use this kit to make model-backed services observable enough to operate. It separates provider health, workflow success, quality movement, cost pressure, and incident response so teams can act on signals instead of collecting dashboards with no owner.

## What it includes

- Event definitions for prompts, retrieval, model routes, tool calls, approvals, fallbacks, cost, latency, and quality.
- Alert rules for quality regression, cost runaway, fallback surge, provider degradation, and approval overload.
- A telemetry schema for traces, route decisions, token spend, source usage, and incident linkage.
- A review agenda for converting anomalies into owner actions.
- A dashboard model for executive, product, platform, and operations views.

## How to use it

Instrument the workflow before scaling traffic. Route every alert to a named owner with a response expectation. Keep service health, model quality, and business outcome metrics visible together so incident review can distinguish platform failure from workflow design failure.