Canary Service: ops-dashboard

What this is

The store-network operations console. Live device-health grid, MCP-server observability, alert state distribution, and adapter-sync watermarks. Two surfaces: REST API for state queries and SSE channel for live updates. Also exposes an MCP server (canary-ops) so AI assistants query operational state on demand.

Tier mix and axis

Property	Value
Port	`:9084`
Axis	B — Resource APIs · C — Agent Surface (SSE + MCP)
Tier mix	Stream (SSE channel for live ops) · Reference (snapshot state, device telemetry, MCP grid) · Change-feed (operational event log) · Stream (silence/unsilence actions)
Owned tables	`app.device_telemetry`, `app.adapter_watermarks`, `app.ops_event_log`, `app.device_silences`
MCP server	`canary-ops` — 6 tools
SSE event taxonomy	device.online · device.offline · device.degraded · adapter.lag · mcp.unhealthy · alert.created

MCP surface (Axis C)

Tool	Tier	Purpose
`ops.device_status`	Reference	Get device health
`ops.adapter_lag`	Reference	Adapter sync lag
`ops.alert_distribution`	Reference	Alert counts by severity/status
`ops.mcp_health`	Reference	MCP server observability
`ops.health_rollup`	Reference	Per-tier health summary (5 tiers)
`ops.silence_device`	Stream	Silence noisy device

Purpose

Surfaces tier-rollup data (5 indicators, not binary green/red) per the cadence-ladder Layer 4 surface commitment. Bases query on Brain wiki consumes from same data.

Dependencies

canary-identity (JWT, super-admin scoping)
Every other Canary service (consumes their /health and /status)
canary-store-brain (presence data for in-store device attribution)

Consumers

Operators / on-call engineers
AI assistants (founder ALX, Cowork, in-store agents)
Brain wiki Bases queries (5-tier rollup)