Insights, dashboards, and alert routing — the core PostHog buildout (gates the Sentry sunset)

[JSONbored]

Part of epic #8286 (Phase 2 — Analytics). The buildout that makes the consolidated data answer real questions — and the alerting gate the Sentry decommission depends on.

• Blocked by Provision a loopover-owned Posthog project for the hosted fleet #7875.

Scope

Mostly PostHog-side configuration (mirrors JSONbored/metagraphed#7765), with in-repo docs capturing what exists so dashboards aren't tribal knowledge:

• Error alerting — alert routing for error spikes and new-issue creation across every Phase-1 surface, plus no-data alerts for the five ORB cron heartbeats (the design the self-host error-tracking sub-issue establishes). This is a hard gate for the Sentry decommission: Sentry cannot be turned off before PostHog alerting is live.
• Error dashboards — breakdown by surface (self-host / hosted Worker / miner / discovery-index / review-enrichment / UI), by the ported operational-tag taxonomy (subsystem, repo-hash, provider/model for AI-adjacent failures).
• MCP usage dashboards — the existing src/mcp/telemetry.ts + packages/loopover-mcp events (tool name, caller type, success, coarse latency) — this data already flows; it has never had a proper dashboard home.
• Web analytics dashboard — once the Umami-replacement sub-issue lands.
• In-repo docs/ (or observability README) page enumerating the dashboards/alerts and what each answers, so drift is visible in review.

Boundaries

• Alert destinations (email/Slack/webhook targets) are maintainer account configuration — documented, not committed.
• No new capture surfaces here — this consumes what other sub-issues emit.

Resources

• Insights, dashboards, and alert routing — the core PostHog analytics buildout metagraphed#7765
• https://posthog.com/docs/alerts

[linear-code]

JSO-5739