M12: Observability baseline — OTel logs/metrics/traces

[ninan-versioning]

Gap — Observability baseline (logs / metrics / traces)

Vertical: enterprise
Revenue-Signal: all
Customer-Type: internal
Critical path: v0.1 Varitea Closed-Loop

What

Thompson Sampling debugging, Kinfolk gate regression analysis, cost guardrail forensics, App Store reviewer issue reproduction — every one of these requires observability that we cannot retrofit later. Establish OTel-instrumented logs + metrics + traces baseline before v0.1 ships.

Scope

• OpenTelemetry SDK in every service (Node + Python)
• AWS CloudWatch as primary metrics+logs backend
• Traces: cross-repo via OTel context propagation (shoploop-shopify → shoploop-core → shoploop)
• Required spans: BrandRulesUpdate, GenerateCreativePack, KinfolkGate, ThompsonSamplingDecision, PublishAsset
• Dashboards: per-merchant cost burn, per-cycle render success rate, Kinfolk gate pass-rate
• Alerts: cost burn 80% of cap, Kinfolk gate pass-rate < 70%, App Store webhook 5xx rate > 1%

Acceptance criteria

• One trace shows Varitea full closed-loop from Shopify catalog ingest → creative publish → GA4 signal → brand_rules update
• Cost dashboard shows live per-merchant burn rate
• Alert fires within 60s of threshold breach in staging

Affected repos

shoploop (primary) → all four other repos