Current state

Architecture

Single server host (Proxmox LXC) runs the full backend, frontend, observability stack, and ingress (Traefik).
A secure tunnel from the server to the edge server exposes the platform to the public internet at the deployment URL.
Two further VPS hosts exist but are unused.
All environments collapse into one: the server serves dev playground, integration testing, and "production" simultaneously.

1,825 vitest unit tests covering frontend slices, middleware, components, and parsers.
147 Playwright end-to-end tests, but against the Vite dev server, not the built container.
11 backend handler unit tests for the LLM parse-ticket endpoint.
*.integration.tc.test.ts suites use Testcontainers. These are the only tests that exercise real wired services, and they only run in CI.
The legacy localhost-or-VETA_BASE_URL smoke suite has been ported to smoke.full.tc.test.ts; see the smoke testing page.

Every PR runs lint, typecheck, vitest, Playwright, CodeQL, secret scan, image builds.
On main merge, CI publishes container images to ghcr.io with the :latest tag.
A systemd auto-pull timer polls the server every ~5 minutes, pulls new :latest images, and runs docker compose up -d when the main-branch SHA changes.
No pre-deploy verification beyond container healthchecks. No staging. No automated rollback. No on-failure alerts (the synthetic probe shipped 2026-05-15 covers that gap; see target architecture for current monitoring state).

LGTM stack on the server (Loki + Grafana + Tempo + Prometheus + Alloy).
Frontend ships warnings and errors to Loki via the gateway to kafka-relay path (after PR #191).
Backend services emit structured logs and OTLP traces.
No SLO dashboards. No alerts beyond the synthetic probe. No formal error-budget tracking yet.

Failures are detected by:

That is the same loop a hobbyist project runs. It is not the loop a regulated trading venue runs.