manual-e2e

name	manual-e2e
description	Drive the real merge0-server binary over HTTP through the whole loop (ingest → triage → review → runner callbacks → telemetry, plus the hosted multi-tenant plane) with no external credentials. Use before shipping server/pipeline changes.

Manual end-to-end run

scripts/e2e-manual.sh boots the real server binary with dev fakes (fake model + fake GitHub — everything else is the shipped code path) and drives ~70 checks over HTTP: auth sweeps, native vendor webhooks, triage runs, approvals, runner callbacks, budgets, telemetry, the SPA, and the hosted control plane with two isolated tenants.

scripts/dev-pg.sh start                      # Postgres on 55432
(cd ui && npm ci && npm run build)           # SPA checks need ui/dist embedded
cargo build -p merge0-server -p merge0-hosted
scripts/e2e-manual.sh                        # or: scripts/e2e-manual.sh <database_url>

Reading failures:

• SPA checks 503 usually means ui/dist was empty when the server was built (rust-embed bakes it in at compile time) — rebuild the UI, then rebuild the server.
• A signal silently missing from triage: check its last_seen against the scout window — e2e payloads need generated now-relative timestamps, not values copied from fixtures.
• Count assertions off after adding a source/scout: expected — new sources change triage arithmetic. Update expectations by content (title match), never by index, and respect the gate's max_work_orders_per_run cap.

When extending the script: keep new checks substring-based via check <desc> <actual> <expected-substring>, generate timestamps with date -u, and never embed real vendor data or credentials.