November 2026 Summaries

Recorded gRPC-web traffic revealed that an API gateway had returned 1,238 INVALID_ARGUMENT errors over two recording windows while reporting HTTP 200, causing conventional HTTP-status searches and passing unit tests to miss failures. Analysis linked several rejected requests to dashboard behaviors such as invalid regex filters, unsupported legacy service parameters, excessive time ranges, and snapshot queries with incompatible filters, all of which displayed missing data or “No traffic” rather than gateway error messages. The team fixed the affected grid and rate-panel error handling, corrected snapshot filter behavior, and strengthened mocks and regression tests so they reproduce the gateway’s actual rejection responses. A coding agent using recorded traffic identified part of one issue cheaply, though human review found additional correctness concerns not evident in the recording. The account recommends searching gRPC-web recordings for nonzero Grpc-Status headers, validating that new tests fail on original code, preserving captured traffic, and sanitizing data before sharing it with AI tools.

Speedscale’s Browser Journeys workflow combines Playwright browser tests with recorded API traffic to help distinguish UI failures from missing, failed, or changed backend responses. The walkthrough uses a sample shopping app to record a successful “Load items” journey, replay it with the real API stopped using mocked responses, and then demonstrate two similar-looking Playwright timeouts: one caused by a nonexistent Checkout button and another by an unrecorded pagination API request, which the report identifies as a mock miss. It also shows how observe mode can route requests to a live API and compare responses against the recording, revealing a backend field-name change that breaks the frontend. The same specs and recordings can be replayed in Kubernetes, either entirely within the cluster or with optional Speedscale cloud reporting that provides journey steps, API evidence, screenshots, and Playwright traces. The guide notes setup requirements, CORS and DNS considerations, generated-test assertion limitations, troubleshooting for aborted requests, and the need for separate configuration identifiers across mocked and real-API modes.