# Review the request that never reaches the server

A backend success ratio covers only what the backend observes. Check which failures happen earlier and how the service will recognise them.

By Cobnex editorial. Published 2026-09-10. Updated 2026-09-11.

## Draw the observation boundary

Trace the user's path through name resolution, network connection, routing, browser execution and the application. Mark where the chosen measurement starts.

If the indicator reads server logs, requests that fail before reaching the server may be absent. That does not make the measure useless, but it limits the claim it supports.

State the boundary in the indicator definition rather than leaving readers to infer it from the data source.

## Add complementary checks deliberately

Synthetic probes can exercise an external path, while browser telemetry can observe parts of the real client experience. Each has limitations in coverage and availability.

Choose checks that address the important blind spots. A probe from one location may miss a problem affecting a particular network or client class.

Avoid combining all observations into one percentage unless their populations and weighting have a clear interpretation.

## Test absent telemetry

Make the page fail before its normal telemetry starts, or use a safe network fault in a test environment. Determine which signal notices the problem.

Then make the telemetry destination unavailable. The monitoring system should distinguish missing evidence from a healthy service.

Check alert paths too. An incident that disables both the application and its only notification route needs a different operational design.

## Connect the result to a response

For each measurement, identify the failure it can reveal and the action or investigation it supports. Remove ambiguous claims that imply complete user coverage where none exists.

Review representative failed and successful journeys with the service owner. Confirm that the objective remains meaningful across the observed boundary.

Approve the indicator when its strengths and blind spots are explicit. A narrower measure with honest supporting checks is more useful than an end-to-end label attached to a query that only counts successful arrivals at the server.

## Sources

- [Google SRE: implementing service objectives](https://sre.google/workbook/implementing-slos/)