A replayable region-locked sentinel set is the simplest foundation for monitoring. Pick a small set of URLs that represent your main templates, lock the locality signals inside a window, and validate with a few gates before you treat differences as market changes.
Start from the slice you want to compare
Define the slice first: market, language, and page type. The sentinel set is a tool to sample that slice repeatedly. If you mix markets or template types in one slice, you will not be able to explain drift.
Keep sentinel count small. A set that you can replay twice inside one window is more valuable than a large set you can only run once.
Lock locality and session shape inside the window
Inside the window, keep locality stable and keep session continuity long enough to avoid layout switches. Rotate exits at window boundaries. Use a retry cap so slowdowns do not turn into clustered retries.
Signals to check before you summarize
Before you summarize results, check three gates: locality consistency, usable-record ratio, and pacing budget. If any gate fails, mark the window non-comparable and record the diagnostic evidence.
Keep the sentinel set maintainable
Rotate a small portion of the sentinel set when you detect a new template variant, but keep a stable baseline. That way, trend comparisons remain anchored even as targets evolve.
FAQ
How many sentinels should I start with?
Start with 10–30 per slice, enough to cover template variants you care about. Increase only when gates stay stable across several replays.
Should I treat missing fields as an error?
Treat it as an input-quality failure. If required fields are missing, the snapshot is not comparable even if the HTTP status is 200.