Measure the action, not just the screen
Measurement keeps the action ID, version, place used, result, and available checks together. This makes release comparisons possible without pretending that use alone proves value.
Separate the questions
- Reliability: did the action finish as expected?
- Use: are people or AI tools using the action, and where?
- Result: did the user get the intended product result?
- Cause: is there enough information to say a specific change produced that result?
Privacy and sources
Collect only the information needed to answer the product question. Keep the source, version, permission to use the data, and limits of what it can show when importing totals or run records.