Preserve measurement source kinds

These examples and illustrative results are independently authored teaching materials, not measured model results.

Use case

A performance review has source code and one lab report. The teaching report gives LCP=2.8 seconds under stated device/network conditions, with no real-user data. Source reveals an unsized hero image. These do not establish a field p75.

Mechanism

Inventory materials with URL, build, capture time and conditions. Label static findings potential impact and leave unmeasured metrics empty. Label lab values with synthetic-run conditions, field values with population/window/percentile, and traces with captured scenarios. Interpret them separately rather than combining source kinds into one metric. Record unavailable or unparsed sections.

Bad example

Claim real-user LCP p75 is 2.8 seconds from source, relabel the lab result as production behavior, and call the image risk observed layout shift.

Good example

Use only the supplied source and lab report: label LCP=2.8 seconds as this Lab run with conditions, the unsized image as a static potential risk, and Field p75 as not measured. List items needing browser or user evidence; do not invent scores.

Why the change matters

The same metric name can describe different populations or runs. Source and conditions are part of its meaning. A static risk identifies a question for verification, not a measurement.

Observable expectation

Teaching output has separate Lab 2.8-second, static-image-risk and Field-not-measured records. Removing the report makes LCP unmeasured too. A new trace gets its own scenario identity rather than replacing user statistics. Trace every number to a material.

Limits

Labels do not validate artifacts; check capture contracts and files. One lab run does not represent users, and device/window/build differences can defeat comparisons. Frozen tooling and data-window descriptions are dated; verify before actual collection.