Measurement Unit and Scope Boundaries
State what a measurement actually counts before using it as evidence for a different resource or behavior.
These examples and illustrative results are independently authored teaching materials, not measured model results.
Use case
A JSONL snapshot totals 1,000 bytes with 800 attachment bytes. Asked about remaining context, this measures persisted storage, not active tokens/headroom.
Mechanism
Declare object, unit, time and classification. Reconcile attachment, other and unclassified bytes including line endings/blanks.800/1,000=80% is storage share only. Observe active context/headroom separately through the applicable runtime or mark unknown, recording version/semantics rather than billing inference.
Bad example
Attachments use 80% of model context with 20% remaining because 800/1,000 bytes are attachments; characters/bytes/tokens are equivalent.
Good example
Report 800 attachment bytes and 80% of this snapshot. Active tokens, headroom and reconnect loading are unmeasured. If an authorized supported runtime surface exists, read and attribute it separately, not by storage conversion.
Why the change matters
Units/loading policy differ, so correct byte arithmetic does not establish tokens. Scope labels preserve useful storage information while preventing unsupported resource claims.
Observable expectation
Teaching 800 attachment+150 other+50 unclassified=1,000. Unclassified is not silently conversation text. Report 80% storage and unknown context separately; rereading a snapshot adds no runtime observation.
Limits
Compression, binaries, caching and input conversion affect relationships. Missing values are not zero. Valid storage measurements remain useful; verify runtime commands/support against actual versions.