Signals and model checks
Model checks are named measurements, not decorative badges. Each one should tell you what was examined, whether it completed and how it influenced the combined reading.
Why several checks
Generators, editors and compression pipelines leave different traces. A semantic model may recognize scene-level patterns that a frequency measurement cannot, while metadata can reveal source context that neither visual check sees.
IA Checker keeps those readings separate before aggregation so a strong direct source signal is not visually flattened into an unexplained average.
Scanner status language
| Status | Interpretation |
|---|---|
| Completed | The configured check returned usable evidence |
| Skipped | The profile or feature configuration did not request the check |
| Unavailable | A required capability or model bundle was not present |
| Failed | The check started but did not complete successfully |
| Timeout | The check exceeded its allowed runtime |
Partial results
Optional scanner failures are isolated. A completed result can still be returned from the evidence that succeeded, with the execution depth recorded internally so calibration and operations can distinguish a quick reading from a deeper one.