Unmeasured values

A document with no text layer has no word count. A reader that cannot detect tables reports no table count. Neither is recorded as zero.

Recording either as zero puts an unmeasured value into every average computed from it, with nothing in the output to distinguish it from a measurement.

So every measurement can say it was not taken:

• A signal that could not be measured is reported as not applicable, with what stopped it, and stays out of the score. The weights are renormalised over the signals that did produce a rating, so a document where six do not apply is not penalised for six missing contributions.
• A readiness factor the run did not measure is dropped the same way. complydoc readiness alone still produces a score, computed from one factor instead of three, and the report states which.
• Coverage from a reader that returns no geometry is null. A value of 0% would describe a blank page.
• A token count with no local encoding is marked estimated.

Previous exception

The entity detector previously reported confidence: 1.0 for every name. The spaCy pipeline in use exposes no per-entity score, so the value was a constant, not a measurement, and it ranked model detections level with checksum-backed ones.

It now reports null, and the evidence tier carries the distinction instead.

Reported limitations

Each run generates a list of limitations from its own execution — components that did not run, categories that could not be scanned, prices without a verification date, sampling — and prints them on the report's first page with a severity.