Evidence interpretation
Confidence cannot fill an unknown field.
Completeness, quality, and confidence describe different properties. Combining them into one optimistic score can hide the most important limitation.
Completeness asks what is present
A completeness review checks whether required fields exist: product identity, price, currency, shipping, seller, condition, availability, delivery, observation time, and source context. A field can be present but still low quality. Conversely, one missing field may be immaterial to one question and decisive for another.
Quality asks whether evidence is useful
Quality includes freshness, source reliability, precision, comparability, and coverage. A current direct observation may be stronger than an old estimate. Ten duplicated variations may provide less independent coverage than several genuinely distinct comparable offers.
Confidence asks how certain the interpretation is
Confidence should describe a particular conclusion under stated assumptions. It should not be used to overwrite weak source material. A researcher may be confident that available evidence is insufficient; that is a valid high-confidence assessment with a conservative outcome.
Use explicit states
Unknown means the evidence was not established. Unavailable means the source could not provide it. Not applicable means the field does not belong to the case. False is an observed negative state. Zero is an observed quantity. Keeping these states separate prevents common errors such as treating absent shipping as free or absent availability as in stock.
Communicate a confidence trail
List the material evidence, missing fields, contradictions, freshness concerns, and assumptions. Then connect each conclusion to those inputs. If an assumption changes, readers should be able to see which conclusion changes with it. Avoid precise percentages unless a tested, documented calibration supports them.
Conclusion
Reliable product research does not compress every limitation into a single score. It preserves the evidence record, names unknowns, and calibrates language to what the record can actually support.