How we analyze reviews

We keep a source's purchase labels separate from what we can establish about the reviewer. A missing or false verified-purchase flag does not prove that a person did not buy the product, and is not by itself a reason to discard their account.

When supported by the source, we report counts for reviews labelled as verified purchases, reviews explicitly described as non-purchases, and reviews with unknown purchase status. The groups and denominator must be stated. A source label is not our independent proof of purchase.

Charts identify the sample actually used. We do not describe a mixed or unknown-status sample as a buyers-only analysis. If a chart uses only source-labelled verified purchases, its count and percentage refer to that subset.

Raw source scores remain in the evidence archive for traceability. Revuei presents review themes and their source, sample and limitations; archive record counts do not establish completed text analysis.

Raw source scores remain in the evidence archive for traceability. Revuei presents review themes and their source, sample and limitations; archive record counts do not establish completed text analysis.

Historical snapshots describe the named source listing at the collection date. They do not establish today's retailer rating or prove that every review concerns the current formula, shade or size. Any known limits belong beside the result.

For each theme, every analysed review is classified as mentioned, not mentioned or undecided. A mention can be positive, negative, mixed, neutral or unclear; a topic name alone does not decide the direction.

Eligible reviews · Q

The distinct source records suitable for the stated analysis after relevant product, text and duplicate checks.

Analysed reviews · E

The eligible records actually processed by the stated analysis version. E is a subset of Q; unprocessed records do not enter its theme percentages.

M / N / U

M means mentioned, N means not mentioned, and U means undecided. These three groups are separate and together cover E.

Direction charts use the theme's mentioned reviews as their denominator when stated. Unclear direction stays separate. Because one review can discuss several themes, percentages across different themes need not add to 100%.

A topic mention is not a clinical finding, an effectiveness probability or an incident rate. For example, mentioning irritation does not necessarily mean the reviewer experienced irritation: negation, comparisons and past experiences matter.

We distinguish repeated archive captures, source review IDs and identical-text groups. Counts represent records under the stated deduplication rule, not a verified number of independent people. We do not silently merge ordinary short phrases from different reviewers.

Source coverage is disclosed for each analysis. Two websites carrying the same syndicated reviews are not automatically two independent samples. An archive count is not the same as a count of analysed reviews.

Product facts and review-theme analyses have separate evidence requirements. An unknown value is not zero. We retain unresolved source information and explain the affected claim rather than treating the whole product as unpublished.

Source text is not reproduced when its permitted use is internal analysis only. We show attributable summaries and calculations within the allowed scope.

Buying links can open the named product's listing so you can choose an available shade, size or other option at the retailer. That choice is yours; a family-level link is not a claim that every option has been separately reviewed.

Some links earn Revuei a commission. This does not determine a product's rating, the interpretation of reviews or its default ranking. Check the retailer for current price, availability and delivery.