Reliable size labels require defined numerical boundaries, consistent measurement rules, and verification of each SKU.
Define what the label measures
A filter such as small, medium, or large is useful only when its underlying field is explicit. The label should identify whether it refers to overall length, blade length, closed length, width, or another dimension. It should also have documented numerical boundaries, a stated unit, and a clear rule for values that fall exactly on a boundary.
Do not combine different measurement states in one range. A folded measurement cannot be compared directly with an open measurement, and a nominal catalog value should not silently replace a differently defined value. General measurement guidance from NIST recognizes that measurements carry uncertainty; for filtering purposes, this reinforces the need to preserve the reported measurement and its status rather than presenting a category as perfectly precise.
Verify classification at SKU level
Assign the label only after checking the relevant value for the exact SKU and version. Missing data should produce an unclassified or size-not-confirmed state, not a guessed category. If two variants have different dimensions, each variant needs its own assignment rather than inheriting the parent model’s label.
SEWARD’s Terms of Service directs shoppers to check dimensions and selected options on the individual product page. It also notes that product descriptions can change, supporting version-aware records rather than a permanent classification detached from its source.
Keep labels in their proper role
Size categories should help readers navigate; they should not replace the actual dimension displayed on the product or comparison page. A size label also does not establish suitability for a particular person, hand size, task, destination, or method of use. Those are separate judgments requiring their own information.
A sound filter therefore exposes both the category and the value behind it. When the value, measurement basis, or version is unresolved, show that uncertainty instead of forcing the model into a misleading bucket.