What the engine uses
The recommendation system reads the same normalized records used by the Product Database and Finder. It can compare product type, intended use cases, padding/lining, closure, straps, back style, material wording, feel tags, construction tags, support context, brand and sizing model.
How similarity is scored
Similarity is calculated from explicit attribute overlap. Higher-value decision attributes receive more weight: intended use, padding, closure and straps matter more than brand identity. Shared feel and construction tags add smaller amounts of evidence. Missing or unknown data never earns similarity points.
| Attribute family | Role in similarity |
|---|---|
| Use cases | High weight; compares practical purpose such as first bra, school, everyday, sports or sensory use. |
| Padding / lining | High weight because coverage preferences can materially change the choice. |
| Closure | High weight; pullover and back-closure products can feel very different in daily use. |
| Straps / back | High-to-moderate weight; adjustable, fixed, racerback and convertible configurations affect use. |
| Feel / construction | Moderate weight based only on shared verified tags. |
| Material | Moderate weight when the published material wording overlaps. |
| Support / type | Moderate weight; helps separate everyday and activity-oriented options. |
| Brand | Very low weight. A same-brand product is not assumed to be the best alternative. |
What the score does not mean
- It is not a comfort score.
- It is not a quality rating.
- It is not a sales-popularity ranking.
- It is not a claim that two sizes are interchangeable.
- It is not evidence that the garments fit the same body in the same way.
- It is not hands-on testing unless a future page explicitly documents such testing.
How sizing is connected
The product engine keeps sizing as a second-stage check. Once a visitor has selected promising garment attributes, the cross-brand size comparison can map the same current measurements into supported manufacturer sizing systems. Measurement overlap is useful context, but the exact product chart and 5-Point Fit Check still control the final decision.
Commercial independence
The current engine does not add points for affiliate availability, retailer commission, price, sponsorship or placement. If commercial relationships are introduced later, they must remain separate from the similarity calculation and be disclosed under the site's Affiliate Disclosure.
Why publish the methodology?
A recommendation system is only useful if visitors can understand what it is optimizing. Publishing the logic makes it possible to challenge, improve and audit the system as the product database grows.