17 · Live
Fit-return Loop
‘This style runs small — sized up from real customer fit reports.’ You see the why.
How it works
Signals in
Customer or brand marks an order returned with a reason; try-on fit feedback (small / true / large); orders kept 30 days count as true to size.
Learned bias
Per SKU after 3 observations, per brand after 10, shrunk toward zero so one report can't swing a size. Recomputed on every return and nightly.
Applied
The Fit Engine adds the bias to the body before sizing; the customer is told a style runs small or large and how many reports say so. Brand analytics show the verdict per SKU.
Real output · sample data
Same body, same size chart. Four real fit reports on the sample slip dress moved its learned bias to +0.9 cm and the recommendation up one size. Fictional model and brands.
Interactive demo
Stretch widens the comfortable negative ease \u2014 on a jersey, snug is the intended fit. Sample chart, not a real brand.
Technical specs
- Inputs
- Returns · fit reports · kept orders
- Granularity
- SKU, then brand
- Safety
- Prior-weighted — evidence, not one opinion