How datfit predicts your size.

1. Inputs

For tops and bottoms, datfit needs height (cm), weight (kg), body type (slim / regular / athletic / curvy), and fit preference (tight / regular / loose). For shoes, datfit needs the brand and size you currently wear in any of the 11 supported shoe brands — that pair maps to a foot length range, which becomes the working measurement.

2. Containment-first scoring

Each size in the brand's chart has a min and max for every measurement it tracks (height_cm, weight_kg, foot_cm). For each candidate size, datfit computes two numbers:

• Outside-distance — how far you fall outside the size's range, normalised by the range's width. If you're inside the range, this is zero.
• Midpoint-distance — how far you are from the centre of the range, again normalised.

The composite score is outside × 5 + midpoint × 0.1. The 5× multiplier is what makes it containment-first: any size that contains you fully outranks any size you fall outside of, even when the outside size's midpoint is closer. This was the core fix to a real bug in v1.0 where 184 cm / 98 kg was returning Adidas XL instead of L.

3. Body-type bias

For apparel only (not shoes), the body type nudges the pick by ±1 size only when the biased size is reasonably close (score difference under 0.6). The intuition: athletic and curvy builds tend to wear one step up at the same height/weight; slim builds tend one step down. The "reasonably close" gate prevents wild overrides when you're solidly inside one specific size.

4. Fit-preference shift

After the body-type pass, fit preference shifts the answer one step in either direction. "Loose" bumps up, "tight" bumps down, "regular" leaves it alone. The result includes a fit note explaining the shift.

5. Confidence rubric

The confidence number you see is computed as follows:

• If the picked size contains your measurements: starts at 92, drops 12 × midpoint-distance.
• If the picked size doesn't contain you: starts at 84, drops 8 × composite score, floor of 50.
• −4 if fit preference is not "regular" (we shifted off the natural pick).
• −2 if body type is not "regular" (apparel only).
• Final clamp: 48–96. Ceiling at 96 because no chart-based prediction is 100% certain.

6. Upload mode

When you upload a size chart instead of picking a known brand, datfit scores you against a generic consensus chart for your gender and garment type — built from averaging all known brand charts in that category. The same body-type and fit-preference logic applies, but the confidence is capped lower (45–88 vs 48–96 for known brands) because we haven't actually parsed your chart's columns yet — that's the OCR work in progress.

Open data

The full size-chart database is public: /data/size_charts.json. Each entry has a brand, gender, type, and a list of sizes with min/max ranges per measurement. Pull requests welcome once the repo is open.