Research / Method transparency
Bodytype algorithm v2: method, thresholds and version history
Bodytype v2 is a deterministic rule set, not a trained model. It takes four girths — bust, waist, high hip and hips — computes three ratios, and applies a fixed order of thresholds to return one styling label. The measurement positions follow ASTM D5219-25 3.3–3.5; the five labels and thresholds are Bodytype’s own rules.
Internal method review (17 November 2026): formulas, thresholds, rule order and test vectors were checked against the current code. This is not external peer review, professional certification or a validation study.
What the rule set does
1. Four inputs, three ratios.
The calculator uses bust-to-hip, waist-to-hip and high-hip-to-hip ratios; it does not use height, weight, age, photos, shoulder width or a trained model.
2. One deterministic rule order.
Inverted triangle, pear, apple, hourglass, then rectangle; each valid input returns exactly one label.
3. The thresholds are Bodytype’s styling rules.
The 0.95/1.05 bust-to-hip window, 0.80/0.88 waist-to-hip thresholds and 0.95 high-hip-to-hip threshold are not ASTM, ISO, GB/T, FFIT or BSAS classifications.
4. v2 changed the measurement definitions, not the thresholds.
Waist, high hip and hips moved to ASTM-aligned positions; the algorithm version was raised so old links are not silently recalculated.
5. Near a boundary, the card shows a geometric distance, not a probability.
A small change in one measurement can change the label; that is a property of threshold rules, not an accuracy estimate.
What this is
A deterministic, reproducible styling rule table computed locally from four girths, with published thresholds, rule order and synthetic test vectors.
What this is not
Not a trained or scientifically validated classifier, not an ASTM/FFIT/BSAS algorithm, and not a medical, health-risk or body-fat assessment.
Inputs and units
Each girth accepts 40–200 cm (15.7–78.7 in). Values are parsed without rounding and converted with 1 in = 2.54 cm; ratios are dimensionless so cm and in give the same label. No gender, photo, height or weight is collected.
| Order | Input | Field | Landmark (ASTM D5219-25) |
|---|---|---|---|
| 1 | Bust | upper | Fullest part of the bust, seen from the side; tape under the arms and across the lower shoulder blades. |
| 2 | Waist | waist | Side waist point between the lowest rib and the highest point of the pelvic bone, found by bending to the side. |
| 3 | High hip | highHips | Midway between the waist level and the fullest hip/seat level (the standard default midpoint). |
| 4 | Hips | hips | Point of greatest protrusion of the buttocks, seen from the side; feet about 15 cm apart. |
If high hip reads larger than hips, or high hip is at or below the waist, the form shows a position-check prompt but still uses the value you enter. These are measurement-position checks, not medical warnings.
The three ratios
Ratios are dimensionless, so the unit cancels and cm/in give the same label. Classification uses the unrounded ratios.
| Ratio | Formula | Purpose |
|---|---|---|
| bustToHip | B / H | Width of the bust relative to the hips. |
| waistToHip | W / H | How far the waist draws in relative to the hips. |
| highHipToHip | HH / H | Fullness of the high hip relative to the full hip. |
Thresholds
Waist-to-hip (W/H)
0.80 and 0.88 include equality.
Bust-to-hip (B/H)
0.95 and 1.05 belong to the balanced window.
High-hip-to-hip (HH/H)
0.95 includes equality; tested only for Hourglass.
| Constant | Value | Meaning | Kind |
|---|---|---|---|
| BUST_BALANCE | 0.05 | Bust-to-hip within 0.95–1.05 (inclusive) counts as balanced. | Bodytype rule |
| WAIST_DEFINED | 0.80 | Waist-to-hip ≤ 0.80 counts as a defined waist. | Bodytype rule |
| WAIST_UNDEFINED | 0.88 | Waist-to-hip ≥ 0.88 counts as an undefined waist. | Bodytype rule |
| HIGH_HIP_CURVE | 0.95 | High-hip-to-hip ≤ 0.95 counts as a low-hip curve. | Bodytype rule |
| VERY_CLOSE_CM | 1 | Stability radius < 1 cm shows “very close” wording. | Display rule |
| HIDE_PROXIMITY_CM | 3 | Stability radius ≥ 3 cm hides the neighbor row. | Display rule |
| MAX_STABILITY_CM | 60 | Upper bound of the stability-radius search. | Implementation |
| RADIUS_PRECISION_CM | 0.01 | Quantization precision of the stability radius. | Implementation |
ASTM D5219-25 defines terms, positions and the default procedure only; it does not define these thresholds, the rule order or the five labels.
The five-rule table
- Inverted triangle
bustToHip > 1.05 - Pear
bustToHip < 0.95 - Apple
0.95 ≤ bustToHip ≤ 1.05 and waistToHip ≥ 0.88 - Hourglass
0.95 ≤ bustToHip ≤ 1.05, waistToHip ≤ 0.80, highHipToHip ≤ 0.95 - Rectangle
All remaining balanced cases
| # | Label | Condition | Equality |
|---|---|---|---|
| 1 | Inverted triangle | bustToHip > 1.05 | Strictly greater; 1.05 does not match. |
| 2 | Pear | bustToHip < 0.95 | Strictly less; 0.95 does not match. |
| 3 | Apple | 0.95 ≤ bustToHip ≤ 1.05 and waistToHip ≥ 0.88 | Includes both endpoints and equality. |
| 4 | Hourglass | 0.95 ≤ bustToHip ≤ 1.05, waistToHip ≤ 0.80, highHipToHip ≤ 0.95 | Includes equality on all three. |
| 5 | Rectangle | All remaining balanced cases | Fallback; always matches if reached. |
Pear precedes Apple/Hourglass, and Apple precedes Hourglass, so the order resolves overlaps. Rectangle is the fallback for balanced inputs with a waist-to-hip between 0.80 and 0.88, or a defined waist but a high-hip-to-hip above 0.95. It is a label, not “unclassified”.
| Input | Ratios | Label | Note |
|---|---|---|---|
| 95/80/95/100 | 0.950 / 0.800 / 0.950 | Hourglass | All three boundaries include equality. |
| 105/80/95/100 | 1.050 / 0.800 / 0.950 | Hourglass | 1.05 is still inside the balanced window. |
| 106/80/95/100 | 1.060 / 0.800 / 0.950 | Inverted triangle | Strictly greater than 1.05. |
| 94/80/95/100 | 0.940 / 0.800 / 0.950 | Pear | Strictly less than 0.95. |
| 100/88/100/100 | 1.000 / 0.880 / 1.000 | Apple | 0.88 includes equality; high hip is not tested for Apple. |
| 100/80/96/100 | 1.000 / 0.800 / 0.960 | Rectangle | Defined waist but high hip > 0.95 fails Hourglass. |
Boundary behavior
Stability radius (single-field shift, cm)
A distance, not a probability, confidence or accuracy estimate.
findNearestShift shifts one field at a time by ±d and bisects for the smallest d that changes the label, quantized to 0.01 cm. It is a geometric distance, not a probability, confidence or accuracy estimate.
| Band | Result-card behavior |
|---|---|
| radius = 0 | Input sits on a rule boundary: “At a rule boundary · nearest labels: …”. |
| radius < 1 cm | “Very close to … · distance”. |
| 1 cm ≤ radius < 3 cm | “Closest other label: … · distance”. |
| radius ≥ 3 cm | Neighbor row hidden. |
| radius = ∞ | No other label reachable within the 60 cm search; hidden. |
| Input | Label | Radius cm | Neighbors | Note |
|---|---|---|---|---|
| 96/72/94/100 | Hourglass | 1.00 | Pear, Rectangle | A small waist or high-hip change can switch the label. |
| 96/72/97/100 | Rectangle | 1.00 | Pear | High hip 97 exceeds 0.95, so Hourglass fails. |
| 90/72/100/100 | Pear | 5.00 | Rectangle | Far from a boundary; the card hides the neighbor row. |
| 100/80/95/100 | Hourglass | 0 | Rectangle | The input itself is on a boundary. |
Test vectors
Synthetic examples verified against the current v2 implementation; they contain no user measurements. The same file is published for download.
| # | B/W/HH/H | Label | Radius | Neighbors |
|---|---|---|---|---|
| 1 | 96/72/94/100 | hourglass | 1 | pear, rectangle |
| 2 | 96/72/97/100 | rectangle | 1 | pear |
| 3 | 90/72/100/100 | pear | 5 | rectangle |
| 4 | 110/80/100/100 | inverted-triangle | 4.76 | — |
| 5 | 100/90/100/100 | apple | 2 | rectangle |
| 6 | 100/85/100/100 | rectangle | 3 | apple |
| 7 | 100/80/95/100 | hourglass | 0 | rectangle |
| 8 | 105/80/95/100 | hourglass | 0 | inverted-triangle, rectangle |
| 9 | 106/80/95/100 | inverted-triangle | 0.95 | — |
| 10 | 94/80/95/100 | pear | 1 | hourglass |
| 11 | 96/72/101/100 | rectangle | 1 | pear |
| 12 | 96/72/72/100 | hourglass | 1 | pear |
| 13 | 100/80/96/100 | rectangle | 1 | hourglass |
| 14 | 95/80/95/100 | hourglass | 0 | pear, rectangle |
| 15 | 100/88/100/100 | apple | 0 | rectangle |
Version history (v1 → v2)
| Item | v1 | v2 | Impact |
|---|---|---|---|
| Algorithm version | v=1 | v=2 | Old links are not silently recalculated under v2. |
| Waist position | Narrowest torso, above the navel | Side waist point between lowest rib and pelvic crest, found by side-bending | Readings may change. |
| High hip position | Top of the hip bone | Midpoint of waist and hip/seat (standard default) | Readings may change. |
| Hips position | Fullest point, feet together | Side-view buttock protrusion, feet ~15 cm apart | Readings may change. |
| Thresholds | 0.95/1.05, 0.80/0.88, 0.95 | Unchanged | Label changes come from readings, not recalibration. |
| Old links | Restore the result directly | Prefill values and ask to re-measure | No confident result is shown for v1 links. |
The measurement definitions changed, so the same numbers mean something different; old links only prefill values and ask you to re-measure. There is no paired v1/v2 population data, so no “share of people who changed label” is claimed.
Nature and limitations
- Deterministic: the same input, unit and version always give the same result.
- Pure: it depends only on the four inputs and rule constants, not on network, time, randomness or identity.
- Complete: every valid input has exactly one label.
- Reproducible: thresholds live in thresholds.ts, rule order in rules.ts, and the page quotes the same definitions.
- Local: no business backend, database or photo upload; measurements are not sent to a server.
- The rules were not fitted from population data; the thresholds are styling rules, not statistical estimates.
- No validation against FFIT, BSAS or any clinical/anthropometric classification is claimed.
- Only four girths are used; no shoulder width, height, weight, posture, fat distribution or garment ease.
- It cannot separate a large bust from broad shoulders; inverted triangle is a proxy label.
- It gives no health-risk, body-fat, disease, weight or “ideal body” judgment.
- The stability radius shifts one field at a time; it is not a measurement-error distribution or a probability.
- Tape position, tension, stance and breathing affect readings; boundary users should re-measure.
Sources and method
- S1 · ASTM D5219-25, 3.3–3.5 (measurement terms and default procedure)
- S2 · Bodytype rules.ts (five rules, order, equality)
- S3 · Bodytype thresholds.ts (thresholds, version, stability parameters)
- S4 · Bodytype stability.ts (radius bisection and reachable labels)
- S5 · Bodytype ratios.ts / units.ts (ratios, conversion, range)
- S6 · Bodytype rule & content library (algorithm v2)
- S7 · Bodytype verify-rules.mjs (rule, boundary, stability, unit tests)
- S8 · Bodytype plain-language / ASTM term crosswalk
- S9 · Bodytype algorithm v2 test vectors release (GitHub + Zenodo version DOI)
Related: calculator, measurement & ASTM, five labels, system comparison, boundary sensitivity, about.
Cite this page
Bodytype.org. (2026). Bodytype algorithm v2: method, thresholds and test vectors (v2, dataset 1.0). Zenodo. https://doi.org/10.5281/zenodo.24304074
Algorithm v2 · updated 17 November 2026. Stable URL: https://bodytype.org/research/bodytype-algorithm-v2. Test vectors: CSV · schema.
Archived release (tag algorithm-v2-v1.0.0): Code on GitHub · Zenodo version DOI 10.5281/zenodo.24304074. Concept DOI: 10.5281/zenodo.24304073. Cite the version DOI for this snapshot; the concept DOI resolves to all versions.