How We Calculate Exercise Similarity

Version 2 draws from 2,056 active exercises and scores a bounded relevance-first candidate pool with a versioned, weighted attribute model.

The 12 Active Factors

Muscle sets use Jaccard overlap. Categorical and numeric fields use the comparison method shown below. These values come from the same metadata owner used by runtime scoring.

Movement subpattern
20% - Exact match, with half credit for the same parent pattern
Movement pattern
10% - Exact category match
Primary muscles
15% - Jaccard set overlap
Equipment category
10% - Exact metadata match before feasibility filtering
Body position
8% - Exact metadata match
Laterality
7% - Bilateral or unilateral category match
Stability requirement
7% - Inverse distance on the stored 1-10 scale
Skill complexity
5% - Inverse distance on the stored 1-10 scale
Force vector
5% - Exact category match
Exercise mechanic
5% - Compound or isolation match
Range-of-motion profile
4% - Exact metadata match
Grip type
4% - Exact match, with half credit for a related grip group

How the Score Is Calculated

Similarity score =
(Movement subpattern match x 0.20) +
(Movement pattern match x 0.10) +
(Primary muscles match x 0.15) +
(Equipment category match x 0.10) +
(Body position match x 0.08) +
(Laterality match x 0.07) +
(Stability requirement match x 0.07) +
(Skill complexity match x 0.05) +
(Force vector match x 0.05) +
(Exercise mechanic match x 0.05) +
(Range-of-motion profile match x 0.04) +
(Grip type match x 0.04)

Runtime scoring sums the weighted factor values, applies a stored equipment-conversion adjustment when one exists, caps non-identical pairs below 1.0, and rounds to three decimal places.

A higher score means the stored attributes are more alike under version 2. It does not guarantee identical training outcomes or individual suitability.

Data, Equipment, and Limits

  • The active library count is read from FitnessVolt's exercise owner at request time: 2,056 active exercises.
  • Each request scores at most 1,000 relevance-selected candidates from that active library.
  • Missing metadata receives no guessed match, except the established bilateral laterality default retained for compatibility.
  • Selected equipment is a feasibility constraint by default; unavailable and unknown requirements appear only in the explicitly expanded view and remain labeled.
  • The score is an exercise-attribute comparison, not medical, rehabilitation, or individualized coaching advice.
Algorithm contract 2.0 | Runtime version 2