How ranks work in Symmetry

The strength score normalizes the estimated 1RM by bodyweight with a fixed allometric exponent k = 0.66 (k = 0.33 for time-based exercises). The 1RM is estimated piecewise based on the set’s reps.

Effective weight per type: weighted exercises use the logged load; bodyweight exercises use (bodyweight + added load) × the exercise’s biomechanical fraction (e.g. pull-up 1.0, push-up 0.65); assisted exercises use max(bodyweight − assistance, 0), and if the result is ≤ 0 the session scores 0 and doesn’t contribute to the rank; timed exercises use the duration in seconds as the magnitude.

High-volume path (push-ups, sit-ups, leg raises, glute bridges): instead of estimating a 1RM, the score is reps × effective weight ÷ bodyweight^0.66.

reps ≤ 5  (Lander):   1RM = 100 × w ÷ (101.3 − 2.67123 × reps)
6 ≤ reps ≤ 12 (Baechle): 1RM = w × (1 + 0.033 × reps)
reps > 12 (Brzycki):  1RM = w ÷ (1.0278 − 0.0278 × reps)
score = 1RM ÷ BW^0.66      · timed: score = seconds ÷ BW^0.33
high volume: score = reps × w_effective ÷ BW^0.66

Per exercise, an exponentially weighted moving average is maintained with k_sessions = 5, i.e. α = 2 ÷ (5 + 1) = ⅓. On the first session, EWMA = score. Although it’s often summarized as “the last 5 sessions,” mathematically there is no hard cutoff: the full history persists with exponentially decreasing weights.

Ramp-up: for sessions 1–4, the exercise rank comes from the best raw score in history (best_1rm_score) and can only go up. From session 5 on, the rank comes from the effective score with the full logic active.

Demotion shield: after every rank-up, 3 sessions activate during which the rank cannot go down. Every promotion resets the counter; promotions are never blocked.

Alongside the EWMA, the system tracks peak_score (historical maximum of the EWMA) and best_1rm_score (historical maximum raw score).

α = 2 ÷ (k_sessions + 1) = 2 ⁄ 6 = ⅓
EWMA_t = α × score_t + (1 − α) × EWMA_(t−1)
EWMA_1 = score_1
peak_score = historical max of EWMA · best_1rm_score = max raw score

With d = days since the exercise’s last session: no penalty until day 60; afterwards, exponential decay with τ = 120 days.

The floor is computed on peak_score: 85% until day 28, a linear slide to 50% at day 56, and a constant 50% from then on.

The effective score that determines the rank is the maximum of the decayed EWMA and the floor.

decay(d) = 1                         if d ≤ 60
decay(d) = e^(−(d − 60) ÷ 120)        if d > 60
floor(d) = 0.85 × peak                if d ≤ 28
floor(d) = linear from 0.85 to 0.50 × peak  if 28 < d ≤ 56
floor(d) = 0.50 × peak                if d > 56
effective_score = max(EWMA × decay(d), floor(d))

Muscle: weighted average of the rank indices of the exercises that train it, with weight = sets × e^(−days ÷ 90) (muscular τ = 90 days). Internally, index 0 is the best rank (Symmetric) and 24 the worst (Iron III).

Muscle group: simple average of the group’s active muscles. The six groups are arms, legs, core (abs, no neck), shoulders, chest, and back.

Global: weighted average of the active groups with weights legs 4, back 3, arms 2, shoulders 2, chest 2, core 2. Requires at least 10 exercises with a rank.

At all three levels, the continuous index converts to a discrete rank with floor (in favor of the earned rank) and the fractional remainder is preserved as progress toward the next level.

w_exercise = sets × e^(−days_since_last_session ÷ 90)
muscle_index = Σ(exercise_index × w) ÷ Σ(w)
group_index = simple average of active muscles
global_index = Σ(group_index × group_weight) ÷ Σ(group_weight)
rank = floor(continuous index) · progress = fractional part

An exercise’s cohort is built per gender — men and women are processed independently — from each user’s best-ever mark on that exercise: one entry per person, all-time, no time window. Outliers are filtered before percentiles are computed, and an exercise’s ranks are only published once the sample is large enough.

Score cutoffs are derived from that distribution by applying the fixed percentile bands in the table below. The band is fixed; the score that delimits it moves with the community.

Recalibration is weekly — Sunday, 03:00 UTC — and the client can cache the cutoffs for up to 7 days. The rank is not a set-by-set real-time leaderboard: it is your position against the latest snapshot of the community.

Iron       III 0–3   · II 3–5   · I 5–10
Bronze     III 10–15 · II 15–20 · I 20–25
Silver     III 25–30 · II 30–35 · I 35–40
Gold       III 40–45 · II 45–50 · I 50–55
Platinum   III 55–60 · II 60–65 · I 65–70
Emerald    III 70–75 · II 75–80 · I 80–85
Diamond    III 85–89 · II 89–92 · I 92–95
Champion   III 95–97 · II 97–98 · I 98–99
Symmetric  99–100 (top 1%)
cutoff(band) = percentile of the cohort’s best-marks distribution