Methodology

Sources

• Federation scorecards — NPC News Online, IFBB Pro League results pages, NABBA archives, WFF results PDFs. Primary source for every placement.
• Wikipedia & Wikidata — for historical edition pages, athlete biographies, and infobox data (height, date of birth, nationality). Cross-checked against scorecards where possible.
• Podcast transcripts — every episode of major female-bodybuilding podcasts is auto-transcribed with Whisper and indexed for quotes. Source attribution links back to the original episode with timestamp.
• Sponsor announcements & brand-ambassador pages — for sponsorship and supplement-stack data.
• Athlete public statements — first-person claims by the athlete (interviews, social captions) explicitly attributed.

Iron Score

Iron Score is a placement-weighted, career-cumulative rating that starts at a baseline of 1200 for every athlete and adds points for each competition result.

score = 1200
     + Σ ( placement_points × tier_multiplier × recency_weight )
         over each season's 3 best results

     — and nothing else. No social reach, no media presence,
       no Hall-of-Fame points, no credit for career length.

placement_points: 1st=100, 2nd=65, 3rd=45, 4th=28, 5th=18, 6th=12,
                  7-10=6, 11-15=2, else=1
tier_multiplier:  apex (Olympia/Arnold/Rising Phoenix) = 6.0
                  pro shows = 2.0
                  amateur qualifier = 1.0
                  amateur regional = 0.5
recency_weight:   1.0 for current year, decays 0.025 per year, floor 0.5
best per season:  only the 3 highest-scoring results of each season count

Revised 28 September 2026. Every result used to count, so the score grew with the number of shows entered: Bikini pros enter six to eight a year, bodybuilders two or three, and the board had a Bikini champion with 45 wins in 72 shows ahead of a ten-time Ms. Olympia. Now only an athlete's three best results of each season count — how well she does, not how often she turns up.

Revised September 2026. The apex tier previously counted ×3 and regionals ×1, which meant a championship could be out-accumulated by regional attendance: forty regional wins scored 4,000 against eight fully-decayed Ms. Olympia titles at 1,200. The tier spread is now 12:1 rather than 3:1 and the placement tail is steeper, so showing up still counts but no longer compounds into a title. Recency decay was not changed.

We chose placement-weighted summation over classic ELO because chess-style ELO penalizes a 10-time champion too heavily for a single sub-podium showing. The full formula is in scripts/recompute-rankings.sh in the public repository.

Podium chance

Before a show, each athlete in the field gets a chance of finishing in the top three. It comes from one thing only: her placings at professional shows. Every athlete has a strength, learned from who she has beaten and who has beaten her, in the same division family; recent shows count more than old ones. Amateur results are left out: beating amateurs says little about a pro line-up. A placing is read from the top: first place goes to each athlete in proportion to her strength, second place the same among those left, and so on. This is the Plackett-Luce model, the standard model for rankings.

What it cannot see: her condition on the day, the judges on the panel, a change of division, an injury. It is what the record says, not a judgement of the show. An athlete with no pro placing in the division gets no number.

Most shows publish no entry list before the day, so odds written in advance are rare (the Olympia publishes one). For every pro show of the season that is over, the show page therefore carries the record's podium chance: worked out after the show, from the pro placings before the month of the show only — exactly the procedure of the test below. Nothing from the show itself, or after it, goes in.

How it was tested. We replayed the 2025 season as if it had not happened: before each month, the model learned only from pro placings before that month and gave odds for every IFBB Pro show in it. The settings were chosen on 2024, so this score was never used to pick them.

Does a 40% happen 4 times in 10? Every podium chance the model gave in 2025, grouped, against how often it came true:

Settings: a result counts half after 2 years; a short record is pulled toward average with weight 0.05. The test passed, so the odds are shown on show pages. The test reruns every week. The code is scripts/placementmodel.py in the public repository.

Heights from NPC classes

At an NPC amateur show, Bikini, Figure, Wellness and Women's Physique competitors are measured for height at check-in and placed in a height class; Women's Bodybuilding competitors are weighed and placed in a weight class. The class each woman was placed in is on the NPC News contest archive. The limits of each class come from the class tables the NPC publishes for every division (Bikini, Figure, Wellness, Women's Physique, Women's Bodybuilding).

The archive shows the class letter but not the table a show used: a show can split a division into 2, 3, 4, 6 or 8 classes. The classes a division actually had narrow it down — a division with a Class F cannot have used a 4-class table. When more than one table fits, we take the widest reading. A height is the overlap of all her classes; if two classes cannot both be true, we show none. Figure divisions with eight classes are left out: the NPC does not publish an eight-class Figure table.

Who is who is not guessed from names: the archive gives every athlete one number that stays the same from her amateur shows to her pro shows, and our pro results already cite her archive pages.

Muscle Dominance Heatmap

Each muscle group on an Iron Card receives a 0–100 dominance score, computed nightly as a weighted composite of six signals:

1. Placement signal (40%) — wins and podiums in shows where the muscle group is decisive in judging criteria for that division.
2. Mention signal (25%) — positively-sentimented references to the muscle group in podcast quotes about the athlete.
3. Statement signal (15%) — first-person claims by the athlete about the muscle group, with source URL.
4. Pose signal (10%) — the mandatory pose at which she placed highest, mapped to its primary muscle.
5. Measurement signal (5%, optional) — when public tape measurements exist for the athlete.
6. Coach archetype (5%) — pattern of physiques produced by the same coach. Tie-breaker only.

Tape measurements for female bodybuilders are rarely public, so we deliberately avoid making them the primary driver. Each cell links to its contributing source URLs on the athlete's profile.

Iron Card

An Iron Card is a stylized identity graphic generated entirely from public data. The silhouette is one of four parametric body archetypes selected by BMI cluster — not a portrait. The heatmap, division code, signature pose, and Iron Score all derive from public facts cited on the athlete's profile page.

We do not generate AI photographs of real athletes. Iron Cards are heraldic representations of public statistics, comparable to the diagrams in a sports encyclopedia.

What we don't do

• We do not estimate measurements an athlete has not made public.
• We do not publish AI-generated photorealistic images of real people.
• We do not invent statistics; missing data is shown as —.
• We do not claim affiliation with any federation. We are an independent reference.

Found something wrong?

Tell us. We log every report, fix what is wrong within 48 hours, and publish the correction.