Where the numbers come from
Most football draft games invent their player ratings. Flawless does not have any. Every figure on every card is a real career record, and every opponent your team faces is rated from matches that were actually played. This page sets out exactly how, because a simulation you cannot inspect is just an opinion with a scoreboard.
Last updated 14 August 2026
The players
The squad list for any (team, decade) cell is built from Wikidata, which publishes structured records of football careers under a CC0 public-domain dedication. For each club we resolve its Wikidata entity, then query every player whose career includes a spell at that club, along with the appearances and goals recorded for that spell.
That gives 85,049 player-records across 841 squads — clubs from six top flights and the national sides, from the 1930s to the 2020s. A player appears in every decade their spell at that club actually covers, which is why the same footballer can be drafted from two different cards. The appearances and goals shown are their career totals at that club, not a per-decade split: the public records are not reliably broken down by season across this whole span, so we report the figure that is actually sourced rather than inventing an apportionment.
The numbers on a card are therefore descriptive, not editorial: appearances, goals, assists where recorded, and honours won. Nobody sat down and decided a player was an 87.
What the Legend score is
The one derived number is the Legend score, and it is a summary of the record rather than an opinion about the player: it combines appearances, goals, assists and honours for that spell, normalised so that figures from a goalkeeper, a defender and a striker can sit in the same list. It is used to rate the squad you assemble. It is not a skill rating, and it does not attempt to say who was better.
The opposition
A drafted XI has to play somebody, and that somebody has to be rated on a scale that means something. Three separate sources feed this, because no single rating system covers the whole board.
European clubs
Club strength comes from ClubElo, which maintains continuous Elo ratings for European clubs from their match results. We take the mean rating per club per decade, so a club is rated as it actually was in that period rather than as it is now.
National teams
For international sides we run our own Elo replay over martj42/international_results, a CC0 dataset of international match results going back to the nineteenth century. Every match is replayed in order, ratings updating as they go.
This produced one artefact worth being explicit about. A pool of teams that only ever plays itself has a fixed total rating, so as the number of rated nations grows over the decades, the average drifts — an effect of the arithmetic, not of football getting better. We correct for it by de-trending each decade against ClubElo’s own drift over the same period, then re-centring so the correction removes the trend without moving the overall level.
Brazilian clubs
ClubElo rates Europe only, so the Brazilian league gets its own Elo replay, over match results from schochastics/football-data (ODC-BY 1.0). That replay has a problem the other two do not: a league that plays only itself sits at its starting rating forever, which would rank Flamengo below every European club we hold.
So it is anchored to ClubElo’s scale through the only fixtures that have ever put the two continents on the same pitch — the 79 Intercontinental Cup and Club World Cup matches between them. Fitting the offset that best explains those results gives +78.8 Elo points, with a standard error of 49.7. We publish the error because it is large: this is the least certain number in the system, and it deserves to be read as an estimate rather than a fact.
How a match is simulated
Your eleven picks are rated into a single team strength, converted to an Elo figure, and adjusted for how well the squad fits the formation you chose. Each match is then resolved from the Elo difference using a Davidson model — a standard extension of the usual win-probability formula that adds an explicit draw term, because football has draws and a model without one will systematically lie about a league season.
Probabilities are read from precomputed fixed-point tables, which is what makes a run reproducible: the same seed and the same XI produce the same season on any device, every time. A league season is 38 matches, home and away, against that league’s rated opposition. A World Cup is a group of four followed by four knockout rounds, with each round drawn from a progressively stronger slice of the field, and knockout draws settled by a shootout.
Formations, and why they score the same
The three formations are risk profiles, not difficulty settings. 3-5-2 draws more and loses less; 4-3-3 trades draws for a wider spread of wins and defeats. Each is compensated so that, at identical squad strength, they produce the same expected points — measured, not assumed, and checked to within a third of a point. What changes is the shape of your outcomes, not how many you should expect.
What we deliberately do not do
- Nothing you choose buys you a better grade. Grades are absolute thresholds. There is no re-roll, reward or setting that improves a result, because the moment there is, no two results mean the same thing.
- Narrowed boards are labelled. Restricting the reel to one league is a recognition aid, but the pools are not equally strong — measured over 3,000 drafted XIs, the spread between the strongest and weakest single-league board is about eight points. So a run drafted from a narrowed board says so on its result and everywhere its grade appears.
- Only the Daily is compared between players. It is the one mode where everyone gets the same spins in the same order, dealt by the server, with the result verified server-side. Everything else is you against the record book.
Corrections
Public data has errors in it, and so, inevitably, do we. If you find a career record that is wrong, or a rating that cannot be right, tell us — the contact address is on the about page. Corrections go into the next extraction.