Pl. 01 / 16Case study
A card-collecting and team-management game for amateur football, from District level to National 1, with the decisive chances played with a finger in 3D.
Presentation film made for this portfolio from game captures (demo account).
Football card games are about the stars. Every Sunday, 2.3 million licensed players play in about 12,000 amateur clubs in France (French federation figures, 2023-24 season), and no game is about them. I wanted a collecting and team-management game where you back your local club, from District level to National 1: open packs, pick your eleven, watch the match in 2D and play the decisive chances yourself in 3D.
Three rules from day one. Nothing real about people: fictional players, no minors, no logo or photo taken from the web. No existing game brand and no real money. And a server that trusts no one with anything of value: packs, credits, rewards. All of it had to run in 3D on a mid-range phone.
The name, the game design document, the monorepo and the engine. The engine is deterministic from the start: same seed, same match, checked under Node, Deno and three browsers.
Biggest risk first: the 3D key moment. You draw the shot with your finger, physics does the rest (spin, bounces, posts, nets), then the camera replays it in slow motion. 60 frames per second on an iPhone 13.
A simulation in phases and duels, readable 2D choreography and 159 announcer lines. 10,000 simulated matches to tune the stats, then the switch to 3D on big chances.
205 clubs and their squads drawn from a seed, generated crests, kits and sponsors, eight card templates, the 3D pack opening and our own footballer.
The server core (profiles, credit ledger, packs and cards in the database), the guest flow, the seven-tab shell, matches replayed and paid by the server, the shop, then going online with a captcha and anti-cheat limits.
A match is a seed, two frozen teams and the player’s inputs. The server replays it bit for bit before paying out credits. To make that possible, the engine uses neither free randomness nor the browser’s maths functions: it has its own.
Players are generated, all adults, and all change if the generator changes: that is a deliberate decision, like the engine fingerprint. Crests and sponsors are invented (“Bar-Tabac du Stade”) and every share image says so.
Credits are earned by playing: 300, 600 or 900 per win depending on the level, 25 per successful key moment. Every pack shows the odds per card and the odds of getting at least one.
The Mixamo dummy only wore a painted-on shirt. I replaced it with an original footballer built by a Blender script: 4,606 triangles, one draw call per player, and a free kick that drops from 93 to 62 draw calls.
Camera, gesture, physics, player walk-out: everything that affects feel has a tuning page in the game’s lab, instead of hard-coded values.
Claude Code wrote most of the code under my direction. I break down each phase, approve the plan before work starts, try every build and record the decisions. That is what let me take on a project this size in five weeks.
The hardest part was not the 3D but trust: as soon as a credit has value, the browser can no longer write anything. Making the engine deterministic on day one made everything simpler later: the server replays, slow-motion replays cost nothing and tests compare a fingerprint. Next step: open the slice to the public and have club players try it.
Colours sampled from the project visuals.