EXP-003PlannedGame

3D Basketball Game

Five-on-five basketball in the browser, with real ball physics, keyboard control and an AI opponent.

Time budget
8h per agent
Spec
v1.0 committed Nov 8, 2026
Agents
3 runs not started
Runs recorded
—
Evaluation
Pending

01 The hypothesis

The hypothesis

Real-time 3D games combine physics, AI behavior, input handling and game rules in one loop. That makes them a stress test of whether an agent can integrate many systems into something that is actually playable.

Questions this experiment asks

  1. Is it a game you can play, or a scripted scene?
  2. Does scoring come from real geometry, with exactly 2 or 3 points and no duplicates?
  3. How capable is the AI opponent on offense and defense?

The challenge

No engine is required and no assets are supplied. Each agent picks its own stack, creates or sources every model and sound, and builds rules that hold up under a 15-step manual test script. The evaluator checks that a two-pointer adds exactly 2, a three-pointer exactly 3, misses don’t score, the clock stops when it should, and a restart really resets the game.

02 The rules

The challenge, as specified

A human can open the deployed URL, start a game, control a team, pass, shoot, score, defend against an AI opponent, see possession change and finish a timed game with a final score.

Required scope Spec: A requirement from the versioned specification.

  • Real-time 3D court with two baskets, markings and ten players in distinct team colors
  • Keyboard control: move, pass, shoot, switch player, sprint, defend and steal
  • A shot timing, aim and power mechanic that affects whether a shot goes in
  • Ball physics with gravity, bounces, and rim and backboard interactions
  • Two- and three-point scoring decided by geometry, plus possession changes, rebounds, steals and blocks
  • Game clock, quarters, shot clock, HUD, menus, pause, restart and a final result screen

Constraints Spec: A requirement from the versioned specification.

  • Eight hours of elapsed development time per agent
  • No human coding, asset preparation or debugging beyond setup, credentials and unavoidable approvals
  • Fictional teams only, with no league branding, real player likenesses or ripped assets
  • No hardcoded demos, canned sequences or scripted score changes
  • Runtime LLMs are not required; state machines and heuristics are fine

Deliverables Spec: A requirement from the versioned specification.

  • Source code, README.md and BUILD_REPORT.md
  • Public deployed URL, with Cloudflare preferred
  • ASSETS.md manifest with licenses and attribution
  • A requirement checklist marking each item PASS, PARTIAL, FAIL or NOT TESTED
  • The goal is an actual playable game, not a static scene, scripted animation, or non-interactive prototype.spec v1.0
  • A smaller complete game is preferable to an elaborate scene that cannot be played.spec v1.0

03 The contenders

Who will take this on

The spec names these agents. Runs haven't started, so there are no repositories, timelines or results yet. Each run will get its own page here once it's recorded.

04 Head-to-head evaluation

Evaluation pending

Scoring rubric · 100 points

Spec: A requirement from the versioned specification.
  1. Gameplay and controls30
  2. Basketball rules and AI25
  3. Physics and ball interactions20
  4. Graphics and animation15
  5. Completeness and stability10
  • Eligibility gate: a build that cannot launch, or cannot show interactive ball movement and scoring, counts as an incomplete game.

05 Evidence

Inspect the artifacts yourself

06 Observations & verdict

What we can say so far

No observations yet. They're recorded during the runs.

No verdict yet.

A winner is only named once every build has been independently tested against the shared criteria, with the score components shown.

Episode

Not published yet

The video for this experiment will be embedded here once it is published.

Get notified on YouTube (opens in a new tab)

Explore more

Other experiments

Discover screen of the Codex-built travel planner, "Elsewhere", from a screenshot the agent committedEXP-001
Runs recordedWeb app

AI Travel Planner

Three coding agents, one 1,245-line product spec, eight hours each. Build a real AI travel SaaS and deploy it.

  • CCClaude Code
  • CXOpenAI Codex
  • CUCursor CLI
Time budget
8h / agent
Spec version
Spec v1.7
Runs
Runs not started
EXP-002
PlannedCreative tool

AI Video Editor

A real browser video editor with a multitrack timeline, AI captions, plain-English edits and a genuine MP4 export.

  • CCClaude Code
  • CXOpenAI Codex
  • CUCursor CLI
Time budget
8h / agent
Spec version
Spec v1.1
Runs
Runs not started
EXP-004
PlannedGame

Project Bulletstorm

A stationary 3D lane shooter where you steer an auto-firing squad, recruit soldiers and survive ten levels.

  • CCClaude Code
  • CXOpenAI Codex
  • CUCursor CLI
Time budget
8h / agent
Spec version
Spec v1.0
Runs
Runs not started