Browser Game

Tank Battle Challenge

A multi-entity browser game case for movement, enemy logic, projectile collisions, and visible combat feedback.

GameCanvasAI enemiesProjectile collisionMobile input

Watch episode

Published video

Tank Battle comparison episode

This provisional comparison episode stitches together the current GLM, Qwen, Doubao, and DeepSeek tank battle runs recorded under the Claude Code framework. Scoring is based primarily on the full gameplay clips, with screenshots used as anchor frames for menu, combat, and failure states.

2m 17s

Watch full episode Watch comparison short Watch best-run short Watch worst-run short

Result matrix

One task, 4 recorded outcomes

The table shows visible results first. Detailed logs and clips can be attached once real runs are recorded.

Agent	Model	Status	Score	Breakdown	Time	Cost	Failure
Claude Code	GLM	Playable	A-84	Runs without fatal errors18/20 Core functionality25/30 Combat reliability18/20 Interaction quality14/20 Code sanity9/10	Review pending	unknown	It still does not prove especially deep restart coverage or a wide range of late-stage combat situations, so the lead is solid but not overwhelming.
Claude Code	Qwen	Playable	B+78	Runs without fatal errors17/20 Core functionality24/30 Combat reliability15/20 Interaction quality14/20 Code sanity8/10	Review pending	unknown	Enemy behavior and battlefield variety look more repetitive than the top run, so it feels dependable but less fully realized under longer play.
Claude Code	Doubao	Partial	C60	Runs without fatal errors14/20 Core functionality18/30 Combat reliability11/20 Interaction quality11/20 Code sanity6/10	Review pending	unknown	A captured blank-field state and shakier consistency across the clip make visual stability and sustained combat reliability harder to trust.
Claude Code	DeepSeek	Partial	D46	Runs without fatal errors13/20 Core functionality14/30 Combat reliability8/20 Interaction quality7/20 Code sanity4/10	Review pending	unknown	Direction handling, thin enemy pressure, and weaker combat feedback leave too much of the real gameplay loop underdeveloped compared with the stronger runs.

Claude Code

GLM

A-

Status: Playable
Score: 84
Breakdown: Runs without fatal errors18/20
Core functionality25/30
Combat reliability18/20
Interaction quality14/20
Code sanity9/10
Time: Review pending
Failure: It still does not prove especially deep restart coverage or a wide range of late-stage combat situations, so the lead is solid but not overwhelming.

Claude Code

Qwen

B+

Status: Playable
Score: 78
Breakdown: Runs without fatal errors17/20
Core functionality24/30
Combat reliability15/20
Interaction quality14/20
Code sanity8/10
Time: Review pending
Failure: Enemy behavior and battlefield variety look more repetitive than the top run, so it feels dependable but less fully realized under longer play.

Claude Code

Doubao

Status: Partial
Score: 60
Breakdown: Runs without fatal errors14/20
Core functionality18/30
Combat reliability11/20
Interaction quality11/20
Code sanity6/10
Time: Review pending
Failure: A captured blank-field state and shakier consistency across the clip make visual stability and sustained combat reliability harder to trust.

Claude Code

DeepSeek

Status: Partial
Score: 46
Breakdown: Runs without fatal errors13/20
Core functionality14/30
Combat reliability8/20
Interaction quality7/20
Code sanity4/10
Time: Review pending
Failure: Direction handling, thin enemy pressure, and weaker combat feedback leave too much of the real gameplay loop underdeveloped compared with the stronger runs.

What this tests

The hidden engineering inside a simple game

Movement and controls

Handles directional movement, firing, and control responsiveness across supported inputs.

Projectile collision

Resolves bullets, walls, tanks, and obstacle interactions reliably.

Enemy behavior

Implements enough enemy logic to create an actual combat scenario.

Game-state management

Keeps score, health, win, loss, and restart transitions coherent.

Visual clarity

Makes tank direction, hits, and combat state readable on screen.

Debug iteration

Requires the agent to use browser feedback to fix visible gameplay bugs.

Unified prompt

Build a playable top-down tank battle browser game. It should include player movement, aiming or directional firing, enemy tanks, projectile collisions, destructible or blocking map elements, score or win-state feedback, game over, and restart. Support keyboard input on desktop and pointer or touch controls on mobile where appropriate. The final result must render visibly, avoid fatal runtime errors, and reset cleanly after restart.

Acceptance criteria

The app starts without fatal runtime errors.
The battlefield, player tank, and at least one enemy tank render visibly.
The player tank can move and fire with responsive controls.
Projectiles collide with walls, tanks, or other required targets correctly.
Enemy behavior produces an actual playable combat loop.
Score, health, game over, win-state, and restart logic work.
Restart clears stale state and starts a fresh playable round.
The final page does not produce console errors during normal play.

Scoring

Readable rubric for video and reproduction

Runs without fatal errors

The app starts, renders, and can be interacted with.

Core functionality

Movement, firing, enemy behavior, collisions, and restart work.

Combat reliability

Hits, damage, and battlefield interactions behave consistently.

Interaction quality

Controls feel usable and the combat loop is readable.

Code sanity

The implementation is understandable and avoids obvious brittle shortcuts.

Run details

What each agent got right and wrong

GLM gameplay clipWatch clip

Screenshots

Menu Progress Game over

Claude Code

GLM

A-84

Runs without fatal errors18/20

Core functionality25/30

Combat reliability18/20

Interaction quality14/20

Code sanity9/10

Status: Playable
Strength: The clip shows the best overall balance of menu clarity, battlefield readability, enemy pressure, pickups, and end-state feedback in the current set.
Failure: It still does not prove especially deep restart coverage or a wide range of late-stage combat situations, so the lead is solid but not overwhelming.

Qwen gameplay clipWatch clip

Screenshots

Menu Progress Game over

Claude Code

Qwen

B+78

Runs without fatal errors17/20

Core functionality24/30

Combat reliability15/20

Interaction quality14/20

Code sanity8/10

Status: Playable
Strength: The clip keeps the combat state readable throughout, with a strong side HUD, clear enemy counters, stable movement, and one of the most legible game-over summaries.
Failure: Enemy behavior and battlefield variety look more repetitive than the top run, so it feels dependable but less fully realized under longer play.

Doubao gameplay clipWatch clip

Screenshots

Menu Progress Game over Bug state

Claude Code

Doubao

C60

Runs without fatal errors14/20

Core functionality18/30

Combat reliability11/20

Interaction quality11/20

Code sanity6/10

Status: Partial
Strength: The run shows an ambitious presentation layer with clear rules text, visible pickups, active enemy fire, and a readable game-over state when the battlefield is working.
Failure: A captured blank-field state and shakier consistency across the clip make visual stability and sustained combat reliability harder to trust.

DeepSeek gameplay clipWatch clip

Screenshots

Menu Progress Bug state

Claude Code

DeepSeek

D46

Runs without fatal errors13/20

Core functionality14/30

Combat reliability8/20

Interaction quality7/20

Code sanity4/10

Status: Partial
Strength: The clip at least establishes a visible battlefield, multiple stages, readable HUD text, and enough movement and shooting to show the intended structure.
Failure: Direction handling, thin enemy pressure, and weaker combat feedback leave too much of the real gameplay loop underdeveloped compared with the stronger runs.

Postmortem

Why Tank Battle Is a Strong Follow-up Visual Case

More entities, more failure points

Tank battle adds enemies, bullets, obstacles, and directionality. That creates more state coupling than simple paddle-and-ball games.

Combat exposes logic gaps quickly

Visible projectile misses, stuck enemies, or broken restart behavior are easy to see in a short comparison clip.

The case is ready for evidence

This draft page now carries first-pass run clips, screenshots, and a generated comparison episode. The current scoring pass is driven by full-motion gameplay evidence first and still remains provisional.

Case package

starter/
tests/
prompt.md
task.md
scoring.md
runs.json
screenshots/
clips/
logs/

Manifest Prompt Scoring Runs JSON

Commands

npm install
npm run dev
npm test

Use this scaffold to collect prompt, scoring, screenshots, clips, logs, and final run summaries before marking the case as published.

Tank Battle Challenge

Tank Battle comparison episode

One task, 4 recorded outcomes

Claude Code

Claude Code

Claude Code

Claude Code

The hidden engineering inside a simple game

Movement and controls

Projectile collision

Enemy behavior

Game-state management

Visual clarity

Debug iteration

Readable rubric for video and reproduction

Runs without fatal errors

Core functionality

Combat reliability

Interaction quality

Code sanity

What each agent got right and wrong

Claude Code

Claude Code

Claude Code

Claude Code

Why Tank Battle Is a Strong Follow-up Visual Case

More entities, more failure points

Combat exposes logic gaps quickly

The case is ready for evidence

Ready for the next visual task?