Documentation

Getting Started

1. Pick a challenge — Browse problems pulled from real open-source repos. Bugs, features, refactors. All based on code that actually shipped.
2. Solve with AI — You get a live editor and a built-in AI assistant. Use it however you want.
3. Pass the tests — Submit your solution and it runs against the test suite. All users see which tests pass and fail. Pro users unlock the full AI collaboration score with dimension breakdowns, badge tiers, and personalized coaching insights.

AI Collaboration

Encouraged

Use AI as a thinking partner. Ask it to explain things, suggest approaches, review your code, help you debug. The point is learning how to direct AI, not avoiding it.

Discouraged

Copy-pasting the whole problem into AI and submitting whatever comes back. The challenges require judgment. AI alone won't get you a good score.

Built-in AI assistant: Every challenge comes with an AI assistant (KIMI K2 Instruct) built in. It sees your full session context including the file you're editing, your conversation history, and the problem brief.

Challenge Formats

• Bug Fix — There's a bug. Tests are failing. Find the root cause, fix it, and change as little as possible.
• Feature Build — You get a spec and an existing codebase. Build the feature. You're scored on how well it fits with what's already there.
• Refactor — The code works but it's messy. Clean it up. Make it faster, more readable, or better structured. Tests still need to pass.

Scoring

Every user gets test results. Pro users get the full picture.

Free

Submit your code and see which tests pass and which fail. You get the full test output with error messages so you know exactly what went wrong. Every challenge, no limits.

Pro

Get the full AI Collaboration Quotient (ACQ) score. Seven dimensions, 100 points total. Plus a badge tier, AI-powered coaching insights, and personalized recommendations to improve how you work with AI.

Pro scoring dimensions (100 points total)

• Comprehension (15 pts) — Did you explore the codebase before jumping in? Reading files, understanding structure, and forming a plan before prompting AI.
• Scoped Delegation (15 pts) — Were your prompts specific and context-rich? Telling AI exactly what you need with the right constraints.
• Critical Review (20 pts) — Did you actually read and improve what AI gave you? Modifying suggestions, catching issues, making the code yours.
• Verification (20 pts) — Did you test frequently and debug when things broke? Running tests early and often instead of just reprompting.
• Communication (10 pts) — Did you narrate your thinking and discuss tradeoffs with AI? Clear intent leads to better suggestions.
• Autonomy (10 pts) — Did your AI dependency decrease over time? Starting with AI help and gradually taking over shows real learning.
• Correctness Gate (10 pts) — Do the tests pass? This is the baseline. Everything else builds on working code.

Badge tiers: Pro users earn badges based on their ACQ score: Just Vibing (0-49), Needs Work (50-69), On Your Way (70-84), and AI Collaborator (85-100).

FAQ

Is Sponge free?

Every challenge is free. All of them. You always see your test results (pass/fail). If you want the full AI collaboration score, dimension breakdowns, badge tiers, and AI-powered coaching insights, those are on Pro.

What does Pro include?

Pro unlocks the full AI Collaboration Quotient (ACQ) score after every challenge. You get a breakdown across 7 dimensions, a badge tier, AI-powered coaching insights that explain your strengths and weaknesses, and personalized recommendations for improvement.

What languages are supported?

Python, Java, and C++. More languages coming soon.

Is there a time limit?

Yes. Each challenge has a time limit shown before you start (usually 30-60 minutes). Your code auto-submits when time runs out. The timer only runs while you're actively coding, not during the brief.

Can I retry a challenge?

You can start a new session on any challenge at any time. Each attempt is scored independently. Your best score is the one that shows on your profile.

What AI model powers the copilot?

KIMI K2 Instruct by Moonshot AI. It sees your full session context including the file you're editing, your conversation history, and the problem brief. It's the same model for every user on every challenge.

Where do the challenges come from?

Open-source repos. We take real bugs, real features, and real refactors from public codebases and package them into problems. Nothing is made up.

What happens if I lose connection?

Your session is saved automatically every 30 seconds. If you disconnect, you can resume right where you left off from the Explore page. Your code, conversation history, and timer are all preserved.

Can teams use Sponge?

Yes. Sponge Enterprise lets teams send challenges to candidates or team members, with structured scoring and reports. Reach out at the Enterprise page to set up a demo.