# BugDr > Source: > Published: 2026-11-01 17:01:43+00:00 ### Free €0/month - All difficulty levels. - 150,000 tokens / day. - AI assistant: Claude Haiku. - Efficiency scoring. - Public leaderboard. [Start now](https://app.bugdr.app) AI can write your code. challenges you to understand it, find what broke, and prove your fix works. The belief Something broke. A customer is waiting. Your first fix looks convincing. Does it actually work? gives you practice codebases built around failures you could face at work. Read the report. Follow the evidence. Use AI. Test your changes. You own the fix. Why Coding puzzles ask you to solve a defined problem. asks you to investigate what went wrong in an existing codebase. | Aspect | Coding puzzle | | |---|---|---| | The task | Solve a standalone problem. | Investigate a failure in existing code. | | AI | Rules depend on the assessment. | Available in the workspace. | | The work | Build a solution from the prompt. | Read, diagnose, change, and test. | | The checks | Does the solution produce the expected output? | Is the failure fixed without breaking the rest? | | What you show | Your solution. | Your fix and the investigation behind it. | How it works Start with the bug report and explore the codebase. Learn what should happen and what happens instead. The cause is yours to find. The timer starts when you start the case. The editor, terminal, and AI assistant run in your browser. Trace the failure. Ask questions. Try a focused change. Run the code. Your prompts, token usage, and test runs are recorded during the session. Run the acceptance checks. Solve the reported problem and keep the rest working. Your first successful result is saved. Replaying the case won't replace it with a higher score. Example case A retry isn't a fix if it charges someone twice. Illustrative case and acceptance checks. Session feedback After a solve, your session debrief looks at your prompts, tokens, and test runs. Compare your approach with other participants on that case and find something useful to try next time. A working fix matters. So does learning a process you can repeat. Example debrief | Measure | You | Comparison group | |---|---|---| | Prompts | 12 | 4 | | Tokens | 3,840 | 1,200 | | First-run pass | No | 68% | Eight test runs before the first pass. Next time, inspect the logs before asking AI for another change. On your next case Read the full stack trace. Identify the likely failure point. Ask for one targeted change. Test the hypothesis. Illustrative feedback and benchmark data. Fewer tokens alone do not mean a better fix. Career paths Choose a target role and follow a structured path through relevant cases. Start with focused bugs, then take on failures that require a deeper investigation. “Get a job” is a stage label, not an employment guarantee. Stage 01 Small bugs. Better instincts. Trace the failure. Make a focused change. Check the result. Pass the acceptance checks and meet the path's efficiency benchmark to move forward. Illustrative cases Roles The connection pool runs out under load. Find out why. The page keeps rendering. The browser keeps slowing down. Two users. One slot. Somehow both booked it. The dashboard is slow. The queries tell the story. The summarizer sends more context than the model can handle. The service keeps running out of memory. Restarting it isn't the investigation. Problem library Duplicate charges. Slow queries. Failed deployments. Browse by role, topic, and difficulty. Every case gives you a brief and acceptance checks, so you know what the fix needs to achieve. More cases coming at launch, October 21. Contests and dashboard Daily, weekly, and monthly contests put participants on the same case. Bring your own approach and see how it compares. Your dashboard keeps your next case and progress in view. Hackathons On special occasions, runs hackathons: every participant gets the same case at the same time. Your fix, your approach, your efficiency — measured against everyone else's. Prizes for the top solvers. Progression Progress through 's practice ranks as you complete cases. A rank reflects your progress on the platform. It does not certify your job level or overall engineering ability. Find your footing in unfamiliar code. Build a reliable investigation habit. Work through more demanding failures. Practice cases with deeper interactions. Look beyond the symptom to the system. Take on the platform's toughest cases. One verified fix at a time. Pricing €29.99/month or €239/year ~€19.92/month when billed annually. Save 33%. €199/month Every Claude model, one daily budget. €12.99/month For engineers who run their own AI stack. Tokens measure how much of the codebase the AI reads. Better engineers use fewer. Payments open after launch. Waitlist opens October 21, 2026. Leave your email for a reminder the night before and a link when it's live. FAQ Yes. Use the workspace assistant to explore code, ask questions, and work on a fix. You still need to understand the changes and check that they work. Use AI. Own the code. They're practice codebases built around failures you could encounter at work. You get incident context and room to investigate without touching a live customer system. The listed library stacks include Node.js, TypeScript, Python, Go, React, Next.js, PostgreSQL, Redis, JWT, and WebSockets. Check the individual case before starting. No. The editor, terminal, AI assistant, and tests run in your browser. A working, reliable fix matters most. Scoring also considers code quality, testing, time, and AI usage. Fewer tokens or fewer lines do not automatically mean better engineering. Your first successful result is final. Take the lesson into your next case. Career paths, AI feedback after each solve, a higher daily token limit, and the listed Pro model options. €29.99 monthly or €239 yearly. A path gives you relevant practice and benchmarks. It cannot promise employment. Hiring decisions belong to employers.