Debugging & Codebase Rounds
More companies now test what most of the job is: reading code you didn't write, finding why it misbehaves and fixing it cleanly. This line takes you from how these rounds are scored, through small planted bugs, to full 45–60 minute mock rounds, in JavaScript, Python, MERN, SQL and Python backends, with read-only Java, C++ and Go. Your code runs in your browser and every check is a real test. Any AI in AI-allowed rounds is optional and uses your own key from your browser.
8 paths · pick one language or stack and stay in itChoose your path
Each path is complete on its own. It starts with the same ten lessons on method and scoring, then stays in one language or stack all the way through. You can switch paths at any time, and your progress in each is kept.
Runs in your browser JavaScript
Plain JavaScript and Node: snippet drills, multi-file bugs, failing-test repos, real open-source packages and every mock.
11 lessons · 26 practice stations
Runs in your browser Python
Python: snippet drills, multi-file bugs, a stock-sync repo, a three-phase round and timed mocks.
11 lessons · 20 practice stations
Runs in your browser MERN
Express, Mongoose and React: middleware order, async errors, stale state, request races and API shapes.
11 lessons · 12 practice stations
Runs in your browser Python backend
FastAPI-style services: shared request state, missing awaits, status codes and cleanup.
11 lessons · 6 practice stations
Runs in your browser SQL
Real Postgres in your browser: joins, NULLs, grouping, paging, indexes and transactions.
11 lessons · 7 practice stations
Read-only · works on a phone Java
Read-only Java and Spring Boot: classic snippets, then a whole service to read and a bug to find in it.
11 lessons · 4 practice stations
Read-only · works on a phone C++
Read-only C++: undefined behaviour, lifetimes and containers, then a whole program to read and a bug to find in it.
11 lessons · 4 practice stations
Read-only · works on a phone Go
Read-only Go: nil maps, loop variables, goroutines and errors, then a whole service to read and a bug to find in it.
11 lessons · 4 practice stations
What these rounds test
Find the code that matters in an unfamiliar repo within minutes, without reading it top to bottom.
Reproduce, map, hypothesise, probe, fix, guard: a protocol that earns credit even when time runs out.
The smallest change at the root cause, a regression test that proves it, and a clear account of why.
It's most useful if you're aiming at FAANG-tier companies or startups like Stripe that run codebase and debugging rounds. Many Indian product-company processes don't include them yet, so check your target company's loop before putting this ahead of DSA and system design.
Interview formats and where they are used
| Format | Company | Our practice | What you get | Where it's hard | Sources |
|---|---|---|---|---|---|
| Snippet debugging assessmentA timed set of small functions, each hiding one logic bug. You read, spot the mistake and fix it before the clock runs out. A useful warm-up for every format below; we found no current official page that ties it to a named company, so we don't name one. | Practice format · no company named | ~20 minAuto-graded | 5–7 short code snippets, one planted logic bug in each | Speed and accuracy: little time per snippet, no one to ask | Last verified: October 2026 |
| Bug squashYou and an interviewer work side by side on a real historical bug in an open-source project, in your programming language of choice, on your own laptop and tools. | StripeConfirmed: The company's own pages describe this round | ~60 minLive with an interviewer | A real open-source project and a real past bug | Finding your way around a large codebase you have never seen | stripe.com ↗ (opens in a new tab)Stripe's own guide describes the round; the guide is undated and not a candidate prep page, so check the current details with your recruiter.Last verified: October 2026 |
| Failing-test repoA small repository written for the interview, with one or more bugs and some red tests. You turn the tests green while explaining what you find. Candidates describe this at several startups, but no company's own page confirms it, so we don't name one. | Practice format · no company named | ~45 minLive with an interviewer | A small custom repo, one or more bugs, failing tests | Working methodically and talking through it while being watched | Last verified: October 2026 |
| Code comprehensionPress reports of an internal Google document describe a pilot round on select US teams for early- and mid-career roles: you analyse an existing codebase to find bugs and improve it, with Gemini allowed. Google's recruiting VP confirmed a pilot but not its details. | Google (pilot)Widely reported: Several reputable outlets report it; no official page yet | ~45 minLive with an interviewer | An existing codebase to read, debug and improve | Understanding someone else's design quickly and checking what the AI tells you | entrepreneur.com ↗ (opens in a new tab)finance.yahoo.com ↗ (opens in a new tab)No official Google page describes this round yet. Both articles cite Business Insider's report on an internal document and a statement from Google's VP of recruiting.Last verified: October 2026 |
| AI-assisted codebase roundMeta says many of its technical interviews now include an AI assistant built into CoderPad that candidates are expected to use, and recommends being comfortable reading, debugging and building on existing code. Our practice version splits the hour into fix, extend and optimise, as candidates describe it. | MetaConfirmed: The company's own pages describe this round | ~60 minLive with an interviewer | Existing code, a built-in AI assistant and tasks that build on each other | Pacing yourself and checking every AI suggestion before you trust it | metacareers.com ↗ (opens in a new tab)The AI assistant and the focus on existing code are on Meta's page; the three-part split is from candidate reports.Last verified: October 2026 |
- Snippet debugging assessmentPractice format · no company namedPractice: ~20 min · Auto-gradedA timed set of small functions, each hiding one logic bug. You read, spot the mistake and fix it before the clock runs out. A useful warm-up for every format below; we found no current official page that ties it to a named company, so we don't name one.
- What you get
- 5–7 short code snippets, one planted logic bug in each
- Where it's hard
- Speed and accuracy: little time per snippet, no one to ask
- Bug squashStripeConfirmed: The company's own pages describe this roundPractice: ~60 min · Live with an interviewerYou and an interviewer work side by side on a real historical bug in an open-source project, in your programming language of choice, on your own laptop and tools.
- What you get
- A real open-source project and a real past bug
- Where it's hard
- Finding your way around a large codebase you have never seen
- Failing-test repoPractice format · no company namedPractice: ~45 min · Live with an interviewerA small repository written for the interview, with one or more bugs and some red tests. You turn the tests green while explaining what you find. Candidates describe this at several startups, but no company's own page confirms it, so we don't name one.
- What you get
- A small custom repo, one or more bugs, failing tests
- Where it's hard
- Working methodically and talking through it while being watched
- Code comprehensionGoogle (pilot)Widely reported: Several reputable outlets report it; no official page yetPractice: ~45 min · Live with an interviewerPress reports of an internal Google document describe a pilot round on select US teams for early- and mid-career roles: you analyse an existing codebase to find bugs and improve it, with Gemini allowed. Google's recruiting VP confirmed a pilot but not its details.
- What you get
- An existing codebase to read, debug and improve
- Where it's hard
- Understanding someone else's design quickly and checking what the AI tells you
- AI-assisted codebase roundMetaConfirmed: The company's own pages describe this roundPractice: ~60 min · Live with an interviewerMeta says many of its technical interviews now include an AI assistant built into CoderPad that candidates are expected to use, and recommends being comfortable reading, debugging and building on existing code. Our practice version splits the hour into fix, extend and optimise, as candidates describe it.
- What you get
- Existing code, a built-in AI assistant and tasks that build on each other
- Where it's hard
- Pacing yourself and checking every AI suggestion before you trust it
We name a company only when its own careers, prep or engineering pages describe the round ("Confirmed") or reputable outlets report it ("Widely reported"). Processes change by team, level and country: ask your recruiter what your round looks like.