“I once screened 400 resumes and interviewed 30 people, and the sameness troubled me more than the weaknesses. If the variable names match, I would retire that take-home and watch them solve a live problem instead: shared code hides, but shared thinking shows.”
“I would treat identical variable names as a strong signal, not a coincidence. At Eightfold we used take-homes as one input among many, and when something smelled off we would either bring the candidate in for a live coding walkthrough or ask them to explain their design decisions line by line. If they cannot, the test has already failed its purpose.”
“I treat a take-home as a screen, not a verdict: if three strangers hand in the same variable names, the signal is about the test, not the candidates. I would move the people who can explain the code live into the next round and retire the leaked prompt.”
“I have seen the same thing from the other side. After reviewing 1,054 SWE resumes, I can spot when a candidate's output has the same scaffolding as everyone else's. Variable names matching across strangers is the tell: either your test is on a public repo, or an LLM is writing the answer from a shared prompt. I would change one variable in the problem statement and see if the next batch still matches.”
“Same typos, same broken line, same formatting artifact from the same parser: I have seen that in resumes, and it means nobody read what the tool made. If the variable names match too, I would bet on a shared generator or leaked answer set before coincidence. I would move them all to a live follow-up and ask them to walk through one decision in the code like I am not technical.”
“I would stop treating the take-home as a detective game and start treating it as a broken signal. Three identical submissions means your test is either public or easy enough that one tool can crack it, and in either case you are not measuring what you think you are measuring.”
“I built Deliveroo's recruiting coordination from scratch, and the one thing I learned is that a take-home stops being useful the moment it becomes a test of who found the same template. If three strangers are handing you identical code, your signal is already gone, so I'd retire that prompt and replace it with a live pairing exercise where they have to reason out loud.”
“I would stop treating the take-home as a signal of skill and start treating it as a signal of access. Same variable names across strangers means the prompt, a solution bank, or a model shortcut has entered the wild, and your test is now measuring who found it, not who can build it.”