Pretty sure this is a more complete, thought-out superset of ideas.

  • no 15-hour task, no 5-min chat. Something in-between.
  • assuming you have some closed source code, maybe give them a function signature to re-implement the body
  • ~1 week “audition” contract (solve-this-real-issue/ PR (pull request
    )     )
    • my take: this is great however IMO not scalable and doesn’t allow you to compare candidates via identical tasks… it may only work if there’s nobody else to compare to. Work-around: have closed source code at a frozen timepoint to ask them to open PRs (pull request
      )       on
  • don’t standardise things irrelevant to what you’re measuring