What the work involves

Day to day, you are reading Java that a model wrote and deciding whether it is correct, idiomatic, and safe to ship. That means running it, not eyeballing it — building a minimal reproduction, checking edge cases the prompt implied but did not state, and writing a short justification a reviewer can audit. Other task families ask you to go the other direction: author a realistic backend problem (a Spring Boot service with a broken transaction boundary, a Kafka consumer with a rebalance bug, a JPA N+1 that only shows under load), write the gold-standard solution, and document why plausible-looking alternatives fail.

Ratings are rubric-driven and comparative. You will often see two or three candidate answers to the same prompt and be asked to rank them with reasoning, or to label a single response for correctness, completeness, instruction-following, and code quality as separate axes. The hardest judgment calls are the ones where the code compiles, passes the happy path, and is still wrong — silent precision loss, a race that needs a specific interleaving, an `equals`/`hashCode` contract violation that only bites inside a `HashSet`.

What the platform screens for

micro1's screen is conducted by an AI interviewer, recorded, and follows up on your answers rather than moving down a script. Expect it to push on specifics: the version of Java you actually shipped on, what your service did under partial failure, how you diagnosed a particular production incident. Generic architecture talk gets probed until it either produces detail or collapses. There is typically a live or timed coding component in Java, and the conversation covers concurrency, JVM behaviour, persistence, and API design.

  • Depth in one backend stack beats surface familiarity with five
  • Willingness to say "I would have to measure that" where it is the honest answer
  • Clear written English — your rationales are the deliverable, not a byproduct

Logistics

  • Fully remote, contractor engagement, invoiced hourly or per accepted task
  • Mostly asynchronous; some projects add a weekly sync or a calibration session in a US-overlapping window
  • Volume fluctuates by project — treat this as variable-load contract work, not a guaranteed weekly quota
  • Screen recording, identity verification, and an NDA are standard before task access

Pay is stated as observed on this listing, not guaranteed. Rates within the band track task complexity and calibration scores over time.