What the work actually involves
You are handed a rubric and a queue. Day to day you open task outputs — data tables, labeled records, model-generated deliverables — and judge each one against explicit criteria: is it accurate, is it complete, does it follow the instructions it was given. Where it fails, you say precisely where and why in writing that someone else can act on. Where the rubric doesn't cleanly cover the case, you make a call and document the reasoning behind it rather than skipping the item.
A meaningful part of the job is pattern recognition rather than item-by-item correction. If forty records in a batch share the same defect, the useful output is an escalation describing the defect class — not forty individual flags. Reviewers who treat every item as isolated tend to be slower and less valuable than reviewers who notice structure. Expect high volume and genuine repetition; consistency on item 400 is what the client is paying for.
What the screen looks for
micro1's screening is AI-led and conversational. It probes for concrete history in data review, QA, or analytics roles — specific datasets, specific error types, specific tooling — and then tests judgment with scenarios where the rubric is ambiguous or two criteria conflict. Spreadsheet fluency (sorting, filtering, basic formulas) is a baseline expectation; SQL and prior annotation or rubric-based evaluation experience are treated as advantages, not requirements. No prior AI work is needed.
Logistics
- Contractor engagement, remote, US-based.
- Largely asynchronous queue work with some written and verbal feedback loops; occasional calibration discussion.
- Hours are flexible but the client values sustained throughput — candidates who can commit predictable weekly blocks tend to stay staffed.
- Observed pay band is $30–60/hr; rates are set per engagement and not guaranteed.