What the work actually is

You write tasks that reproduce the artifacts of a real VC seat, then evaluate how well a model handles them. A single task might be: here is a seed-stage company's data room summary — build a bottoms-up TAM, flag where the founder's top-down number breaks, and draft the two paragraphs of an IC memo that would actually change a partner's mind. You supply the gold-standard answer and the reasoning trace behind it. Later you grade model attempts against a rubric you helped calibrate, writing qualitative feedback that names the specific analytical failure rather than scoring it 3 out of 5.

The failure modes you are hunting are particular to this domain. Models pattern-match to comparable-company logic without checking whether the comp is structurally similar. They multiply a market number by an arbitrary penetration rate and call it a bottoms-up build. They produce founder assessments that read as flattering summary rather than judgment. They compute post-money ownership while ignoring the option pool refresh. Your job is to catch these consistently and explain them in a way that generalises into training signal.

What the screen looks for

micro1's process is AI-led and conversational, with follow-ups that go deeper on whatever you claim. It probes for: whether you have personally written memos that went in front of partners, whether you can walk a market sizing from first principles under pushback, whether you understand cap table mechanics beyond the summary tab, and whether your written feedback is specific enough to act on. Vague seniority claims get unwound fast. Naming real sectors, stages, cheque sizes, and the decisions you got wrong reads far better than a polished narrative.

Logistics

  • Contractor engagement, fully remote, asynchronous — no standing meetings beyond occasional calibration sessions with project leads.
  • Work is drawn from a queue; contributors typically commit 10–20 hours a week, though volume fluctuates with project phase.
  • Expect an unpaid or lightly-paid calibration exercise before live work, and rubric revisions mid-project as gaps appear across sectors and stages.
  • No AI or ML background required. Deal judgment and writing are the deliverable.