Get paid for what you already know.
Frontier models are strikingly good at the parts of your job that are written down, and strikingly bad at the parts that aren't. We pay practising professionals to write down the second part — properly, once, with their name on the record.
Practitioners, not tutors
The distinction that matters is whether you currently do the work — or did it recently enough that your judgment is still calibrated to how it's actually done.
Finance & markets
Investment banking, private equity, equity and credit research, quantitative trading, FP&A, restructuring. Analyst through MD — seniority changes the task, not the eligibility.
Engineering, software & math
Software and systems engineering, hardware, data science, competition mathematics, formal methods, applied physics. Production experience counts for more than credentials here.
What a task actually looks like
Not survey questions. Not rating chatbot answers. A brief, the tools you'd normally use, and a bar you can see before you start.
# a representative brief
task LYD-4471 · finance.lbo
brief Build a five-year LBO from the attached financials and term sheet.
Include a debt schedule with a cash sweep.
tools Spreadsheet workbench · filings corpus · your own working notes
you'll be 1. Every output cell traces to an input or a stated assumption
checked on 2. The sweep repays in the correct order of priority
3. Every hardcoded figure carries a source note
4. The model balances and the IRR reconciles
also asked A short note on what you'd flag to a deal team, and what
you deliberately didn't model.
estimate 90–150 minutes
You see the checks before you start. There is no hidden rubric and no mystery scoring. If you fail a check, you're told which one and you can revise — that's the part contributors tell us is different from every other data platform they've tried.
Four steps, and we pay for the trial
We're deliberately slow at the front and fast afterwards. Screening properly once is what lets us skip constant supervision later.
Apply
Email us with your background and the domains you'd work in. A CV or a LinkedIn profile is enough to start.
Credential check
We verify employment history and professional registration where the domain has one. Everything is under NDA from this point.
Paid trial task
One real task, graded against the production bar. It is paid whether or not it passes, because your time is worth the same either way.
Calibration
If it passes, you're onboarded to a domain track. Your first few submissions are reviewed closely, then you work at your own pace.
What contributors ask
Fair questions, all of them.
What does it pay?
Per task, priced by domain and difficulty, and quoted before you accept a brief — you always know what a task pays before you start it. Rates rise with demonstrated quality rather than with time served. We publish exact rates to applicants at the credential-check stage rather than on a public page, because they vary widely by track.
How much time does it take?
It's project-based and genuinely flexible. Most contributors do this alongside a full-time role, taking a handful of tasks a month. There's no minimum commitment and no penalty for going quiet for a while.
Will this conflict with my employer?
We design tasks so that you're never asked to reproduce your employer's proprietary material, use their systems, or disclose confidential information. Tasks are built from public sources and constructed scenarios. That said, check your own employment agreement — outside-work clauses vary, and that call is yours to make.
Who sees my work, and does my name go on it?
Your work is delivered to the client that commissioned it. Records carry your credential class — for example "VP, sponsors coverage" — not your name. You're identifiable to us, not to the client, unless you ask otherwise.
What happens if my submission is rejected?
You're told exactly which check failed and why, and in most cases you can revise and resubmit. Rejection isn't a strike — the graded feedback loop is how calibration happens. Contributors whose work is consistently rejected without improving do stop receiving briefs, and we'll tell you plainly if that's happening.
Is my work training AI to replace me?
The honest answer is that these systems are being built either way, and the version built without practitioners is the more dangerous one — confidently wrong in ways only someone who does the job can spot. We think the people who understand the work should be the ones defining what "correct" means for it. If you disagree, we'd genuinely rather you didn't apply.
Put your expertise on the record
Tell us what you do and where you do it. We read every application ourselves.