×
Synthetic demoReal jobs for
what you're already great at.
Some of these job shapes are already paid in adjacent AI markets. Others are still hypotheses we're testing. Every one is directed work: you do a task, you get paid for your verified performance, and you keep what you make. Pick the missions that fit your games and your rank. The better you are, the better the work.
Buying, not playing? The lab-facing page is here
Verified skill is what keeps you in the expert tier that pays by the hour, instead of the commodity piecework that pays pennies.
CS2 expert sessions, in the game you already main
One paid session, about two hours, built around the queue you already grind. You play your normal Premier or FACEIT matches with your POV and your own mic recording, then hand us the match share code so we can pull the official demo and stats. A short calibration set on our trainer anchors your mechanical baseline. Talk through your reads live if that's natural for you, or stay quiet and do it afterward in the replay review, where a coach pulls up six or eight of your decisions and you walk through what you saw, what you considered, and why you took the shot you took. Either way you're paid for the time, EG coaches score the session, and the panel includes a Major-winning former pro and coach. Your results stay private.
Pro or ex-pro, the anchors of the cohort
FACEIT level 10 or high Premier rating, recently active
Experienced and credible, building toward the top tier
Six players from the first cohort continue into a two-week coached sprint: one named weakness, one coach, a short practice plan, and a before-and-after measurement. That pays another $200 on top of the session rate.
Getting paid is boring on purpose. You onboard once through EG's contractor workspace, the same one EG already uses to pay people: tax forms handled, invoice generated for you when your session is accepted, payment against that invoice. Wages are never paid in points or credits, and there's no marketplace balance to wait on.
We send a recording profile. One import, done forever. It puts your mic on its own track, so only your voice is captured.
Normal Premier or FACEIT. Hit record. If you already stream or clip, this is nothing new.
One copy-paste from the CS2 client. We pull the official demo and line it up with your recording.
Sit down with a coach and talk through your own decisions. Most players say this part is the fun part.
What we record, and what you control. The official match demo from Valve or FACEIT, your POV, your own mic, and the transcript, all tied to your verified account. Your mic only: teammates' voices are never captured unless your whole five-stack signs up and consents together. You sign a consent that says exactly who can see what before anything is captured. No public leaderboard, your identity is never sold, and the one-click delete works here the same as everywhere else on Meijindex.
Ten ways your skill gets paid.
Left column of each card is the job. Right column is why a lab writes the check, which is the question most “get paid to game” pitches quietly skip. And every rate is labeled with where it comes from: posted means a live public posting pays this today, anchored means a named buyer publicly pays these rates for this shape of work, and modeled means the demand is documented but the rate is our estimate against the posted anchors.
Queue up against an AI agent and play it straight. Win or lose, every round you play is recorded as the human benchmark.
Labs need a real human score to measure their agent against. You're the human reference score, and a lab can't fake one.
Play like normal, but say your reads out loud. Why you rotated, what you heard, the bluff you called. Just talk while you cook.
The reasoning behind the play is the scarce data. Labs pay a premium for the 'why,' not just the input log.
Get handed a game you have genuinely never played and figure it out on camera. No prep, no guides, just you learning live.
Researchers need to watch a skilled human meet something new. Your verified history is the only proof the game is actually new to you.
A lab wants a specific moment: a 1vN clutch, a comeback from a lost position, a weird bug handled cleanly. You go make it happen.
The rare situations are exactly what models fail on and what generic footage never captures. Commissioned collection fills the gaps.
Pick up a new game and log your climb over a few weeks. Same you, from first day to competent. Small stipend the whole way.
How a human actually acquires a skill is a live research question. It's impossible to capture without a verified, persistent identity.
Team up with another player. One of you calls the shots in plain language, the other executes. Then you swap.
This maps natural language straight to in-game action, the core of teaching an agent to follow instructions. It's how DeepMind's SIMA was trained.
Try to break a studio's AI opponent or bot. Find the exploit, expose the dumb behavior, document how you did it.
Studios shipping AI need to know how it fails before players do. Expert adversarial testing is worth real bounties.
Play through AI-generated levels and mechanics and call it: fun or not, fair or broken, ship it or scrap it. Your gut is the product.
This is the job xAI posted at $45 to $100 an hour. Expert taste is what teaches a model what 'good' feels like.
Show the model how it's done. Clean demonstrations from a top player, the reference other AIs try to imitate.
Expert demonstrations carry a big premium for imitation learning, and pro-level gameplay has no public price yet. We haven't found a public rate for pro-tier demonstration, so ours is modeled until a real deal sets one.
Pilot real machines through a headset and controller.
Tesla and Figure pay $25 to $35 an hour for teleoperation. Whether game skill transfers to it is untested, and we are not claiming it does.
What they'd actually ask you to do.
The ten job types above, aimed at games you already play. Each one is a real task a lab or robotics team would commission, and the reason they pay for it.
Clutch a 1vN and talk through every read, or run pure aim-and-reaction sets on command.
Split-second tactical decisions under pressure, plus a human reaction baseline that FPS agents get measured against.
Execute a site take with full utility, then narrate how you read and adapt mid-round.
Multi-agent coordination and decision-making with incomplete information, the hard part of any team agent.
Aerial saves, 50/50s, and recoveries, tagged frame by frame with your inputs.
Physics intuition and 3D control. We would want to test this against robotics work rather than assume it carries.
Get to diamonds from a fresh spawn, thinking out loud the whole way.
Long-horizon planning and tool use. The canonical benchmark agent labs already train on (see OpenAI's VPT).
Drop into an experience you have genuinely never played and figure it out live.
Generalization: how a skilled human learns a brand-new task and interface from scratch. Your history proves it's new to you.
Build-battle a bot, then explain your rotations and resource calls.
Real-time construction and adaptation, planning and building while you fight.
Narrate the teamfight call and the macro rotation, objective by objective.
Strategic long-horizon reasoning coordinated across five players toward one win condition.
Run a raid encounter's mechanics role by role with your fireteam.
Complex, multi-step team execution under time pressure, the kind of coordination we would want to test against multi-agent work.
Real money, on rails built for it.
Your passport is free and matching is free. You onboard once through EG's contractor workspace, the same machinery EG already uses to pay people: tax forms handled, an invoice generated for you when your work is accepted, payment against it. Labs pay on 30 to 60 day terms, so we're building instant pay on top: your money the moment the work clears review, funded by PEAK6 rather than by your patience, for a small optional fee. That's the only fee you'll ever see here.
Instant pay · on the roadmap, launches with marketplace volumeThe work board above is the paycheck. This is the dividend on top: later, you could license your verified history on your terms, cohort-aggregated progression data to AI labs (modeled $4–11/mo), anonymized research data to universities ($1–3/mo), and paid playtest cohorts matched on your real history ($8–25/study). Honest math: that's pocket money for data you already generated, tens of dollars a month at best, never a living, and buyer demand for this shape of data is still unproven. Always opt-in per use, always cohorts, never your identity, only publisher-partnered, and every license stamps a signed consent receipt you can audit and revoke.