How we know this
This job isn’t in the U.S. government’s occupational survey yet, so this page carries no official pay figures. Everything below is drawn from published job descriptions and industry guides, each one linked at the foot of the page. The work described is real and sourced. What’s missing is the survey’s rating of which parts of the job matter most — so read this as a picture of the work, not a ranked list.
Think you get this job?Six quick questions on how it really works — with a hint and the reason behind every answer.
Test yourself →The work, task by task
These are the real jobs-to-be-done, not a wish list. Each task shows where we found it, and the prompt underneath is written for that exact task.
Analyze game data to identify balance issues
+Run the latest week of match logs to find champions, items, and abilities whose win rates drift more than…
Analyze game data to identify balance issues
+Run the latest week of match logs to find champions, items, and abilities whose win rates drift more than three percentage points versus expected, flag matches with >500 players, and attach the top five suspect scripts and a short one-paragraph impact note for design.
The tools that do the workSageMathweb
Pasted it? Good. When the reply comes back, push once: ask it to sharpen the weakest part. — Did this prompt help?
Design or build systems and workflows for testing
+Build a test harness that spawns reproducible 100-player sandbox matches with controlled load, random seeds,…
Design or build systems and workflows for testing
+Build a test harness that spawns reproducible 100-player sandbox matches with controlled load, random seeds, telemetry hooks for ability use and damage frames, and a CSV output schema so QA can run nightly regression tests and I can iterate scenarios.
The tools that do the workDockerweb
Pasted it? Good. When the reply comes back, push once: ask it to sharpen the weakest part. — Did this prompt help?
Translate complex problems into structured models
+Convert the vague designer note about ‘burst feel too strong at level 8’ into a model that defines metrics…
Translate complex problems into structured models
+Convert the vague designer note about ‘burst feel too strong at level 8’ into a model that defines metrics for burst (peak DPS, cast frequency, kill share) over level bands and returns parameter ranges to test in simulation runs.
The tools that do the workSageMathweb
Pasted it? Good. When the reply comes back, push once: ask it to sharpen the weakest part. — Did this prompt help?
Check assumptions and document results for review
+Audit the last balance report: list every assumption about player skill, match length, and item acquisition,…
Check assumptions and document results for review
+Audit the last balance report: list every assumption about player skill, match length, and item acquisition, run sensitivity checks that flip each assumption, and produce a two-page review with failed assumptions highlighted for the design lead.
The tools that do the workClaudeweb
Pasted it? Good. When the reply comes back, push once: ask it to sharpen the weakest part. — Did this prompt help?
Refine measurement methods based on evidence
+Adjust our damage-per-second measurement to use time-weighted active windows and exclude periods of crowd…
Refine measurement methods based on evidence
+Adjust our damage-per-second measurement to use time-weighted active windows and exclude periods of crowd control, run the change on the latest dataset and report how many abilities shift more than one standard deviation in ranking.
The tools that do the workSageMathweb
Pasted it? Good. When the reply comes back, push once: ask it to sharpen the weakest part. — Did this prompt help?
Record outcomes in structured logs for comparison
+Log every match outcome into the canonical comparison ledger with fields for build, map, skill bracket, seed,…
Record outcomes in structured logs for comparison
+Log every match outcome into the canonical comparison ledger with fields for build, map, skill bracket, seed, and key metrics, validate the schema against previous runs, and export a differential CSV that highlights runs to revisit.
The tools that do the workDockerweb
Pasted it? Good. When the reply comes back, push once: ask it to sharpen the weakest part. — Did this prompt help?
Says who?
These are the pages we read to build this. Open any of them and check us.
Related careers
Same family of work — each with its own tasks and prompts.
The LLOS Work Atlas is the world's largest evidenced task library — a map of human work, with a ready prompt behind every task. 1,774 careers · every task named by the sources that witnessed it — O*NET, ESCO, real job descriptions, Wikipedia — and the deepest tasks by several at once. And it is honest about limits: where AI cannot help, the map says so.
The rest of the map
Same library, five ways in.
Copyright © LLOS.ai · 2026 — original pedagogy, voice, and design — all rights reserved.
Built on public evidence: O*NET®, ESCO, Wikipedia, U.S. Bureau of Labor Statistics, ILOSTAT. All sources & licenses
Built on public evidence: O*NET®, ESCO, Wikipedia, U.S. Bureau of Labor Statistics, ILOSTAT. All sources & licenses