Administrative workflows with consequences you can inspect.
Maydan’s work models government and enterprise HR processes as applications an agent can operate, with synthetic records, resettable starting states and deterministic checks after action.
What is inside an environment?
A working browser application, a synthetic population, parameterised tasks, state reset and a verifier that reads the resulting database. The goal is to test a sequence of decisions and actions across records, not a text answer.
Saudi employment and residency administration is the first environment area. Example state includes employers, workers, dependents, contracts, documents, eligibility conditions, fees and prerequisites. Wider GCC work is a direction for scoped projects; this page does not imply completed country packs for every GCC state.
Why workflows need state
Many failures appear only after several steps: a date changes eligibility, a blocker must be resolved first, or the right action for one record would damage another. Instances can also require no action. A saved state check can distinguish those outcomes.
What stays out of the environment?
The records are synthetic. No real person, employer or government record is used, and the environment does not connect to a live government system. A custom scope requires its own rule provenance and agreed boundaries.
Broader environment work includes Arabic interfaces and matched Arabic and English task design. The currently public sampler and current calibration evidence are English only; they do not establish an Arabic performance difference.