Mahal Systems ScienceWorld
Fits teams who want a procedure-following environment with partial-progress scoring; the 30 task types have about 7,000 parametric variations, so it resists memorization better than ALFWorld.
What it asks the agent to do
- run an elementary-school science experiment in a text simulator
- measure or change a property (boil water, test conductivity, grow a plant)
- classify or find an object that matches a stated property
What to borrow from it
grading designsandbox design
Caveats
Simulated science, not a business workload; size counts task types, not variations; results are hard to compare across papers because step budgets differ.
Details
| Workloads | science, education, embodied, planning |
|---|---|
| Interaction | multi turn scripted |
| Environment | api sandbox |
| End state checked | Yes |
| Tools | text world action interface |
| Grading | programmatic check |
| Size | 30 |
| Data | synthetic |
| Personal data | none |
| License | Apache-2.0 |
| Harness | own repo |
| Status | active |