Mahal Systems AgentBoard
Fits teams who want partial credit instead of pass/fail for multi-step tasks. Borrow the progress-rate metric (sub-goals reached) and the grounding-accuracy breakdown.
What it asks the agent to do
- complete household or science tasks in a text world
- solve a text adventure or PDDL planning problem
- shop or navigate on simulated websites
- answer questions by calling tools or operate a tool to change state
What to borrow from it
grading designtaxonomy
Caveats
Nine environments all wrap earlier benchmarks, so much of it is familiar to current models; the dataset is GPL-2.0, which some teams cannot redistribute.
Details
| Workloads | general assistant, embodied, games, web browsing, tool calling |
|---|---|
| Interaction | multi turn scripted |
| Environment | mixed |
| End state checked | Yes |
| Tools | text world simulators, web simulators, query and operation tools |
| Grading | programmatic check |
| Size | 100 |
| Data | mixed |
| Personal data | none |
| License | Apache-2.0 (code); GPL-2.0 (data) |
| Harness | own repo |
| Status | active |