One person, eight systems, three shared keys.
Nothing in the schema says the HR record and the directory account are the same person. 52 of the 127 tasks cross at least one boundary and have to infer it — 11 span three platforms or more, 75 stay inside a single system.
Identity Security Posture Management
127 identity-security questions over Osbench Labs, a frozen synthetic enterprise spanning eight platforms. Every task is classified by theme, scope, and structural complexity, over the platforms it touches, and every answer comes from the environment’s own data, so there is a right answer to check against.
- Published modality
- Text-to-SQL
- Native-vendor modality
- The same world, through APIs and CLIs.
The first public contract gives the agent a frozen relational snapshot and records each read-only query as evidence.
The framework keeps the environment, tasks, and answer criteria fixed while the interrogation surface changes. Request traces then play the evidentiary role of SQL traces.
Composition
every bar is a share of 127 tasks
Theme
Scope
Complexity
a scale, not a ranking
Platforms
a task can span several
Run the benchmark yourself.
One binary ships the task set, the pinned dataset, and the grader. The score you see locally follows the same public contract as the board.