SCREENCOMPUTER USE
Software Task Screenflows
Desktop task recordings that connect interface state, user action and task outcome across a continuous workflow.
View datasetSCREENERROR RECOVERY
Error Recovery and Correction Screenflows
Successful task traces show what works. They say much less about what a system should do when the interface responds unexpectedly. Error Recovery and Correction Screenflows focus on that missing part of the trajectory: the failed action, the visible error state and the path that follows.
By preserving the context before and after a failure, the dataset can support models that need to recognise an error, choose a correction and continue the task. Error text can also be represented when it contributes directly to the recovery decision.
Technical structure
- Pre-failure context linked to the failed action
- Error state and subsequent correction or retry sequence
- OCR-derived error text where relevant
- Final continuation, completion or abandonment state
Typical applications
Agent robustness, error detection, recovery-policy learning, failure classification and retry behaviour.
Delivery note
Failure types and recovery patterns are selected according to the systems and tasks the project needs to cover.
Delivery
The exact package depends on the collection scope. Where relevant, delivery can include task-specific records, manifests, provenance fields, stable identifiers, SHA-256 hashes, MLCommons Croissant 1.0 metadata and loading instructions.