· 0.0.1-alpha.14
Workflow evidence, benchmarks and clearer run accounting
Build versioned datasets and benchmarks from real workflow evidence, inspect usage and estimated costs, and review inputs and waiting time in Studio.
Before updating
Finish active and paused runs and verify retained-resource cleanup before updating. Stop services, back up state and validate workflows before restarting.
Usage and costs reflect retained provider observations and available pricing; missing coverage is not zero usage, and estimates are not provider invoices.
Dataset exports are analysis artifacts. Migrating execution requires workflow-specific adapters and separate preservation of databases, workspaces and external files.
This alpha targets Linux verification. Native macOS and Windows, Claude and legacy Microsandbox require separate release-specific evidence.
Changes
- Capture workflow evidence, freeze versioned datasets and export analysis manifests with retained provenance and logs.
- Evaluate workflow changes with native benchmark trials, datasets and evaluator workflows.
- Inspect recorded agent token usage and estimated workflow cost, with explicit coverage limits.
- Distinguish elapsed workflow time from time spent waiting for human responses in Studio.
- Inspect bounded input previews and open the full recorded workflow input from Studio.
- Preserve native evidence membership and cursor consistency through concurrent registration and pruning.
- Keep human previews closed after keyboard dismissal and improve native child and component-review compatibility.
- Use expanded consumer guidance, evidence skills and documentation with workflow code and visualization tabs.
Review compatibility in Studio or run laufwerk update --check with an updater-aware CLI. Older installations need an initial manual upgrade. Stop services and verify recovery before changing packages.