Concepts overview
Read these pages in order to understand the runner-first model.
Recommended reading order
- Mental model — workflow and ownership
- Native model and framework runners
- Framework runners
- Data model
- Evidence and trajectories
- Scores and gates
- Annotation — labeling captured sessions (span / trace / session)
- Environment bundles and dependency tapes — executable worlds for agent eval
- How it works
By persona
Framework owner
| Topic | Page |
|---|---|
| First run | Quick start |
| Runner packaging | Prepare a framework run |
| Promptfoo specifics | Promptfoo integration |
| Runner semantics | Framework runners |
Platform operator
| Topic | Page |
|---|---|
| Trust boundaries | Trust boundaries |
| Storage and retention | Storage and thelake |
| Runtime model | Kernel and hosts |
CI / release owner
| Topic | Page |
|---|---|
| Compare and promote | Compare and promote |
| Result status | Result status |
| Events | Events |
Key principle
Softprobe owns the outer workflow and environment, frameworks own inner evaluation semantics.
text
framework suite + subject + environment + runner
→ WorkflowVersion → framework runner
→ native result bundle + evidence
→ Softprobe compare/gate/governance