What a workflow run stores
Each workflow run stores:
- workflow snapshot
- input snapshot
- run mode
- status
- final output
- step runs
Each step run stores:
- step snapshot
- rendered prompt
- output text
- model provider and model name
- token usage when available
- runtime
- error details
- retry count
Why snapshots matter
Snapshots make workflow runs stable over time.
If you edit a workflow tomorrow, yesterday's run still points at:
- the old workflow structure
- the old prompt templates
- the old input shape
This makes comparison, debugging, and review much easier.
How step reruns work
PromptPlan supports rerunning a single step from the run view.
You can rerun with:
- a prompt template override
- input overrides
Current rerun policy is snapshot-first:
- the rerun is based on the original saved workflow snapshot
- prior completed step outputs are reused from the saved run
- the rerun produces a new step run record instead of overwriting the old one
What changes after a rerun
When a rerun succeeds:
- the new step result becomes the latest visible result for that step
- the workflow run final output is recalculated from the latest successful outputs
- older attempts remain preserved in history
When to rerun a step
Rerun a single step when:
- the prompt needs a small correction
- one output is weak but earlier steps are still good
- you want to test a model override on one step
- you need a cleaner export without rerunning the entire chain
When to rerun the whole workflow
Run the full workflow again when:
- your source inputs changed materially
- you changed upstream logic that affects many later steps
- you want a clean end-to-end comparison between two workflow versions