Workflows
A workflow is a calibrated simulation pattern: it bundles a methodology, a population spec, and the aggregators that convert agent responses into a shipping number you can act on.
What's inside a workflow
Every workflow has a citation (primary source), a classifier output (what methodology fits best), a calibration history, and a K-sample confirm result. The result: a number in the response shape you expect, with accuracy stated up front.
Versioning
Workflows are semver'd. A patch bump means a re-calibration pass against the original source. Minor bumps add new scenario parameters. Major bumps change response shape.
Picking a workflow: manual or auto
Two ways to point a simulation at the right workflow. POST /v1/simulations with an explicit workflow_id when you already know which calibrated pattern matches your question — browse the catalog at /dashboard/workflows to find one. POST /v1/simulations/auto with a natural-language question when you want Simulix to pick for you — the auto-router scores the catalog against your question and picks the highest-confidence match, falling back to a 422 routing_low_confidence response with the top alternatives so you can pick explicitly. Try both from the dashboard at /dashboard/run.
Custom workflows
Coming: train your own workflow on your own ground truth. If the catalog doesn't carry the methodology you need, the alternative today is to run against the closest match and treat the gap as a known accuracy delta. Full custom-training is on the post-launch roadmap.