Skip to main content
The Contexts page is the operator workspace for each recommendation surface. It lets you create a context, attach a ranking recipe, review the runtime pipeline, and manage pre-query filters from a single slide-over editor.

​
Open the context editor

1

Go to Contexts

Open Console > Contexts.
2

Create or edit a context

Click New Context to create one, or click an existing row to edit it.
3

Work through the tabs

Use Basics, Recipe, Pipeline, and Advanced to configure the surface.
4

Save changes

Use the sheet footer to save the context. Recipe and pipeline edits are managed inside their embedded editors.

​
Basics tab

Use Basics for the surface definition. Fields include:
  • Context Name - the operator-facing name, such as Home Page Feed or Related Articles.
  • Description - optional notes for operators.
  • Recommendation Type - the intended recommendation mode for the surface.
  • Model Family - the model family the resolver should prefer.
  • Retrieval Algorithm - the first-stage candidate generator for the context.
For existing contexts, Basics also shows Currently serving, which resolves the model actually used by requests after the fine-tune, family base, and general fallback chain.

​
Recipe tab

Use Recipe to attach or edit the ranking strategy for the context. A context can start with:
  • a context-only setup with no recipe
  • a blank ranking recipe
  • a use-case preset
  • a reference template
Recipes can configure candidate sources, signal overrides, rerank behavior, discovery, guardrails, and suggested events. Switching tabs keeps staged recipe edits mounted while the editor remains open. If the context has no recipe, attach one from the Recipe tab. If a recipe exists, the embedded recipe editor opens inside the sheet.

​
Pipeline tab

The Pipeline tab appears when editing an existing context. It embeds the pipeline editor in context-locked mode so the operator can review and tune the stages used by that surface. Use it to check:
  • candidate generation limits
  • feature enrichment
  • scoring
  • rules stage behavior
  • post-processing and page size
If you are creating a new context, save it first before configuring its pipeline.

​
Advanced tab

Advanced is for pre-query filters. These filters narrow the candidate pool before scoring. Use pre-query filters for hard constraints such as:
  • only recommend items from a specific inventory pool
  • exclude unavailable content before ranking
  • restrict a surface to a metadata category
Use the Rules Engine instead for post-scoring controls such as boost, bury, pin, dedupe, cap, grouping, or reorder.

​
Quality checks before launch

Before sending traffic to a context:
  • confirm the model family and retrieval algorithm are intentional
  • verify the currently serving model
  • attach or review the ranking recipe
  • confirm suggested events exist if the recipe depends on them
  • inspect the context pipeline
  • test at least one representative user through the recommendation API
  • use Explainability to validate a few ranked results