Skip to main content
This walkthrough connects the ranking features into a single sequence. You create a context for one surface, attach a recipe, inspect the pipeline that serves it, apply a business rule, and then explain an individual result. Everything below happens in the console; no application changes are involved. Work through it on a surface you recognise — a homepage feed, a related articles rail, a product detail rail, a continue-watching row — so the results are meaningful to you as you go.

​
The sequence

1

Start with a surface

Open or create a context for the surface, such as Homepage Feed, Related Articles, Product Detail Rail, or Video Continue Watching.
2

Decide what the surface is for

Write down what it should improve: discovery, CTR, conversion, retention, long-tail coverage, freshness, or editorial control. The recipe you choose next follows from this.
3

Attach a recipe

Open the context’s Recipe tab. Candidate sources, signals, discovery settings and guardrails all attach to the surface here.
4

Inspect the pipeline

Open the Pipeline tab to see the runtime stages a request passes through: candidate generation, enrichment, scoring, rules, ranking and post-processing.
5

Apply a business control

Add one rule — boost fresh content, filter unavailable items, cap sponsored items, or pin a launch item. Rules run after model scoring, so relevance still determines the ordering within what the rule allows.
6

Explain a result

Use Explainability on a single user-item pair to see the retrieval score, which rules applied, the status of each pipeline stage, and the feature contributions.

​
What the walkthrough demonstrates

Completing the sequence exercises each of these behaviours:
  • Ranking behaviour is configured per surface, so two contexts can optimise for different outcomes.
  • Operators change ranking in the console, without an application release.
  • Business rules run after model scoring, so control and relevance compose rather than replace one another.
  • The pipeline exposes the runtime path a request took, stage by stage.
  • Explainability makes an individual recommendation inspectable, which is what you need when debugging an unexpected result.

​
A worked example

For a news or media catalogue:
  1. Create or open a Homepage Feed context.
  2. Attach a discovery-oriented recipe.
  3. Look at the candidate generation and rules stages in the pipeline.
  4. Add a rule that boosts recent editorial priority content, leaving relevance to order the rest.
  5. Request recommendations for a sample user.
  6. Explain one recommended item, and read the score, the matched rules and the pipeline stages.