GuidesRecipesAPI ReferenceChangelog
Guides

Models Library

The Models Library is where you create and manage all scoring models for your Trustfull products. It gives a full view of which models are active across products and allows building custom models.

Models Library Overview

The model list is split into two sections:

  • Standard models: system defaults provided by Trustfull, one per product. Active by default and cannot be deleted.
  • Custom models: models you have created. Each can be published to become the active model for its product.

The Active indicator shows which model is currently scoring live traffic for each product. Only one model can be active per product at a time.

Creating a Custom Model

There are three ways to create a custom model.

From Training Set

The fastest and recommended path to a model tailored to your data. Trustfull analyses a training set of your labelled historical lookups and auto-generates custom rules based on the fraud patterns in your data, no rule writing needed.

Requirements: the training set must contain at least 10,000 records.

  1. Click Create Model in the top right of the Model Library page
  2. Select From Training Set and choose the Training Set to use
  3. Click Create Model and the model will be generated in minutes

You can also start this flow from the Training Sets page: click the context menu on any training set and select Create Model. You will be redirected to Model Library with the training set pre-selected.

From Library

Copies an existing model (standard or custom) with all its rules pre-loaded, so you can build on top of an existing strategy.

  1. Click Create Model and select From Library or click Copy on any model row
  2. Select the model to copy
  3. Click Create Model and you are redirected to the model editor with all rules pre-loaded

The model is named as "[source model] copy #[ID]". Modify the rules and publish when ready.

From Scratch

Creates an empty model with no rules. Use this when you want to define your scoring strategy from the ground up using the rule editor with autocomplete, a rule bank, and no coding required.

  1. Click Create Model and select From Scratch
  2. Choose the product
  3. Click Create Model and you are redirected to the model editor where you can create all the rules starting from scratch.

The model is named automatically as "Custom [product] #[ID]". Add rules and publish when ready.

Edit and Publish a Model

Edit Model

Every time a model is created or you click Edit on an existing model, the model editor opens. In this page you can add, remove, and toggle rules on or off.

Each rule is built from conditions and an action. For each condition you can select a signal from the fixed set of signals Trustfull retrieves for that product (NB you can find the explaination of each signal in each product's data schema eg Onboarding). You can create simple rules with a single condition, or combine multiple conditions using the AND / OR operator.

Once conditions are set, you define the action: the effect that triggers when the conditions are met. There are two effect types:

  • Add/Remove points: adds or subtracts a weight from the score (e.g. -50 points if phone_has_apple OR phone_has_whatsapp equals True)
  • Limit to cluster: caps the record at a specific cluster regardless of its score (e.g. limit to Moderate if a condition is met)

The rules list shows each rule's description, action weight, product, signals used, and an on/off toggle.

Compare Score with Active Model

As you add or modify rules, Trustfull automatically shows two donut chart comparing What If Analysis the new model against the currently active one. The comparison updates in real time and shows how the model would score the last 1,000 lookups, highlighting shifts across score clusters. This makes it easy to assess the magnitude of rule changes, identify affected risk bands, and validate your strategy before publishing.

Publish

When the model is ready, click Publish from the model editor to make it the active model for its product. This replaces the previously active model and takes effect on live scoring immediately.

⚠️

Review the What If Analysis before publishing to understand the impact on your live traffic.

Deleting Models and Training Sets

To delete a custom model, click the context menu on any row in the Model Library and select Delete. Standard models cannot be deleted.

To delete a training set, click the context menu in the Training Sets page and select Delete. If the training set is linked to an existing insight or model, deletion is blocked and delete the associated insights and models first.


📚 Resources

Updated about 5 hours ago