An incident is triggered by a rule in a Decision Strategy (e.g payments.strategy). Decision strategies are designed to be the last piece of logic executed prior to an event being created.
When defining a rule in a decision strategy, you are given the opportunity to define one-or-more remediations - these include
- adding or removing of labels to key identifiers on the event
- setting of attributes
- adding the event to an incident
Triggering an incident
An event must have its result terminated with a rule in order to trigger an incident. When an event is terminated you must specify an initial disposition (Pass, Reject, Alert, Review, Challenge)- this is an indicator of what the engine or your rules believe an outcome to be. It is important that you set a disposition as this enables Darwinium to later determine KPIs such as false positive rates.

Rules below a result terminate are still evaluated, but no longer will determine the decision strategy result or trigger additional incidents. Their other remediations, such as labels and attributes, don't run either. So the higher priority rules and queues should be configured higher in the strategy.
Note: You will still see the signal of those lower rules if they fire.
When triggering an incident, you are given a number of options in the rule:
- The queue the incident will appear. Details on queue configuration can be found in Configuring Queues and Teams
- The mechanisms under which subsequent events matching the rule criteria will be grouped into incidents. An example of this might be grouping subsequent events with the same primary IP address into the same single incident.
This is highly configurable:
- Infer from rule - observes the rule logic and picks the best means of grouping. This is particularly useful when decisioning from a feature value. Note: If the rule condition contains no general features with
event_typeandwith_the_same, the "Infer from rule" produces exactly the same config as "Do not group": one incident per firing event. - Do not group - means that each subsequent incident will contain only 1 event
- Manually specify grouping - enables the rule author to specify the common event identifiers and attributes (such as a device signature, IP address, username, or a combination of these) under which subsequent events will be grouped into incidents for review
Manually specifying grouping
When you choose Manually specify grouping, three fields decide which other events join the incident:
- For events - the event types that can join. All events accepts any type, Same as current only the type of the event that fired the rule, and Specific event types only the types you pick.
- Condition - an optional filter, written as a search query in the same format you use to search events in the Portal. Only events that match it join the incident. For example,
payment.amount > 1000groups only payments over 1,000. Leave it empty to group every event that matches the other two fields. - With the same - the identifiers or attributes an event must share with the one that fired the rule, such as a customer token or device signature. When you pick more than one, an event must share all of them.
The rule's own condition decides whether an incident is raised. Condition here only narrows which other events are grouped into it.
In the example below, the rule rejects the event and raises an incident in the 3. High Risk - Reject queue. Events of any type then join that incident when they share both the account's customer token and the device signature. This lets an analyst review everything one customer did on one device as a single case, rather than one event at a time.
