Version 4.13 includes a change to engine behaviour.
Please read the following post to determine whether this may impact your knowledge maps.
This version is scheduled for release on the following dates:
Community: Monday 16/05/2022 at 5pm
Enterprise: Monday 23/05/2022 at 5pm
As part of this release we’ve fixed a bug in the engine where end-users were being asked questions that were unnecessary, because the engine had already established a fact where the certainty could not be improved.
This defect only relates to object-specific rules, which refers to a rule where the object instance is specified. This article aims to demonstrate where this change applies, so users can determine whether queries to their knowledge maps are affected and whether it may break existing tests.
To confirm, this behaviour does not change the result. It means the result is arrived at in less time with less interaction from the end-user.
Using a simple example, this knowledge map contains 2 rules that determine whether a person has a high approval rating. (The rbird file is attached for reference. Just run a query on the relationship ‘awarded approval rating’).
In this example we can determine if a person has a high approval rating in 2 ways, as described by the 2 rules.
They are either classified as having a high approval rating
IF they have funds of over £1m (Rule 1)
OR they have funds of between £500k - £1m AND assets valued at more than £750k. (Rule 2)
When processing rule 1, the end-user will be asked if the subject has funds of more than £1m. When answering yes with 100% certainty you would expect the engine to return a result of ‘High’. However it was continuing to the next rule and would ask if the subject had funds between £500k - £1m, which is unnecessary.
If the end-user answered yes to this 2nd question and also yes to having assets of more than £750k, the engine would still return a result of ‘High’ with 100% certainty, with the evidence showing this was satisfied by rule 1. This demonstrates rule 2 did not need to be processed as the fact was established and the certainty could not be increased.
The change delivered in v4.13 means the engine will no longer continue processing these rules when it cannot increase the certainty of the fact.
In the attached example, if the 1st question was answered with 99% certainty or less, then rule 2 would be processed because there is a chance to improve the certainty of it.
If any automated tests have been created where they test the expected interaction (i.e. the questions asked for a given scenario), this change may result in the tests failing.
Please contact us if you have any questions or require assistance determining the impact of this change or updating automated tests.
The Product Team