Decision & Reasoning Engine
Use context, rules and models to determine what should happen.
Lesson 14 of 17 · Find it on the platform map
01 What it is
What this layer does
The decision and reasoning layer turns context into a choice. It gathers what is known about a situation, applies calculation, models and business rules, and produces a set of options with a recommended course of action. In SCIKIQ this layer retrieves context, reasons with LLMs and models, calculates and simulates, evaluates rules, policies and constraints, trains and scores predictive models, and returns options and a recommendation.
02 Concepts
Four ideas to hold on to
Decision intelligence
The practice of treating a business decision as something designed, with defined inputs, logic, options and outcomes, rather than an informal judgement. It sits between analytics, which describes, and automation, which executes.
Hybrid reasoning
Combining deterministic logic such as calculations and rules with probabilistic reasoning from models and LLMs. Each covers the other’s weakness: rules are exact but rigid, models are flexible but can be wrong.
Simulation and what-if analysis
Estimating what would happen under each option before acting, for example the margin effect of a discount or the stock effect of a reorder. It turns a single answer into a comparison of consequences.
Constraint evaluation
Checking each candidate option against hard limits — budgets, policies, contractual terms, capacity — so that options which break a rule are removed or flagged before they are recommended.
03 In SCIKIQ
What the layer contains
04 What good looks like
Signs it is working
- Every recommendation shows the options considered, not just the chosen one.
- Each option lists which rules, policies and constraints it was checked against and the result.
- Calculations and simulations can be rerun on the same inputs and give the same figures.
- Predictive model scores used in a decision record which model version produced them.
05 Diagnostics
Questions to ask your team
- 1
For our most frequent decisions, which parts are rules, which are calculations and which are model judgement?
- 2
Can we show a reviewer why a recommendation was made, including the context it was given?
- 3
Where do we simulate consequences before acting, and where are we still guessing?
- 4
Who owns the rules and constraints the engine evaluates, and how are changes to them approved?
06 Keep going
Related reading
— Questions
Frequently asked
Is this layer just an LLM?
No. An LLM is one reasoning tool among several; the layer also calculates, simulates, scores predictive models and evaluates rules, so that numbers and policy limits are not left to a language model.
Does the engine take the action itself?
No. It produces options and a recommendation; the Governance Gate decides whether the action may proceed and the Action Fabric carries it out.
Where do the predictive models come from?
Predictive models are trained and scored within this layer, and their scores feed the options alongside rules and calculations.