Glossary
Chain of reasoning.
The sequence of inferences an AI follows to arrive at an answer: "X is true because A, A is true because B and C, B comes from source D." Lets the user follow the logic step by step, not just see the final claim.
A model that shows its reasoning is a model you can argue with. A model that only shows its conclusion is a model you have to either trust or distrust, with nothing in between.
What this looks like in practice
Why a conclusion alone isn't enough.
"Yes, this customer should be offered the discount" is a claim. The user has no way to evaluate it without seeing the logic that got there. Was the customer's revenue checked? Were past discounts factored in? Did the model weight the renewal date correctly? Without the chain, every answer requires either blind trust or independent re-verification.
Where chain of reasoning differs from chain of thought.
Chain of thought is the model's internal step-by-step process while generating. Chain of reasoning is the explanation it shows the user. They overlap, but the framing is different. Internal = debugging. User-facing = trust and education. Both matter. Most products do neither well.
The trap: reasoning that sounds rigorous but isn't.
A model can fabricate reasoning the same way it can fabricate facts. "I considered X, then Y, then arrived at Z" can be invented after the fact to make a guessed answer feel justified. Real chain-of-reasoning has to be tied to actual retrieved sources at each step, not generated narration. The difference is structural, not stylistic.
Show the logic. Earn the trust.
Every Lumen answer can be expanded to show the reasoning path: what was retrieved, how it was weighed, what the final answer is based on. You don't have to trust the conclusion. You can follow the work.
Bring your knowledge.Simply ask.
CASA Certified · SOC 2 Infrastructure · GDPR-Aligned · Enterprise Ready