Once you deploy a decision tree, you can do more than run it. You can ask it why. SMEme answers from logic, not a guess, so it can tell you which facts decided a job and what would change the outcome. °FAI: Ask the Tree, Own the Tree.

Part 1 covered why rules don’t belong in prompts. Part 2 showed how to build a tree.

Questions a chatbot can’t prove.

SMEme’s homepage lists six questions you can ask a running tree. Here’s how they’d work on an underbase call.

What follows? The tree calls for a full underbase, rules it out, or tells you it still needs a fact.

Which facts decided this? SMEme gives you the smallest set of answers that forces the outcome. Maybe garment color and ink color decided it, and fabric didn’t matter.

What if a fact changed? Swap one answer and see if the call holds. Would it still need a full underbase on heather gray?

What would have to change? SMEme shows you the edits that would reach a different outcome.

What can’t change the result? If an open question can’t move the outcome, SMEme tells you, and you stop chasing it.

What can this tree reach? You see every outcome the tree allows.

A prompt gives you an answer. It can’t prove the logic behind it.

So you’re teaching an AI through literal examples of your complex variables, AI decision support.

You own the tree.

The knowledge in a decision tree belongs to the shop that built it. You can download every tree you build. Arista Labs publishes the core engine as source available under a fair-code license, so you can host it yourself.

Why this matches my pipeline.

A print order advisor recommends mesh and flashes, and a person makes the call. Everything stays read-only until it clears accuracy gates the factory sets.

Each factory also decides how much of its data touches a third party. That’s an ownership question too. I wrote about why the owner has to make those calls in °FAI: AI Adoption Is a Leadership Decision.

What this really shows: your tribal knowledge is your competitive advantage. The right tool helps you encode it.

Tools used:

• Claude (read the SMEme docs, drafted this article)
• SMEme (AI decision support)