--- title: Guided Tree Completion description: "The last human pass over an AI-built competency tree — one decision at a time, with the instructional-design expert beside you, until every end is mapped and the tree is ready for Learning Object creation." created: 2026-10-25 updated: 2026-10-25 authors: ThinkingCap topics: [Research Development] status: published canonical: https://console.thinkingcap.com/guest/Research-Development/Skills/Guided-Tree-Completion order: 1 --- # Guided Tree Completion You brought us a competency framework — authored it, imported a CASE package, or adopted one. Our downward inference worker took it from there: it walked every node, folded in what ESCO, O*NET, OaSIS and the legacy taxonomy know about that territory, and decomposed your claims downward until each branch ended in something real — an atomic, teachable skill with a Rich Skill Descriptor behind it — or stopped honestly where the evidence ran out. That gives you a complete *draft* of a competency and skills map. Guided Tree Completion is how you make it *yours*. ## How it works Open any tree in **Our Frameworks**. If it still owes you decisions, you'll see a **Complete tree** button in the header, with the count right on it. The wizard tells you exactly what's outstanding — how many nodes are pending review, how many open dead ends remain — and then walks you through them, one node at a time, in tree order. At each step you see the node itself: what it claims, where it came from, how confident the worker was, and the evidence it stood on. Then you decide: - **Agree** — this belongs in the tree, as written. - **Reject** — it doesn't. The whole inferred branch below it goes with it. Nothing is ever deleted; rejections are superseded and auditable, forever. - **Add a child** — your knowledge, supplied by hand. If the node is a dead end, this is how it stops being one: the moment you name a real sub-skill below it, downward inference can continue from there. - **Infer Down** — where the tree allows it, send the worker deeper. You are never alone in the wizard. On the right of every step sits your instructional-design expert. It opens each decision with a short, plain explanation of the question in front of you — what this node claims, why it ended up here, what your options actually mean — and then it's yours to interrogate. Ask it anything. Push back on it. It advises; the buttons decide. Work a little and leave, or go to the end. Every decision saves the moment you make it, so there is no wrong place to stop. The end is worth reaching: when the queue is empty, the tree is **fully realized** — every node reviewed, every end mapped. That is the moment a competency and skills map becomes ready for Learning Object creation. Everything we build downstream — experience envelopes, curricula, instructional designs, and the learning objects themselves — attaches to this tree. Completing it is the last step that makes all of that sound. ## Why we ask you to finish it We could hide the seams. Most of this industry does. The AI's inferences go live provisionally because momentum matters — a tree that waits for permission never grows. But provisional is not the same as ratified, and we will not pretend otherwise. The wizard exists because *a map this important deserves a human pass before anyone builds on it* — and because that pass should take an afternoon, not a committee. And we will be straight with you about something bigger. Every part of this machine stands on strong science — decades of occupational research, published labor-market frameworks, explicit confidence scores, and a provenance trail that shows its reasoning at every step. You can open any node and ask *why*, and get an answer with receipts. But until real-world outcome tracking closes the loop — until we can watch learners actually succeed or struggle against these maps and measure the fitness of what we inferred — all of it, ours and everyone else's, remains well-reasoned conjecture. That is not a ThinkingCap confession. It is a reality the whole of edtech has to face. So don't be discouraged when an inference is wrong. Some will be. Wrong is cheap here — reject it, add what you know, let the worker run again — and wrong is *informative*, because the moment outcome data flows back into this picture, every one of those corrections becomes evidence. The only unforgivable state is a tree nobody looked at. Let outcomes guide your judgments, and let your judgments teach the machine. Complete the tree. Then we build.