Node Transformation

Sometimes the fastest way to work is to outline first and fill in later — drop a batch of node titles, keywords, or rough sentences onto the mindmap, then ask the agent to transform them into proper claims with citations. This is what we mean by node transformation.

It's a tight loop: human supplies structure and intent, agent supplies evidence and prose.

When to use it

  • You know the structure you want and don't want the agent re-deriving it from scratch.
  • You're working from your own notes, lecture, or interview transcript and want to convert them into citation-backed claims.
  • You want to draft fast and let the system verify and cite afterward.

Step 1 — Lay down the scaffold

Create the nodes manually. Don't worry about polish — write each as a phrase, a keyword, or a one-sentence draft.

Examples of useful scaffolds:

Drought hypothesis
├─ pollen evidence supports drought ~1200 BCE
├─ Aegean and Levantine cores agree
├─ critics: dating uncertainty
└─ counter: tree-ring evidence

Each child is rough — exactly what the transformation step is for.

Step 2 — Select the subtree

Select the parent node. The agent's context will then include the parent plus all of its draft children.

Step 3 — Issue the transformation

A clear instruction for this workflow:

Transform each of the children under the selected node into a full citation-backed claim, drawing only from the KB. Preserve the structure I've laid out — don't add or remove nodes. If a child can't be supported by the KB, flag it instead of dropping it.

A few important phrases in there:

  • "Preserve the structure" — without this, the Editor may consolidate or expand the tree.
  • "Don't add or remove nodes" — keeps your scaffold intact.
  • "Flag it instead of dropping it" — surfaces gaps you can address by adding sources, instead of silently losing the node.

Step 4 — Review the result

Walk the subtree node by node:

  • Are the citations actually relevant?
  • Did the agent inflate a rough phrase into more than the evidence supports? Confidence scores help here — low confidence on a transformed node usually means the original phrase was speculative.
  • Are flagged nodes (couldn't support from KB) ones you want to either drop, add sources for, or keep as your own assertion?

Variations on the same pattern

The same transformation step is useful in several other forms:

Polish drafts

You've written a few nodes in your own words. Ask the agent to keep meaning intact, tighten the prose, and add citations from the KB:

Polish each of the children under the selected node. Don't change the substance — tighten the prose and attach any supporting citations from the KB. Flag any claim that can't be supported.

Convert outline to claims

You have an outline of bullet points (not yet on the map). Paste them in as a child list and ask the agent to convert them into a structured subtree:

Here are nine bullet points from my lecture notes paste. Add each as a child node under the selected parent, then transform each into a citation-backed claim from the KB.

Convert transcript to claims

For interview or recorded transcript work:

Under the selected node, I've pasted ten quotes from an interview. For each quote, create a claim that paraphrases the speaker's position, preserve the original quote as a sub-node, and attach any KB citations that contextualize the position.

What makes this workflow efficient

You're paying the agent for the part it's good at (retrieval, citation, prose-tightening) and skipping the part where you'd otherwise debate structure with it. The scaffold is your contribution; the transformation is its.

What's next