Skip to content
RebelCore Docs

How the Tree works

The Tree is the heart of every project. It’s RebelCore™‘s main abstraction — the place where raw, messy data becomes something the system can reason about.

Where it sits in the flow

The Tree is the third stop in a four-stage journey:

Dataset    →    Project    →    Tree    →    RebelCore™ Agent
(raw data)     (container)     (gold        (inference /
                               abstraction)  action)
  1. You start with raw data — files you’ve uploaded and curated into a dataset.
  2. The dataset lives inside a project, which is the unit of work for one customer engagement.
  3. The project renders as a Tree — that’s this page.
  4. From any node in the Tree you can hand work to the RebelCore™ Agent for inference.

What the Tree actually is

The Tree is a gold abstraction tool. It takes the messy reality of imported data — multiple files, inconsistent column names, partial records, mixed schemas — and lifts it into a clean, labelled hierarchy that downstream tools can act on.

Concretely, each leaf in the Tree (a spec) is a labelled vector set: rows of data attached to the labels you chose when you created the silver dataset. Those labels are what make the data inference-ready — the agent uses them to know which rows belong together, which columns matter, and what kind of question makes sense to ask of each spec.

If a node in the Tree is built and complete, it means:

  • The data has been normalised into a consistent shape.
  • The labels (Label 1, Label 2, …) are attached and consistent.
  • The result is a vector set — a known shape of rows + columns ready for inference.

The agent doesn’t read your raw spreadsheets. It reads the Tree.

Opening the Tree

  1. Open any project from the Projects page.
  2. The Tree opens automatically.
  3. The URL ends with ?projectId=… — bookmark it to come straight back to this view.

Reading the Tree

The Tree is laid out top-to-bottom (or left-to-right, depending on your view setting):

  • Datasets sit at the top — these are the modules you built during curation.
  • Sheets branch from datasets — typically one per worksheet/CSV.
  • Specs are the leaves — individual table or schema specifications, each one a labelled vector set.

Click any node to open the Blade — a side panel with details and AI suggestions.

Merging across multiple modules

When a project carries more than one module, the tree merges them by shared label. Each label that appears on at least one module becomes a first-class branch under Imported Data → Dataset. The modules that carry that label sit under it:

Imported Data
└── Dataset
    ├── Customers          ← labelled in Module A and Module B
    │   ├── Module A
    │   │   └── Sheets / Columns
    │   └── Module B
    │       └── Sheets / Columns
    ├── Sales              ← only labelled in Module A
    │   └── Module A
    └── (Unlabelled)       ← any module with no labels
        └── Module C

Two modules are merged under the same branch only when they share the same label record (i.e. the same Label.id from your curation step), not just the same name. Same-name labels created independently for different customers stay distinct on purpose.

A module that carries multiple labels appears under each of its labels — selecting it under one label is independent of selecting it under another, so you can include the module in one labelled context without including it in the other.

A single-module project still goes through this code path — it just ends up with one module under each label, which matches the historical layout.

The Blade

When you click a node, the Blade slides in from the right. It has two tabs:

  • Details — labels, type, sample rows, status, and the live weight of this node within the current selection. See Details & weights for the full breakdown and why weighting matters before you build a semantic dataset.
  • Chat — AI suggestions for the node, plus a chat box for asking questions about them. See Tree chat & AI suggestions for details.

Tips

  • Use the search box at the top of the Tree to jump to a node by name.
  • Specs marked as incomplete need data or settings before they can be built — open them and follow the prompts.
  • The Open in RebelCore™ Agent button preserves project + spec context so you can drop straight back into RebelCore™ afterwards.

What’s next

If something looks wrong (missing data, unexpected structure), go back to Importing data and re-check the source files — fixing it usually means re-curating the batch.