Build 360 Table

Causal discovery requires a single, coherent table that represents the domain you want to model. If your data lives across multiple datasets — customers in one file, transactions in another — you need to join them before the engine can work.

RootCause calls this joined table a Data View. The friendly name for it in this workflow is the 360 Table: one row per entity of interest, with all the variables you want to analyze in a single flat structure.

The good news: RootCause builds it for you. After the ontology runs, it has already identified which datasets share common identifiers and can recommend a join recipe ready to use in one click.

---

The recommended path

After the ontology finishes processing, open the Datasets section in the top navigation bar.

Two single-source Data Views were created automatically during ontology processing; the joined dataset was created from the ontology's recommendation.

You will see:

• Auto-generated views — one per dataset, created automatically, with zero operations (a direct pass-through to the raw data).

• Recommended Datasets — on the Ontology page's right panel, RootCause identifies datasets that share an identifier and proposes a join. Click the recommendation to create the joined view instantly.

The recommendation is derived from the ontology: when two datasets share the same identifier concept (for example, Customer Id), RootCause proposes joining them on that field, with the larger dataset as the main table.

---

The Operations editor

Clicking a recommendation creates the view and opens it in the Operations editor.

The joined view with its Operations panel: two source datasets joined on Customer Id, 4,998 rows out. The join type, key columns, and table order are all editable.

The editor shows your recipe as a visual graph:

• Source nodes (blue cards) — one per input dataset.

• Operation nodes — transformations applied in sequence. The recommended view starts with a single JOIN node: "Left join on Customer Subscription info - customerID = Customer Details - customerID."

• Bezier edges connecting each source into the join.

The recipe is fully editable. Each node has pencil (edit) and trash (delete) controls. You can change the join type from left to inner, switch which table is primary, or add further operations — a filter to exclude inactive records, an aggregate to roll up transaction-level data to customer level, or a derived column calculation.

The accept-as-is path takes one click. The edit path uses the same interface. You do not need to rebuild from scratch to change what the platform inferred.

---

Adding operations manually

To build a view from scratch, or to add steps to an existing view, use Add Source and Add Operation in the top-right of the Operations panel. Available operations include:

Category	Operations
Data combination	Join, Time Series Join, Concatenate
Filtering and cleaning	Filter, Remove Duplicates, Drop Nulls, Impute Nulls, Clip Outliers
Reshaping	Aggregate, Sort, Transpose Panel
Column manipulation	Split Column, Drop Columns, Replace Values, Cast Type, Normalize
Categories	Map Categories
Time series	Interpolate Time, Time Series Aggregate, Frequency Analysis, Cyclic Aggregate, Vertical Diff, Date Span, Cumulative Sum
Arrays	Decompose Array, Consolidate into Array, Explode Array (long), Aggregate Array
Custom logic	SQL Query

The specialized time-series and array operations, briefly:

• Time Series Join — join two sources on time, with exact, nearest (as-of), or granularity-based matching.

• Transpose Panel — collapse a long-format panel dataset into a single wide-format time series, one column per environment and value.

• Frequency Analysis — run FFT or Welch's method over a time series to surface its frequency content.

• Cyclic Aggregate — group rows by a cycle component (hour of day, day of week, …) and aggregate, revealing the shape of a cycle.

• Vertical Diff — the change from each row to the next along the time axis (first difference).

• Date Span — the interval between two date columns in the same row, in the unit you choose.

• Cumulative Sum — a running total along the time axis.

• Decompose Array — spread an array column into one scalar column per position, or one boolean column per distinct value.

• Consolidate into Array — merge several scalar columns into one array column.

• Explode Array (long) — unnest an array column into one row per element.

• Aggregate Array — reduce each row's array to a single value (mean, sum, min, max, median, or count).

Operations execute in sequence — each takes the output of the previous step as input. Preview results after each addition to verify the output before continuing.

---

What's next

When the view looks right, you're ready for causal discovery. It runs when you build a Digital Twin from this view — either click Create & build twin on the Ontology page's recommended dataset, or start from Digital Twins → New Digital Twin and pick the view in the builder.

Next step: Step 4: Build Causal Graph