The Organization home. Each card is a workspace — a self-contained analytical project.
Creating a new workspace. A descriptive name helps RootCause suggest relevant analysis as you work.
A new workspace opens on Sources. The top navigation bar carries the six main sections, which map directly to the seven-step workflow.
The Import Manager lists all available connectors. Sources marked Setup need credentials configured before use.
Drop files onto the zone or click to browse. You can add multiple files at once.
Files staged and ready. The Upload button shows the number of files queued.
Current Imports. Each file reports its size, rows, and columns once the upload completes; View opens the imported source.
Sources after processing: both datasets imported and analyzed. The journey card in the corner offers the next steps — understand why, build a report, or build an app.
The ontology network. Each colored cluster is a dataset (Customer Subscription info with 15 concepts, Customer Details with 7); the node where they meet is the shared Customer Id identifier.
A recommended dataset, tagged Static twin: the platform has detected the join between the two sources. View all recommendations opens the full card, the one-click path to Steps 3 and 4.
The Table view: concepts as a sortable table, the fastest way to check that identifiers and time fields have been correctly detected. Here Customer Id is the only classified concept, an identifier shared by both datasets.
Two single-source Data Views were created automatically during ontology processing; the joined dataset was created from the ontology's recommendation.
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 Relationships view: the causal graph beside the relationships table — 31 relationships here, with a further 140 candidate connections tested and ruled out. Clicking any node reveals what drives it and how much each driver contributes.
The Digital Twins list. Click New Digital Twin to begin.
The twin builder. Choose a method, pick the data, add what you know, then review and run.
Training progress. The stages run in sequence; each can take from seconds to several minutes depending on dataset size.
The twin Home after training: Configuration (19 variables, 31 relationships), Versions (v1.0.0), Simulations, Graph explorer, and Model evaluation (69.6% model fit).
The New Simulation panel. Type a plain-language question to auto-generate a configuration, generate Quick Start suggestions, or choose a simulation type directly.
The seven simulation types available on a static twin.
A completed Explanation. The Key Driver Analysis ranks each variable's causal contribution to churn: Contract leads at 44.4%, then InternetService (25.4%) and OnlineSecurity (17.0%), each with a range and a confidence rating.
The Reports home lists every report in the workspace with its owner and status. New report starts a draft.
Generate with AI: describe the report and type @ to ground it in your data — Enter generates.
A generated report in the editor. Sections are structured, the Contents rail tracks the outline, and the highlighted numbers are evidence links back to the underlying runs. The report can be edited freely after generation.
The import manager. Choose file upload for one-off imports or a database connector to keep data in sync automatically.
The dataset detail view. Every column carries its type and value distribution; the panel on the right reports the sources, concepts, and operations behind the view.
The upload area. Drag and drop files directly, or click to browse.
The dataset view shows the detected schema, a preview with per-column statistics, and the summary panel: here the joined customer view, 4,998 rows across 21 columns.
The import manager. Select a connector type to begin the connection setup.
The builder's opening steps: pick the Data View and the platform marks which twin types the data supports, here Static as Recommended with the time-based types disabled.
The four twin types, by time dimension and number of environments.
The Type selector after picking a data view: Static is marked Recommended, and the types the data cannot support are disabled.
The Advanced settings panel and the Review & run step: training options and engine choices above the plain-language recap. Temporal and multi-environment twins add controls to the same screen.
A first run with a poor fit — 17.1%. The averaged score is the headline, but the variable you care about may score differently.
The retrained twin after dropping City to leave 19 environments — fit rose to 29.1%.
A trained Digital Twin's Home. The cards on the right report 19 variables and 31 relationships, version v1.0.0, and a 69.6% model fit — and open each section.
Two versions compared in the Separate view. The Comparison Overview reports size, structural overlap, and evaluation for both models — here v1.1.0 carries 32 relationships to v1.0.0's 31, sharing 28 edges.
The merged diff view. The 28 shared edges render grey; the coloured edges are unique to one model or flipped direction between them — exactly the differences re-discovery introduced.
The twin Home. The causal graph fills the canvas; the card panel on the right summarizes the state of the model.
Variable Details for Churn. Contract, InternetService, OnlineSecurity, and TechSupport are its direct drivers; prediction quality and the learned distribution sit below the summary.
The full Evaluation view. The Model health panel carries the headline verdict; the charts and table break it down per variable.
The Relationships panel beside the graph. The header shows the total relationship count (31 here); the footer reports the 140 connections tested and ruled out.
The Config tab. All sections are read-only; Modify model opens the builder to change them, starting a new version.
Version History. Here v1.1.0 is Trained and Current at a 71.7% model fit, with v1.0.0 preserved below it; select a row to inspect a version or switch back to it.
Modifying a twin. The builder opens seeded with the current version's configuration; the method chosen in step 1 decides how the next version is built.
A version in training. Each of the five stages reports progress; the run can be cancelled.
A finished version — here v1.1.0, trained and current at a 71.7% model fit. The model fit score reports how well this version fits the data — see Reviewing Model Quality.
The management list. Open a twin's Versions card for the per-version breakdown.
A completed Prediction run: the AI summary states the headline probabilities, and the predictions table below covers every input scenario, here 3 of them, with 95% confidence intervals.
The Simulations panel on the right lists recent runs against the twin; New Simulation opens the type picker.
One target is enough; multiple targets run in the same simulation.
The setup form, empty — no target chosen yet (0 of 19), no input records below.
Three scenarios filled in, one column per record. The banner above the grid shows the causal filter at work: only the 7 variables with a causal path to the target remain as inputs; the other 13 were filtered out.
The Simulations panel lists every run against this twin, here nine completed, with search. Click any entry to reopen its result.
The empty setup form. Sections build the experiment top to bottom.
Choosing the variable, the change type, and the new value. The Estimated Intervention Effect preview shows how the chosen change shifts the variable's distribution before you run anything.
A metric on the SQL tab. The query must return a single value column; click Test SQL to run it against the underlying data; here it passes with a baseline churn rate of 0.241. A second metric, overall_churn_rate, is stacked below.
A completed configuration. The summary mirrors the experiment back in plain English: the change (Contract to Two year, applied only when Contract equals Month-to-month) and the two churn-rate metrics it will be judged on, both aiming lower. Validate checks the setup; Run Simulation launches it.
A run in progress; the sidebar restates what is being changed and measured. Static interventions usually complete in seconds.
The top of a completed run. The AI summary states the headline effect: churn falls from 0.24 to 0.05, a −78.05% change flagged statistically significant, above the Intervention Impact chart and per-metric table.
Segment Comparison in the result. The fiber-optic internet segment drops from 0.40 to 0.11 while segments that barely churned barely move: the effect is real but concentrated where the change bites. Every section exports as PDF.
The New Simulation screen. Each type opens its own setup form.
Natural Language. Describe the metric in plain English, say the share of customers who churn, and Generate SQL writes the query.
SQL. Test SQL confirms the baseline before the run: here, the current churn rate is 0.241.
The completed configuration. The summary mirrors the run back in plain English: Minimize Customer churn rate by adjusting Contract, OnlineSecurity, and TechSupport, with both constraint sections left empty. Validate checks the setup; Run Simulation launches the search.
The top of a completed run. The AI summary states the recommendation: a three-lever change cutting churn 87.1% from the 0.24 baseline, with all constraints satisfied, above the three headline stats. It also notes the cost side plainly: the solution requires 3 interventions.
The rest of the result: the three recommended changes, the baseline-vs-optimized chart (0.24 → 0.03, −87.14%), and the All Solutions table with the baseline as row 0. Note the OnlineSecurity recommendation: the optimizer searches every level present in the data, including a missing/blank level, so inspect recommended values before acting on them. Every section exports as PDF.
The empty setup form. The stepper at the top walks the sections in order; nothing runs until a target and at least one record are set.
A target outcome, Churn set to No. For categorical targets the match is exact; numeric targets add tolerance and direction modes. The Baseline states section, covered next, starts below.
Field Input. Each record is a column; set a value for every variable to define a starting state.
File Upload. The CSV must contain every variable the twin uses as a column.
The populated form: target Churn = No and the baseline records in place. Run Simulation launches the search across every baseline record.
The result list, one block per record. Sample #1 reaches Churn = No by moving Contract to One year and OnlineSecurity to Yes at 81.22% confidence; Sample #2 moves Contract to Two year and OnlineSecurity to Yes at 94.95%. The sidebar restates the run: Churn = No across 2 starting records, at most three variables changed per scenario. Every section exports as PDF.
The Explanation setup opens on the mode step: Directional, Discovery, or Impact. The mode chosen here decides which variables the next step asks for.
Input. Both a Cause variable (Contract) and an Effect variable (Churn) are set; Step 3 Segments is left empty to run across the whole population.
Output. The headline card quantifies the single relationship, Contract shifting Churn: Yes from 22.7% to 35.5%, with the AI summary spelling out the effect size and its significance above it.
Deeper in the same result: Causal Pathways attributes 97.8% of the effect to the direct path, and Segment Analysis breaks the effect out by contract type.
Input. Discovery needs only the Effect variable (Churn); the Segments section is identical across all three modes.
Output. The ranked drivers: each row carries its contribution, range, and confidence, and expands into the full pathway breakdown.
Input. Impact needs only the Cause variable (Contract).
Output. Contract's downstream footprint is dominated by TotalCharges at 98.3%; every other affected variable, including Churn, carries a sliver of the total impact.
Root Cause Analysis runs against a trained Digital Twin — the same causal model that powers every other simulation type.
The Root Cause Analysis setup as it opens: the anomalous variable to diagnose on top, the observed data to diagnose it against below.
The setup filled in: MonthlyCharges as the anomalous variable, with the observed customer records to diagnose — one column per record.
The run page as the analysis lands: the AI summary and the per-sample table arrive together. The sidebar restates the run — diagnosing MonthlyCharges across 4,998 samples, here at a 1.00% target FPR.
A completed analysis. The AI Summary is the headline verdict; the table below shows the working: observed vs expected, severity, and the anomalous flag for every sample.
Anomaly Scan runs against a trained Digital Twin — the same causal model that powers every other simulation type.
The Anomaly Scan setup as it opens. There is no variable to name — the scan checks every variable in every record you provide.
Observed data on the Dataset tab: pick a dataset and its records load as the samples to scan; here each of the three carries 4,998 rows. Only columns matching the twin's variables are imported. The detection settings follow below.
The scan in progress — this run scans every variable across 4,998 samples at a 1.00% target FPR with Auto-RCA enabled. Expect minutes, not seconds.
A completed scan. The headline states the count, the charts show where the anomalies sit, and the samples table below lists each flagged record with its diagnosed cause.
The temporal twin's simulation menu, with generated Quick Start suggestions above the type cards. Choose Forecast to project variables forward in time.
The start of the Forecast setup: pick the targets to project — each selected target becomes its own chart in the result.
The configuration summary confirms what is being forecast, over what horizon, at what confidence — the last check before running.
A completed Forecast. Each target gets its own chart with the projected trajectory and its confidence band; the summary table reports the net change over the horizon.
The Assistant is an orchestration layer. Your question triggers precise tool calls to causal engines — the numbers come from the math, not from AI pattern-matching.
A data query in the Assistant: the average churn rate is 26.3% — 1,313 of 4,998 customers churned — with the query results table attached as clickable evidence.
The Applications list. New application opens the builder.
The builder's first step: pick the surface your audience will consume. Each card lists example uses of that type.
The start path and the basics: a blank canvas, an AI-scaffolded flow, or a template, then a name and one-line description.
Organisation Details: the name, the wide logo, and the square icon, each replaceable in place.
The three default roles and their per-area levels. Create Role defines a custom one.
Sharing a workspace: add users or whole roles at Read, Write, or Full access. The owner is always listed.
Team Members with the Invite User action, and Domain-based joining below it.
The API page lives on the Organization home, under Manage.
The API page opens on Getting Started: the authentication example in cURL, Python, or JavaScript, with the API reference links below and the other sections stacked underneath.
Creating a key: a recognisable name, a workspace scope, and only the permission scopes the integration needs — here 6 of the 24.
Manage Keys: every key with its scopes, expiry, last use, and the copy and revoke actions.
The MCP Integration section generates the setup for each client with your platform's real URL — here the one-command Claude Code path.
Registering a confidential client: name, exact redirect URIs, an optional workspace lock, and the scopes the app may request.