Monitoring & results¶
The Projects list¶
Projects shows your folders and projects at a glance: a status indicator, a progress bar while a project runs, and small icon badges showing whether Vectorization and/or Smart Match were used — so you can tell how a project was configured without opening it:
The badges under a project name¶
Small icons under the date record how the project was configured, so you can read its setup without opening it. Hover any badge for its description:
| Badge | Meaning |
|---|---|
| ⚡ | Swift mode: auto-detected boost, trained immediately, dedup + quota + balance on download |
| 🌐 | Express mode: census-amplified boost — each boosted case is a distinct real person |
| ▤ | Vectorization applied: some columns were grouped into a single multi-select field |
| ✨ | Smart Match enabled |
Vectorization happens automatically: when the platform detects that a set
of columns belongs to one multi-select question (Q3_1, Q3_2, Q3_3), it
groups them into a single multi-select field instead of treating each option
as an independent question. The badge is how you know it was applied to a
given project — there is no setting to turn it on. In the charts dashboard a
grouped question is shown with its field count, e.g. Q3 (3), and reports
one combined set of metrics for the whole group.
Open a folder to see its projects. Status chips at the top filter by state (Created, Training and predicting, Finished):
A project moves through: Created → Ready to be processed → Being processed X% (the variable currently being trained is shown under the bar) → Finished.
The project detail screen¶
Click a project name to open it. This is the control room for the models:
- Top strip — real (Train) vs. synthetic (Predict) distribution for key demographics, plus the Smart Match recommendation badge.
- One row per question — with its model, histogram (train vs. predicted), per-question config (Tune model) and classes.
- Bottom bar — Remove answers, Clear text fields, the Smart Match toggle and Pause.
- Quality certificate — a panel with a plain verdict on the finished data (viable or not, scores, delivery checks). See Quality certificate.
Green check marks appear per question as processing completes:
Choosing a model per question¶
Every question defaults to Random Forest, and you can switch it:
Available models: Random Forest, Random Forest Lite, Linear Regression, Decision Tree, Logistic Regression, TabPFN, CatBoost.
The DATASETS menu¶
The DATASETS button (top right) opens the project's datasets — options enable as they become available:
Train Dataset is available from the start; Predicted Dataset, Post Process Dataset, Quota Manager and Quote dataset activate once processing produces them.
Quality: the charts dashboard¶
Every project includes a comparison dashboard showing, question by question, how the synthetic (predicted) answers compare to the real (train) answers. It is available early — even before a project finishes training — and also feeds the Smart Match recommendation (see The prediction engine). Open it with the bar-chart icon in the project row:
- The overview chart plots Average Deviation, Average Relative Deviation and Coverage for every question.
- Below it, each question shows Deviation, Coverage and Accuracy plus a bar chart comparing distributions — click View all to expand:
The three bars per answer option are the training data, the predicted data and the merged result — they should look nearly identical.
For a one-screen verdict on the whole project rather than question-by-question charts, use the Quality certificate.
Downloading results¶
Download Merged in the project row offers every output. What each file actually contains is described in What's inside each download:
| Option | Content |
|---|---|
| Full Merged File | Real + synthetic respondents in one Excel file |
| Full Merged File (SPSS) | The same, as an SPSS .sav |
| Postprocess | Optional cleanup steps applied to the generated rows before merging — see Post-processing |
| Quota Manager | The quota-adjusted dataset |
| Predicted Data Only | Only the synthetic respondents |
| Quality and performance report | Four-sheet audit of the delivered rows: fidelity, coverage, crosstab fidelity, coherence, duplicates and near-copies, plus which model answered each question — see its chapter |
| Current Predict | The current prediction snapshot |
Compare executions¶
When the feature is enabled for your organization, the project detail carries a Compare executions panel: "Keep the same data and profiles. Choose generation and speed independently; exact numeric rules remain protected in both modes."
| Control | Options |
|---|---|
| Generation | Selected models — each question answered by its own model — or Joint generation · experimental, which generates the answers together |
| Data preparation | Normal or Accelerated |
| Random seed | Fixes the run so it can be reproduced |
| Compute KNN fit on training data | Off by default; it costs extra time and, as the label says, does not measure generated quality |
Save for this project stores the settings; the panel reminds you to save before starting.
To compare rather than replace, give a Comparison copy name, choose Models for the copy (Keep each selected model, KNN, or Traditional models) and press Create comparison copy. The copy runs on the same data and profiles, and the Comparison results table lines the runs up: execution, generation, data preparation, prediction time, marginal fidelity, profile fidelity, numeric rule violations, observed block deviations and rows in duplicate groups. Each row links to that run's own project detail.
Executions lock
Once an execution has started it is locked: "This execution is locked. Changes below apply only when you create a new copy." The settings, Remove answers, Clear text fields and the Smart Match toggle are disabled from then on — comparisons happen between copies, never by editing a run that already produced data.
The results describe the stored generation, before any optional download filters such as post-processing.
Comparing projects¶
Compare projects (top of the Projects list) takes comma-separated project IDs and shows their key metrics side by side — status, average accuracy, deviation, coverage and number of fields:
