We want business users (aka end-users) to validate machine-generated data and make corrections as needed, based on their domain expertise. From a business perspective, there are two sub use cases: we may use the human-reviewed, machine-generated data…
Machine-generated data would be stored in the output dataset of an existing data pipeline. Each row would correspond to an item to validate. Columns would include:
Instead of exporting this dataset to Excel, we want end-users to access a web interface to validate and correct the data. In addition to the above columns, we would want 2 feedback columns: one to mark rows as valid (via checkboxes) and one to write comments.
Validation columns are used to indicate that a human saw what the machine did for a given row, and had the opportunity to make corrections or to fill in missing values if needed.
The webapp’s backend implements special behavior when a cell from a column named “Validated” or “Reviewed” is edited: values of all editable columns from the same row are logged (even if they weren’t edited). This allows the editlog to include not just the information that the row is valid, but also to record the actual values that were validated. This is particularly useful when those values were generated by an algorithm, because they may change if the algorithm changes.
As a result, there will be no missing value in the machine-generated and human-reviewed columns that are present in the edits dataset, for rows marked as valid.
You must be familiar with the initial How to Use guide before following the steps below.