A round is a segment of a project designed to move a document set closer to a more relevant and defensible state. Most rounds are performed to train the system to recognize the difference between responsive and non-responsive documents, but every project also includes a round dedicated to quality control, specifically to check on the accuracy of Relativity's coding decisions.

Note the following key Sample-Based Learning concepts:

  • Coded – a document is referred to as coded or tagged when it has a choice set in the Designation field. Reviewers perform this part of Sample-Based Learning.
  • Categorized – a document is referred to as categorized when it has a category set by Analytics. This is the automated part of review visible in the system-created fields after your first categorization.

This page contains the following sections:

See the following related topics:

Preparation before starting Sample-Based Learning rounds

Before starting a round refer to the following:

Types of rounds

There are four types of rounds. Each one has a purpose and can occur throughout the process. The type of round works in tandem with the saved search selected.

  • Control set rounds - For Relativity to automatically measure precision, recall, and determine the F1 score, it is necessary to set aside a control set of documents not utilized for training the system (not set as examples or excerpted). It is a best practice to create a control set as early as possible in the project, ideally as the very first round. See Control set round.
  • Training rounds - This is used to train the system. This is generally used initially prior to any other rounds, with the exception of the Control Set, if used. Training will occur until a significant portion of documents have a categorized value. See Training round.
  • Pre-coded seed rounds - To utilize previously-coded documents, you can select pre-coded sets of documents. These documents cannot be part of a currently active control set round. See Pre-coded seed round.
  • Quality control (QC) - This is sampling used to verify the categorization process. This will use one of the saved searches that includes documents with a value applied by the categorization process. Use quality control after you’ve already sufficiently trained the system. See Quality Control (QC) round.

Monitoring round progress

Deleting a round

If you need to clear any mistakes made during round creation, you can delete a round for which no sample documents have been newly coded without removing any data from the project or workspace.

The option to delete a round is available underneath the Finish Round option on the console when a sample set has been created but no new documents have been coded for designation or issues. It is available under the Start Round option only when a round was canceled by the system.

Delete round button

To delete a round:

  1. Click Delete Round.
  2. Click OK on the confirmation. The confirmation informs you that deleting a round:
    • Permanently removes it from the project
    • Deletes the sample set saved search
    • Disassociates the sample set documents previously associated with the round
    • Removes the value set for the Use as example field on the sample documents
  • When you initiate the deletion of the round, all affected reports update in real time.
  • A deleted round does not display in reports.
  • Relativity removes a deleted round from the list of rounds.
  • Round deletion is audited.
  • The next round created after a round is deleted replaces the deleted round and has the same number.
  • If you code or un-code the only coded document(s) in the round, the Delete Round option becomes available again.
  • You can delete a round canceled by the system, but only the last round that was canceled.