Privilege results

Once the Populate Privilege Results pipeline step completes successfully results will begin to populate on documents at roughly 2,000 documents per hour. To check if the results have been populated, run a search that looks at Priv::Prediction is SET and Priv Project=[the name of the project you are running] and confirm the document count matches number of documents in the original project saved search.

Some results are populated to fields directly on the Document object, while others are populated to aiR for Privilege Document Results Object fields and then reflected onto the Document object through the multiple object Priv field. See all fields that are name spaced with "Priv::".

Results will never be overwritten or updated after the completion of Populate Privilege Results.

Result fields

Following is a list of all the results fields that are populated:

Document field	Field type	Description	Notes
Priv	Single result object	Links the document to its corresponding aiR for Privilege Document Results object, which is used to reflect results onto the Document object via the Priv:: namespace.	This field can be ignored.
Privilege Project	Multiple object	The project where this document was analyzed.
Priv:: Prediction	Single choice	Whether a document is Privileged or Not Privileged.
Priv::Category	Single choice	The category of privilege is dependent on the Record Type field set during General Settings. Documents with an Email record type will have the following Not Privileged Categories: Not Privileged Junk Document Documents with an Email record type will have the following Privileged Categories: Wholly Privileged Privilege Redaction Privileged Individual / No Privileged Content Borderline All other documents will have the following Not Privileged Categories: Not Privileged Junk Document All other documents will have the following Privileged Categories: Attorney Involvement Related to Legal Matter Legalese / No Privileged Content Borderline	For a list of privilege categories and their descriptions see Privilege categories.
Priv::Rationale	Long text	A description for why the document received the prediction it did.
Priv::Considerations	Long text	A description of what might change the AI’s prediction.
Priv::Citation	Long text	Excerpt from the Document which demonstrates the prediction.
Priv::Error Details	Long text	Contains error details for result level errors	For a list of all document errors and their descriptions see How document errors are handled.
Priv::Ungrounded Citation Present	Yes/No	When the aiR results citation cannot be found anywhere in the document text. Review the Ungrounded Citation and corresponding results with caution.	One of the citations has been identified as an Ungrounded Citation. Carefully review the Ungrounded Citation in the Priv::Citations field, Rationale, and Consideration. For more information, see Ungrounded Citations and Conflations
Priv::Draft Log Description	Long text	Draft Privilege Log entry.	This functionality is currently in Advanced Access. Only documents that have a Decision of Privileged and are not in the Category of Errored will receive a Log Description.
Priv Recipient Aliases (all email segments)	Multiple object	Lists all aliases who are recipients on an email. Includes all email segments.
Priv Sender Aliases (all email segments)	Multiple object	Lists all entities who are senders on an email chain. Includes all email segments.
Priv Recipient Entities (all email segments)	Multiple object	Lists all entities who are recipients on an email
Priv Sender Entities (all email segments)	Long text	Lists the entity that is sending the email

Note: Searching in the UI across direct document fields vs. those reflected from the aiR for Privilege Document Results object fields is a slightly different experience.

Privilege categories

Following is a list of privilege categories and their descriptions:

Document type	Category	Description
Attachments/Standalone Documents	Attorney Involvement	Document includes information that reflects attorney advice.
Attachments/Standalone Documents	Related to Legal Matter	Document reflects information about an ongoing legal matter but does not include advice from an attorney.
Attachments/Standalone Documents	Legalese / No Privileged Content	Document includes lots of legal jargon and legal content, however does not reflect the advice of an attorney.
Email	Wholly Privileged	Email in its entirety reflects privileged information or reflects attorney advice. For example, an entire email chain is a discussion between an attorney and a client regarding legal advice.
Email	Privilege Redaction	Email contains privileged information or attorney advice, but not all of its content is privileged. For example, an email chain that has privilege in the last in time email, but earlier emails in the chain do not contain privilege.
Email	Privileged Individual / No Privileged Content	Email includes privileged individuals, but no attorney advice is being sought or given.
Email/Attachments/Standalone Documents	Borderline	Inclusion of privileged information is unclear.
Email/Attachments/Standalone Documents	Not Privileged	No clear explanation for privilege
Email/Attachments/Standalone Documents	Junk Document	Contains no useful information to analyze. For example, random characters or an empty document.

How document errors are handled

If aiR for Privilege encounters a problem when analyzing a document, it will not return results for that document. Instead, it scores the document as -1 and returns an error message in the Error Details column. Your organization is not charged for any errored documents.

Possible error messages

Following are possible error messages and their descriptions:

Error message	Description
Failure to parse results	The large language model (LLM) encountered an error.
System error occurred	The large language model (LLM) encountered an error.
All citations have been identified as Ungrounded Citations.	All the Citations for this document have been identified as an Ungrounded Citation. For more information, see Ungrounded Citations and Conflations.
All citations have been identified as conflations.	All the Citations for this document have been identified as a conflation. For more information, see  Ungrounded Citations and Conflations.
Document text is empty	The extracted text of the document was empty.
Document text is too short to generate results	There was not enough extracted text to analyze in the document.
Document text is too long to generate results	The document's extracted text was too long to analyze.
Model could not process document	A communication error occurred between the large language model (LLM) and Relativity.

If you receive an error message, carefully review the document for privilege manually.

Ungrounded Citations and Conflations

Two types of errors deserve special mention:

Ungrounded Citations—occur when the aiR for Privilege results citation cannot be found anywhere in the document text. This is usually caused by formatting issues. However, just in case the large language model (LLM) is citing sentences without a source, aiR for Privilege marks it as a possible Ungrounded Citation.
Conflations—occur when the aiR for Privilege results citation comes from something other than the document itself, but is still part of the full prompt. For example, it might cite text that was part of the Prompt Criteria instead of the document's extracted text.

When aiR for Privilege receives the analysis results from the LLM it checks all citations against the document’s extracted text to search for Conflations or Ungrounded Citations.

If a document has a combination of at least one valid citation and an Ungrounded Citation:

Review all results with caution.
All Valid Citations and Ungrounded Citations will be written back to the Priv::Citation field.
All Ungrounded Citations will be flagged in the Citation Field.
All Rationale and Considerations will receive a warning: [Based on Ungrounded Citation]
The field “Priv::Ungrounded Citation Detected” will receive a flag of Yes.

If a document has a combination of at least one valid citation and a Conflation:

Conflations will be omitted from the Priv::Citaton field.

If a document consists of all Ungrounded Citations and/or Conflations without a valid citation:

The document will receive a Priv::Category of Errored.
Please review the document manually.

Propagating to family members

aiR for Privilege focuses on analyzing the text within the four corners of each document. This means that privilege determinations are based solely on the content of the individual document and are not automatically propagated to family member documents such as email attachments or related threads.

To code family members for privilege, we recommend first reviewing all the documents with a prediction of Privileged independently and then reviewing the family members of documents you coded as privileged.

Reviewing family members for privilege

Following is the recommended approach for reviewing family members for privilege:

Initial coding:
- Review all documents that our tool has predicted to be privileged to make your final privilege determination.
- Search Conditions:
  - Priv::Prediction = “Privileged”
Creating a saved search of related documents:
- Once your initial coding is complete, create a saved search of all the documents you coded/confirmed to be privileged.
- Modify your search to include family members such as attachments, email threads, etc.
- Search Conditions:
  - Privilege Coding =“Privileged” (or equivalent field you used in your review for the final privilege calls.)
  - Related Items = “+Family”
Review Documents
- Carefully review the related documents. Review the privileged document and individually tag family members as privileged where necessary, ensuring that all documents in the family group are coded correctly.