The schema does much of the work
A vague column such as “Relevance” asks the model to invent your review criteria. A better definition states the issue, what qualifies, what does not, and what evidence the cell must contain. One wide cell containing date, narrative, interpretation, and recommendation is difficult to audit.
Harvey's release notes document Review Table features, including source-snippet exports and cell verification changes over time. Check the current help material for the account you use. The column design below is platform-independent; it is not a claim that a particular tool will extract it perfectly. [source]
Define a first table
| Column | Definition |
|---|---|
| Document ID | Stable identifier from the manifest |
| Document date | Date of this document; do not substitute an event date |
| Notice discussed? | Yes, no, or uncertain, using the stated issue definition |
| Relevant event | Neutral account of one identified event; preserve multiple events separately |
| Evidence | Exact passage and location |
| Qualification | Condition, dispute, or ambiguity affecting the event |
| Review status | Unreviewed, checked, correction required, or inaccessible |
If the tool uses one document per row, multiple events may require a separate event table or an explicit list within the event cell. Do not force the underlying record into a shape that conceals distinctions.
Write a column instruction
Identify whether this document contains evidence about receipt of the revised delivery schedule. Distinguish the original schedule from the revised schedule. A reference to sending alone is not evidence of receipt. Return yes, no, or uncertain; quote the relevant words and identify their location. If an attachment is required to identify the schedule and is missing, return uncertain and explain why.This definition gives both positive and negative boundaries. It also specifies what to do when the evidence does not resolve the issue.
Calibrate on contrasting documents
Use five to ten documents: a clear positive, a clear negative, an ambiguous reference, a missing attachment, and an email chain with repeated text. Review every cell in this sample. Revise the definition where an ambiguity caused a predictable error. Keep the sample and its reviewed answers for future comparison.
When the definition changes, identify which previously generated cells depend on it. Regenerate or re-review them. Otherwise the table can mix inconsistent coding standards while presenting a uniform surface.
Try it: repair a column
Column: “Is this helpful to our case? Be generous.” What is wrong with it, and how could it be improved?
Check your understanding
It lacks an issue definition and invites one-sided selection. Replace it with a bounded question, supporting passage, adverse or qualifying content, and review status. If you want a separate advocacy assessment, keep it separate from the extraction and explain the criteria. Evidence organization and strategic evaluation serve different purposes.