Legal Services Offshore blog
Set a document review coding rule before the first batch
A coding guide that gives reviewers the same definitions, examples, and escalation path before document review begins.

Write the coding question in plain language
A coding rule is only useful when the question is clear. Write the question the review must answer, such as whether a document responds to a specific request or whether it contains a particular type of communication. Name the population, the date range, and the sources included. If the question changes during the review, version the rule rather than quietly editing it.
Give each code a definition and a counterexample. A reviewer should know what qualifies, what does not, and what to do when the document sits between the two. Use the same terms the review platform uses so the code and the export do not drift apart. Keep the guide short enough to use during a review but specific enough that two reviewers reach the same result on a calibration set.
Calibrate on a small set before production
Select a calibration set that includes ordinary documents, edge cases, and at least one document that should be escalated. Ask two reviewers to code the set independently, then compare their answers. Discuss every disagreement and decide whether the rule, the example, or the reviewer's reading caused it. Record the decision and update the guide.
Do not start the main batch until the calibration set is stable. If the reviewers still disagree after the discussion, the question may be too broad for the current rule. Narrow it or add a reviewer step. A short calibration period is cheaper than correcting thousands of documents later, and it gives the firm evidence that the coding standard was understood.
Make escalation part of the code set
Some documents should not be coded by the first-line reviewer at all. Identify the triggers in advance: a new legal issue, a communication with counsel, a document that may be privileged, a request for advice, or a conflict with an earlier decision. Give the reviewer a clear action, such as flag and route, rather than asking them to choose the closest code. Record the trigger in the coding guide so the decision does not depend on memory.
Record the escalation reason and the owner who will decide. Do not let an escalated document disappear from the queue or get coded provisionally without a note. If escalations cluster around one document type or one request, the coding guide may need a new category or a narrower scope. The escalation path is part of the rule, not an exception to it.
Audit the output before the batch closes
Sample completed documents across reviewers and code values. Check that the code matches the document, the supporting note is present, and the escalation path was used when required. Compare the sample with the calibration decisions. If a reviewer applied a code consistently but differently from the guide, the guide may be ambiguous; if the reviewer ignored the guide, address that separately.
Record the sample size, the findings, and the corrections. Keep the reviewed version of the guide with the batch output. A coding rule that cannot be audited is only a preference. The audit closes the loop by showing whether the written standard was actually applied and whether the next batch needs a clearer instruction.