Skip to main content
Metrics are available on each Issue Classification step. Select the step from the Flow tab, then select the Metrics tab. Each classification step has its own model covering the tags on that step, so a workflow with several classification steps has metrics for each of them. Review them separately — the steps carry different tags and may well perform differently.
Issue Review differs from Privilege Review in one way that shapes how metrics should be read. A privilege model has a single negative outcome to measure against, so one overall score is meaningful. An issue catalogue has no equivalent — every tag is its own question. Read the Tags tab first and treat the overall numbers as context rather than as the result.

​
Analyzing coding errors

Before the Prediction Overview and Error Analysis tabs have data, you need to run a feedback batch. This instructs the system to compare the AI’s coding against your reviewers’ decisions and generate explanations for any mistakes. To run feedback, select the documents you want to evaluate and click Analyze Errors. The process runs in the background. Once complete, the tabs will reflect the results. You can re-run feedback any time to refresh the data when additional review work has been completed.

​
Model Metrics

This tab shows how the step’s model is performing based on a set of documents that have already been reviewed and coded. Summary cards at the top:
  • Total coded documents: the number of documents that have been manually reviewed.
  • Total coded positive: how many of those carried a tag from this step.
  • Total coded negative: how many carried none.
Relevance Score chart — a bar chart showing how many documents fall into each confidence range. Bars to the left of the cutoff threshold are classified as negative; bars to the right are classified as positive. You can drag the threshold slider to explore how changing the cutoff affects your results.

​
Prediction Overview

This tab gives you a high-level summary of what the AI coded across your entire document set, and how much of that work has been reviewed by a human. Summary cards:
  • Total Documents: the total number of documents the AI has scored.
  • Positive and Negative counts for the step’s tags.
Tags Analysis chart: a stacked bar chart breaking down predictions by tag. For each tag:
  • The green portion shows documents that have been manually reviewed.
  • The gray portion shows documents that have not yet been reviewed.
This chart is the fastest read on two things: which issues are carrying the population, and which still have unreviewed coding. A tag with a tall bar and no green has not been checked by anyone.

​
Tags

This tab reports the model’s performance for each tag individually, which is the view that matters most in an issue review. For each tag it reports:
  • Whether the tag trained, and its cross-validated performance
  • Precision and recall against each control set, where the reviewed sample was large enough to measure them
  • A performance curve, where the sample was large enough to produce one
Tags with too few reviewed documents are reported as such rather than given a misleading score. If a tag matters to the matter and has no metrics, the answer is to review more documents carrying it, not to read the overall model number in its place. Use this tab to decide where to spend QC effort: a tag with strong precision and weak recall needs a broader instruction, and the reverse needs a narrower one.

​
Error Analysis

This tab is where you can dig into the AI’s mistakes, understanding why errors occurred.

​
Filters

Three filters at the top let you focus on a specific subset of results:
  • Tags: narrow down to a specific issue tag.
  • Error Type: filter by:
    • All Errors: show everything
    • False Positives: documents the AI tagged, but a reviewer said should not have been
    • False Negatives: documents the AI missed
  • Confidence Level: focus on predictions the AI was Low (0–60%), Medium (60–80%), or High (80–100%) confident about.
Filter by tag before anything else. Errors pooled across an issue catalogue rarely have a single cause, and the pattern only becomes visible one tag at a time.

​
Error Categories Distribution

A pie chart showing which error categories appear most frequently across all mistakes. Each slice represents a category of reason the AI got it wrong. Hover over any slice to see the exact count and percentage. Error categories can be customized in the Configuration tab.

​
Error Analysis Table

A detailed row-by-row table of each document the AI was evaluated on. Columns include: You can apply additional filters using the filter bar.