> ## Documentation Index
> Fetch the complete documentation index at: https://labs.laer.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Configure and run an Issue Review workflow

Once you have uploaded your data to Epiq AI and created an Issue Review workflow from a template, you are ready to configure and run your Issue Review workflow.

An Issue Review workflow starts with a single Issue Classification step. You build out the rest of the pipeline to match your review protocol.

To configure and run an Issue Review workflow, complete the following procedure.

1. In the **Project space** menu, select **Workflows**.
2. Select your Issue Review workflow.
3. Select the **Input Data** tab and import the dataset containing the documents to review.
4. Select the **Steps** tab and configure the workflow's steps.

<Accordion title="Link a review protocol">
  The protocol is the master list of coding instructions for the review. Linking it lets each step import the slice of instructions it is responsible for, and records where each instruction came from.

  1. Open the step's instruction input and select the **Protocol** tab.
  2. Select **Select existing** to choose a protocol already in the collection, or **Import new** to upload one.
  3. Select the protocol.

  Linking a protocol does not import anything on its own. Import the instructions onto each step separately, as described below.

  If your matter has no written protocol, skip this step and enter tags and instructions directly on each step instead.
</Accordion>

<Accordion title="Issue Classification">
  This step codes documents against a set of tags. Each tag carries an instruction stating when it applies, and a document may receive zero, one, or several tags.

  1. Select the **Issue Classification** step.
  2. Select **Instructions** and do the following:
     1. Review the **General Instructions**. This is the standard applied to every document — edit it to record how the catalogue as a whole should be applied.
     2. To import from the linked protocol, select the **Protocol** tab, select the instructions this step should carry, and select **Import selected**. Use **Select all** to take the whole protocol where it fits within the step's limit.
     3. To enter instructions directly, select the **Instructions** tab and add each tag with the instruction stating when it applies.
     4. Edit imported or entered instructions as needed for your matter.
  3. Select **Tags** and confirm the tags on the step and their grouping.
  4. Select **Model** and do the following:
     1. Update model settings as needed.
     2. Select **Metrics**.
     3. Modify error categories as desired and select **Update Categories**.
     4. Select **Update**.
  5. Select **Output** and choose the column this step writes its tags to. Create a new column, or select an existing column to merge this step's tags into it.
  6. Optional. Click **Run Step** to run this step individually, or wait to deploy the full workflow after you configure all steps.
  7. Select the back arrow to return to the **Configuration** tab.

  A step holds up to 50 tags. When your catalogue is larger, add another Issue Classification step and import the next slice of the protocol onto it.
</Accordion>

<Accordion title="Issue Field Extraction">
  This step captures structured fields for each document. It runs after the classification steps and treats each document's predicted tags as context, so extracted values reflect the coding already applied.

  1. Select **Add Step**, then select **Issue Field Extraction**.
  2. Select the step, then select **Instructions** and do the following:
     1. Review the **General Instructions**.
     2. Import field instructions from the linked protocol, or enter the fields directly with a description of what each should contain.
     3. Edit the fields as needed for your matter.
  3. Optional. Click **Run Step** to run this step individually, or wait to deploy the full workflow after you configure all steps.
  4. Select the back arrow to return to the **Configuration** tab.

  A step holds up to 50 fields, and field names must be unique across all Issue Field Extraction steps on the workflow. When your field catalogue is larger, add another extraction step with a disjoint set of field names.

  This step is optional. A workflow with classification steps alone is fully runnable.
</Accordion>

<Accordion title="Add or remove a step">
  1. On the **Configuration** tab, select **Add Step**.
  2. In the **Add a step** dialog, select **Issue Classification** or **Issue Field Extraction**.
  3. Configure the new step as described above.

  To remove a step, open its card and select the delete action. Deleting a step permanently removes its tags and instructions.

  You cannot delete the workflow's only Issue Classification step. Field extraction steps can be removed freely.
</Accordion>

5. Select **Deploy** to run all steps in order.

<Info>
  Running the classification steps on their own first is the recommended sequence. Classification establishes the coding, and field extraction uses that coding as context — so it is worth confirming the tags are being applied correctly before running the extraction pass over the population.
</Info>

### Review and correct results

Issue coding is iterative. Use the following procedure to review results and refine the configuration.

1. Select the **Steps** tab and select a classification step.
2. On the **Results** tab, review the tags applied and the model's reasoning for them.
3. Apply or remove tags on documents you have decided on. Reviewer decisions take precedence over the workflow's predictions.
4. If a tag is being applied too broadly or too narrowly, select the **Instructions** tab and edit that tag's instruction rather than correcting each document individually.
5. Click **Run Step** to re-run the step with your corrections.
6. Re-run any Issue Field Extraction steps afterwards, so the extracted fields reflect the corrected coding.

<Warning>
  Re-running a step clears and rebuilds that step's results. Where a later step depends on an earlier one, re-run them in order so the output stays consistent.
</Warning>

### Update the workflow when the protocol changes

When the protocol is revised, import the affected instructions again onto the step that carries them. Instructions matching what the step already holds are updated in place rather than duplicated, so re-importing after a protocol edit is safe.

Re-run the affected steps to apply the revised coding standard.