Section 9: QA Process - How to Review an Agent

Why QA Matters

Building a great agent doesn’t stop at setup — it ends with accuracy.

Your QA process is what ensures your agent:

Extracts the right data
Formats it cleanly
Grounds its outputs for traceability
Scales confidently across files

This section walks through the tools and best practices available in Go to streamline the review process and make sure your agent delivers clean, accurate, and auditable results.

How to QA an Agent Like a Pro

Use the Review Tab to Organize Your Outputs

The Review tab gives you a visual, drag-and-drop layout where you can:

Resize, move, and rearrange property widgets
Create a custom output layout that aligns with your QA flow
Group key metrics visually (e.g. “Financials” vs. “Risk Factors”)

This is your QA dashboard. Use it to:

Focus attention on the most important outputs
Quickly validate values across rows
Organize views around specific personas (Analyst, Legal, Ops)

Leverage AI Citations for Verifiability

Text and Number properties support AI Citations, which allow you to:

Highlight where an output came from in the source file
Jump directly to that page or sentence
Visually verify the bounding box in context

Best Practice:

For any property pulling directly from a file (e.g. NOI, Cap Rate, Address), always enable citations in the tool settings. It makes verifying outputs 10x faster.

Prefer Single & Multi-Select Where Possible

Using Select properties improves QA in two major ways:

Efficiency: Fewer possible outputs → faster review
Clarity: Clearly defined dropdowns make errors obvious

You can also:

Color code options for better scannability
Route based on select values (e.g. “Needs Review” view)

Bonus: They also reduce token usage and hallucination, improving both performance and cost.

Use the “Expand Entity” View for Side-by-Side QA

Clicking Expand Entity on any row opens a split-screen review mode:

Left: The source file (PDF, webpage, contract, etc.)
Right: All extracted properties listed vertically

This is ideal for:

Spot-checking values in full context
Viewing citation targets inline
Quickly editing outputs or reviewing model reasoning
Validating collections or long form extractions

It’s your go-to view when reviewing:

A new agent for the first time
Long documents or financial tables
Multi-property summaries like executive memos

Suggested QA Best Practices

Combine these workflows for a scalable and reliable review loop:

Best Practice	Why It Works
Start with Review View	Visually prioritize your most critical outputs
Always test with 3–5 diverse, heavy files you know inside-out	Surfaces edge cases early and accelerates QA by using files you’re already familiar with
Add Selects to flag rows	Route exceptions to “Needs Review” views for faster triage
Use JSON + Python for grouped logic	Simplifies debugging and makes structured outputs easier to validate
Document what “good” looks like	Aligns team members reviewing outputs and prevents subjective QA bottlenecks

Pro Tip: Use files that differ in structure, length, and formatting — but that you’ve personally reviewed before — so you can immediately tell when something’s off.

Summary: A QA Checklist

Before you call your agent “done,” ask:

Are all key properties visible and organized in Review view?
Are citations enabled on all text/number fields that pull from source documents?
Are select fields being used wherever options are discrete and finite?
Have you opened a few rows using Expand Entity to verify side-by-side behavior?
Have you run real files and caught 1–2 edge cases?
Are your outputs clear, traceable, and scalable?
Do you have sub-workflows with human-in-the-loop review stages?

If the answer to all of these is yes — you’re ready to deploy.