Discovery Testing

The Discovery Agent (Disco) is built for open-ended, explorative testing. Unlike the Verification Agent, which follows strict goals and test cases, Disco works with minimal instructions and adapts its behavior as it explores the game.

Disco is especially useful when:

1. You don’t have test cases yet: it can be used to surface issues, inspire new test cases, or just get broad testing coverage quickly.
2. You need quick checks:
   • Testing a new feature before deciding if it’s worth writing a full test case
   • Verifying a bug fix (“I fixed ‘A’ - can you check if it’s really gone?”)
   • Testing game areas that aren’t covered by regression test cases
   • Quickly exploring updates without setting up structured tests

Exploration Scope

Disco decides what to test based on what it encounters in the game. It starts with a basic plan, then updates it dynamically as it discovers new areas, screens, and features.

You can run Disco in two ways:

Unbounded Exploration

Run Disco “out of the box” with no scope (leave the goal and finish condition of the test case empty). The agent explores freely, trying to cover as much of the game as possible.

• Great for broad coverage when you just want to know “does anything break?”
• Quality improves if you provide game context (rules, mechanics, hints) so Disco knows what to expect and can make smarter decisions

Scoped Exploration

You can guide Disco by giving it a specific Exploration Scope. This is useful when you want focused testing without writing a full test case. For example:

• “Explore the shop and try buying items”
• “Test the new character editor”
• “I fixed a bug where the reward didn’t match the description — can you check that?” Scoped exploration lets you quickly validate new content, patches, or specific systems without heavy setup.

Exploration Budget

To control the depth and duration of exploration, you define an Exploration Budget using min and max turns in the test configuration.

What to Expect

While exploring, Disco automatically reports issues it encounters, such as:

• App crashes or freezes
• Localization problems (missing translations, incorrect text)
• Functionality issues (broken buttons, flows, menus)
• Contradicting behaviors (e.g., reward text says “100 gold” but only grants 50)
• Critical visual bugs (missing textures, broken UI elements)

Disco won’t catch every issue that a structured test case would find, and maybe even sometimes find false positives. But it’s a great way to quickly surface obvious problems and gain confidence that core systems are working as intended. Plus, reviewing the run and just dismissing any false positives is way quicker than going through the game yourself.

In short: Discovery is your flexible, lightweight way to test, whether that means catching bugs early, validating new content, or building confidence through exploration.