Intelligence Lab

Implemented

The building blocks in the current codebase, including Market discovery, tracking and comparisons. Availability depends on permissions, configuration and connected data.

• Guardian: versioned rules, source findings and purchase handling checks.
• Analytics: Roblox metric readings, SDK events, funnels, retention, segments, experiments, remote configs, stability signals and advertising readiness checks.
• Agents: questions and investigations with tool context, proposals and approval controls.
• Studio context: tasks, releases, LiveOps plans and manually entered ad and creator campaign plans.
• RobloxBench: a public suite for comparing AI models on Roblox development tasks (Luau, APIs, security, analytics). Scores appear only when published runs exist; empty boards stay empty.

Brain is the flagship direction for bringing these tools together. The homepage demo uses fictional example data to show the intended workflow. It does not read a connected game.

Active research

Open investigation topics for this refresh, with no completed study or measured improvement claimed.

• Distinguishing an unusual change from a normal busy or quiet day.
• Explaining possible causes while making uncertainty and missing data clear.
• Checking whether a recommended next step is useful to a studio, with supporting evidence.

These topics are not shipping commitments. We will publish methods and limitations alongside any results.

Future plans

• Compare recommendations with what a team actually shipped and what happened afterwards.
• Make it easier to follow evidence across analytics, audits, campaigns and LiveOps.
• Connect Market signals, creator campaign context and game outcomes while keeping public estimates distinct from measured revenue.

Plans may change. No automatic learning from outcomes, guaranteed growth or launch dates are promised.

Quarterly benchmark placeholders

No benchmark results are published here. Q1–Q4 are reporting slots without an assigned year, committed delivery date or implied completed evaluation. For model task scores, see RobloxBench.