Roblox House Stack — Research Frontier: the Autonomous Dev Pipeline

Overview

Every problem below serves one studio-wide direction: agents shipping verified features end-to-end with near-zero owner time. Everything here is OPEN or CANDIDATE — nothing is shipped; each problem states why current practice fails, what asset the house stack already provides, the first three concrete steps in the current project, and a falsifiable "you have a result when..." milestone.

The recurring bottleneck: in house-stack projects, feature work ends code-done with verification owed — cards pile up in the Testing lane of docs/BOARD.html blocked on the same gate: a human opening Roblox Studio. Measure it in your project before choosing a problem: open the board, count Testing-lane cards, and check each card's evidence status (evidence bar per roblox-validation-and-qa). When all agent-executable work in that lane is exhausted, that gate is the frontier. (Learned in a shipped firebit game, 2026: five features sat simultaneously code-done, all waiting on one manual playtest gate.)

When to use

• Choosing what to work on when feature work is blocked on owner verification.
• Asked "what are the open problems / research directions / what can be automated?"
• Starting work on scripted playtests, headless sims, telemetry-driven tuning, or test coverage as a project (not a one-off).

When NOT to use