blushneybusiness-glitch/roblox-omniskills/rbx-qa-devices

rbx-qa-devices

When to load

Publish gauntlet phase 7 (rbx-qa-gauntlet loads this automatically), any "works on my machine" UI report, before enabling console access, and whenever a game has only ever been run in a default 16:9 Studio viewport.

The tier ladder

• T1 — low phone (the design floor): 2–3 GB RAM, weak GPU, small screen, touch only.

Hunt here: load time to first input, memory ceiling, touch targets, text size. Load-order races hit T1 hardest — a 15-second load widens every timing window, so re-run the rbx-qa-scale race audit with T1 in mind.

• T2 — mid phone / tablet (the majority): hunt sustained problems, not first-minute

ones — session-long frame stability and battery-eating permanent full-screen particles.

• T3 — desktop (the dev's lie machine): everything works here; verify only ultrawide

UI anchoring and that keybinds are discoverable.

• T4 — console: full gamepad path through EVERY menu including purchase prompts

(GuiService.SelectedObject chains), and text readable at couch distance — 1.5× the desktop size rule.

Studio emulation

Test tab → Device Emulator. Presets emulate resolution, DPI, safe-area insets, and input type — they do NOT emulate CPU, GPU, memory, or thermals. An emulated pass proves layout only. The closest Studio gets to T1 performance is graphics quality level 1 plus a deliberately skeptical eye; a real low-end device test outranks both.

Aspect-ratio sweep

Minimum three emulated checks per UI surface: a 19.5:9 phone (390×844 class, notch + safe area), a 4:3 tablet (1024×768), and 16:9 desktop. Offset-positioned UI that looks centered at 16:9 clips or drifts at 19.5:9; scale-based layout with UIAspectRatioConstraint / UISizeConstraint survives all three. Phone floor: touch targets ≥ 44×44 px, body text ≥ 14 pt equivalent.

Input types

TouchEnabled, KeyboardEnabled, MouseEnabled, GamepadEnabled are capabilities, not identities — a touch laptop reports keyboard AND touch, a phone can have a gamepad paired. Never branch "if TouchEnabled then mobile"; adapt live per UserInputService:GetLastInputType() and LastInputTypeChanged. Every action bound through ContextActionService gets its touch button for free — verify the buttons don't stack over the jump control. Console test: unplug the mouse mentally and drive every flow with selection navigation alone.

Memory tiers

Watch client memory in the Developer Console (F9 → Memory) during the device pass. T1 devices get killed by the OS well before their nominal RAM — a client sitting near 1 GB is already in crash territory on a 2 GB phone, and the crash produces no error log, just a silent disconnect that shows up as churn. Textures and meshes are the usual bulk; budgets live in rbx-perf-mobile and quality-bars.

Sharp edges

• CRITICAL — Trusting emulation for performance: smooth in the emulator, slideshow on a

real T1 phone. Fix: quality level 1 + the rbx-perf-mobile budgets, and one real-device run before publish when hardware exists.

• HIGH — Fairness gated on effects that quality level 1 strips: at quality 1, shadows

and many particles vanish — if the enemy telegraph or killbrick warning is particle-only, low-end players die blind. Fix: duplicate critical signals in geometry, color, or UI.

• HIGH — Startup platform detection (TouchEnabled ⇒ mobile UI forever): misclassifies

touch laptops and gamepad-on-phone players. Fix: react to LastInputTypeChanged.

• MEDIUM — Testing one aspect ratio and shipping: the 19.5:9 notch class is the most

common phone shape and the most common clip site. Fix: the three-ratio sweep, every UI.

Load with

rbx-qa-gauntlet (the device phase), rbx-qa-playtest (run the emulated-viewport scenarios), rbx-perf-mobile (T1 budgets), rbx-ui-mobile (touch targets, safe areas), rbx-input-mobile + rbx-input-gamepad (the input paths under test).