Repository navigation
Cycle 10 research gate: audit current AI, gaming, and accessibility pain points #71
Description
Goal
Run a fresh AI, gaming, and accessibility evidence sweep; check every credible signal against OpenController's actual ownership and existing issues; select only concrete, testable improvements. This is a decision gate, not a promise to manufacture implementation work.
Research log (2026-10-05)
Sources are individual reports, not prevalence estimates. Source coverage was broad across GitHub issues, Steam Community, OpenAI Community, and prior-cycle GitHub sources; it was not exhaustive.
| Source | Observed pain and workaround/counterevidence | OpenController fit and confidence |
|---|---|---|
| OpenAI Codex #44074, 2026-09-09 | One Windows Computer Use user reports game controls only held for about 1–4 ms despite requested timed key presses, and asks for bounded holds/down-up plus release on cancel/session close. No reliable workaround. | Detailed, high-confidence report for that backend; weak product fit. It concerns desktop keyboard/mouse injection; OpenController controls virtual gamepad state and already has bounded timed presses and cancellation handling. |
| Codex #35510, 2026-07-26; #46471, 2026-09-18 | Game input not received despite successful key calls; and Computer Use unable to target/capture a Godot window. A follow-up on #35510 reports Satisfactory hotkeys failed while Notepad/Calculator worked. | Detailed cases, but OS/desktop input and window capture are outside SDK ownership. Not evidence for keyboard injection or game acknowledgement in OpenController. |
| OpenAI Community: Computer Use game play, 2026-09-26 | One user says Pokémon Silver emulator setup, dialogue, and movement consume much of Computer Use allowance; asks for lower-cost/longer sessions. | Provider quota and emulator session issues, not controller SDK work. |
| Proton #10160, 2026-09-19; #10187, 2026-09-27 | Reports include controller enumeration failure and Xbox Elite reconnect behavior across Proton versions; no reliable fix in the reports. | Detailed individual cases, but Wine/Proton enumeration and game-specific reconnect state are upstream. |
| Godot #123704, 2026-09-22 | Steam Deck Gaming Mode lost built-in controls on one Godot 4.8 dev build while Desktop Mode/dev5 worked; the report traced to bundled SDL/environment integration and was closed with an upstream fix path. | Specific reproduction, low recurrence evidence; engine/SDL integration rather than OpenController output. |
| SDL #16420, 2026-09-30; #16441, 2026-10-04 | One report describes Switch 2 Pro motion stopping after about an hour with buttons still working; another reports one 8BitDo Bluetooth LE device not enumerating on macOS while USB works. | Narrow driver/timestamp and physical-input enumeration reports; no matching OpenController-owned defect. |
| Sunshine #5670, 2026-09-10 | One streaming user reports shoulder-button repeat or no controller after a virtual-HID migration; local controller works. | Host streaming/driver behavior, not OpenController report serialization or readiness evidence. |
| Steam Community: one-handed/foot play, 2023-05-02 | One user asks about trackpad/joystick movement while using a mouse for aim to learn play with one foot; a reply says simultaneous use worked for them. | High confidence for this specific Steam Input anecdote, but no demonstrated OpenController gap or hardware compatibility. |
| Steam Community: arthritis/carpal tunnel, 2016-11-21 to 2017-02-04; “painful”, 2015-11-08 | Reports conflict: remapping, gyro, high sensitivity, and lighter grip helped some; another buyer returned the controller because its physical layout was uncomfortable. | Firsthand but old, device-specific ergonomic reports. Configuration cannot address physical fit, and OpenController cannot validate suitability. |
Coverage limits
Reddit and its public mirrors, Google/Bing/DuckDuckGo search, Xbox.com, AbleGamers, SpecialEffect, Anthropic Community, and Google AI Developers endpoints were blocked by the outbound proxy (403). Steam Community pages were reachable, but search later returned 429. GitHub, OpenAI Community, and Hacker News searches were reachable. No first-party OpenController user telemetry or hardware/game verification is available. These gaps do not prove demand is absent.
Repository audit and ranking
A Luna repository audit found no new, distinct candidate beyond existing issues #8–#70. The externally reported keyboard timing is not a keyboard injection gap in this gamepad SDK; upstream enumeration, SDL, remote-play, and ergonomic reports remain outside the SDK's verified ownership. Recent code findings #33 and #34 are already fixed in PRs #69 and #70.
| Candidate | Decision | Reason |
|---|---|---|
| New Cycle 10 implementation issue | None selected | No externally supported report mapped to an uncovered, OpenController-owned behavior with a separate testable acceptance criterion. |
| #46 helper readiness handshake | Remains deferred | The process adapter marks connected after spawn, but adopting a readiness protocol still requires a helper-owned acknowledgement/version contract. Current reports are host-specific and do not resolve that dependency. |
| #60 persistent-state expiry lease | Remains deferred | Godot/SDL reports concern consumer/driver behavior, not a verified OpenController stale-state incident; lease timer races and cleanup remain material risks. |
| #21 host/game compatibility, #28 latency boundaries, #62 Windows VHF test path | Remain deferred | Hardware, a named consumer, or supported driver test path is needed to make trustworthy claims. |
DAG and decision
flowchart TD
R[AI, gaming, accessibility source sweep] --> A[Repository and duplicate-issue audit]
A --> D{Uncovered SDK-owned candidate with evidence?}
D -->|No| C[Record no-build decision; retain deferred leads]
D -->|Yes| I[Create scoped implementation issues and prerequisites]
I --> W[Parallel Luna build lanes in isolated worktrees]
W --> V[Exact-head validation and independent review]
V --> M[Merge in dependency order and close cycle]
For this gate the decision is No. No implementation issues or PRs are created from these signals. The next recurring sweep should seek current, repeated reports with a concrete SDK boundary, and may revisit #46 or #60 only when their prerequisite/evidence changes. This avoids turning unrelated upstream failures or isolated anecdotes into product claims.
Acceptance checklist
- Search AI-agent/game control, gaming/platform, and accessibility sources; retain direct links and dates.
- Record observed behavior, workarounds, ownership fit, confidence, and material source gaps.
- Audit against current repository and issues Preserve agent context and diagnostics in replay errors #8–fix(core): reject negative action durations #70 with an independent Luna lane.
- Rank existing deferred leads and reject duplicates/speculative new work.
- Record the no-build decision and preserve the recurring research/build/review/merge process in goal issue Run a recurring evidence-backed AI and gaming improvement cycle #55.
- Merge the roadmap closeout in PR docs(roadmap): close cycle 9 and record cycle 10 decision #72 (
fb70a63ea962fca5ae06e2ab2815a84a6e01bc63); close gate Cycle 10 research gate: audit current AI, gaming, and accessibility pain points #71 and milestone 9.
Activity