Cycle 5 research gate: select the next evidence-backed improvement #29

Goal

Complete Cycle 5 as a bounded, evidence-led cycle: refresh AI and gaming workflow reports, audit current OpenController ownership and gaps, rank candidates, then select one measurable improvement or record a no-build decision. End with independent review, required checks, merge, roadmap closeout, and a Cycle 6 research gate.

Cycle 4 implementation and closeout are complete: issue #22 merged in PR #25 at 8eb744c, and PR #30 closed gate #23. This gate is active.

Fresh source log (2026-10-04)

These are dated public reports and requests, not prevalence estimates. External host and application failures are not evidence of defects in OpenController.

Coverage limits

• HN Algolia and OpenAI Discourse JSON were reachable for targeted searches. Reddit, Google search, and SDL Discourse search returned HTTP 403.
• Steam Community search worked, but the relevant results found were old and low value for this decision.
• GitHub issue/API searches covered Steam for Linux, SDL-related candidates, Selkies/ROMM, Jellyfin Web, Cockpit, Tanks, and latency seeds. Search was targeted, not exhaustive.
• No clear current recurring AI-agent-specific controller problem was found. No result count, comments, or reactions are used as a prevalence estimate. Accessibility-community coverage remains open.

Repository audit and ranking

Audit target: integrated main at 62e8b95, including the core ControllerHub, browser exports, native bridge identity, tests, latency fields, and current README. The ControllerHub README called hub.createController, which the public class does not implement; it exposes add/get/has/list/states/disconnectAll. Existing tests covered two dry-run controllers with distinct IDs and profile entries but did not assert separate state, duplicate-ID rejection, or reuse after disconnectAll. The SDK does not discover physical devices or own browser/OS slot identity.

Decision and implementation DAG

Selection: issue #27 corrected the README API example, added a typed runnable counterpart, made duplicate rejection cover concurrent additions and failed construction cleanup, explicitly asserted per-controller state isolation, tested ID reuse after disconnectAll, and documented caller-owned IDs plus the requirement to await additions before disconnectAll. It does not add physical discovery, automatic mapping, native adapter sharing, or host/browser reconnect claims.

Issues #26 and #28 remain deferred. Issue #27 closed when PR #32 merged; PR #31 records the decision. This gate now awaits its roadmap closeout. Cycle 6 gate #35 is open and blocked by both #29 and selected implementation #20.

flowchart TD
  A[Cycle 4 closeout PR #30; gate #23 closed] --> B[Refresh public evidence]
  B --> C[Audit main 62e8b95]
  C --> D[Select #27; defer #26 and #28]
  D --> E[Correct ControllerHub docs and test logical ID lifecycle]
  E --> F[Independent review and required checks]
  F --> G[Merge implementation PR stack]
  G --> H[Roadmap closeout and Cycle 6 gate]
  H --> I[Close Cycle 5 gate #29]

Checklist

• Refresh dated reports across AI/game-agent forums, HN, OpenAI Developer Community, Steam, GitHub projects, and related controller workflows; record Reddit, Google, SDL Discourse, Steam Community, and accessibility coverage limits.
• Record source dates, workflows, stated workarounds or absence, evidence strength, and limits; count the Selkies/ROMM cross-post once and make no prevalence claim.
• Audit current main, ControllerHub docs/code/tests, browser exports, native identity, and latency measurement boundaries.
• Rank candidates by user value, evidence strength, project fit, cost, and risk; select Fix ControllerHub docs and test logical ID lifecycle #27 and defer Verify controller routing with a user-run browser hardware check #26/Measure input latency at explicit SDK boundaries #28.
• Define the selected scope, acceptance, non-goals, test plan, and dependency before implementation.
• Delegate the decision record and implementation to gpt-6-luna agents in isolated worktrees; fetch at integration points. PR docs: record Cycle 5 research decision #31 records the research decision; PR fix(core): clarify controller hub logical ID lifecycle #32 implements the selected issue.
• Implement Fix ControllerHub docs and test logical ID lifecycle #27 in PR fix(core): clarify controller hub logical ID lifecycle #32; independent gpt-6-luna review found no blocker after fixes, GitHub CI passed, and the builder reports core tests (62/62), typecheck, runnable typed example, and full release:check passing. Issue Fix ControllerHub docs and test logical ID lifecycle #27 closed on merge at e6d5769.
• Merge the Cycle 5 roadmap closeout and close Cycle 5 research gate: select the next evidence-backed improvement #29. Cycle 6 gate Cycle 6 research gate: select and ship one bounded reliability improvement #35 and selected issue Preserve cleanup and failure context in native test #20 are open behind Cycle 5 research gate: select the next evidence-backed improvement #29; Cycle 6 research gate: select and ship one bounded reliability improvement #35 also depends on Preserve cleanup and failure context in native test #20.

Acceptance

• Current-main source evidence and code audit are linked and accurately scoped.
• One bounded issue is selected; deferred issues remain outside active work.
• Issue Fix ControllerHub docs and test logical ID lifecycle #27 tests and corrects the real ControllerHub API contract without claiming hardware identity or host behavior.
• The implementation is independently reviewed, required checks pass, and PRs docs: record Cycle 5 research decision #31/fix(core): clarify controller hub logical ID lifecycle #32 follow the dependency graph.
• The roadmap and next gate exist before Cycle 5 closes.