Cycle 3 research gate: rank evidence-backed AI and gaming opportunities #16

Goal

Run the next evidence-led AI and gaming improvement cycle for OpenController. End with a reviewed, bounded change merged to main (or a documented no-build decision if evidence does not justify implementation), an updated opportunity record, and a concrete next-cycle goal. Issue creation is a research backlog, not build approval. The gate selected #14 for implementation, merged in PR #19 at f2baa6b. PR #24 records roadmap closeout and closes this issue after Cycle 4 gate #23 existed. Candidate #15 remains deferred outside the active milestone.

Starting state and duplicate check

Reviewed the roadmap docs/roadmap/ai-gaming-opportunities.md and issues #5, #7, #8, #9, and #10. Cycle 2 implementation PRs #12 and #13 have merged; the updated status and merge commits are recorded below.

Cycle 2 issues #8 and #9 were implemented and merged as PR #12 and PR #13. Issue #10 remains deferred. Cycle 3 must avoid duplicating the merged slices and rank candidates against current code.

Initial evidence log (2026-10-04 review)

These are dated public reports and a targeted sample, not a survey or prevalence estimate. Separate reports within a single project/ecosystem are not independent replications. Engagement counts and repeated examples do not estimate the number of affected users.

Coverage limits

• Reddit search and DuckDuckGo returned HTTP 403 on 2026-10-04; no Reddit posts were reviewed. Steam Community search returned a generic page without a useful verifiable dated result.
• An accessibility-focused GitHub issue was reviewed, but it concerns another product. An assistive-gaming article returned HTTP 403; only a Hacker News comment was accessible. No accessibility evidence specific to Agent Fighter was found.
• The examples above are self-reports and feature requests. No thread count is a prevalence estimate. External pain signals remain limited; Keep key SDK documentation examples runnable #14 was selected because a repository-owned documentation-coverage gap was verified, not because community prevalence was established.
• Issue Explore host-side physical gamepad input for OBS overlays #10's macOS physical-input idea and the Agent Fighter regression baseline work remain deferred from Cycle 2. Re-open them only if refreshed evidence and repository feasibility materially improve.

Initial candidate backlog and provisional ranking before the gate

The initial backlog was intentionally small because most reports concerned upstream behavior, overlapped Cycle 2, or lacked repository fit. The gate selected #14 as a preventive, repository-owned CI improvement and deferred #15; it did not treat external reports as prevalence evidence.

Cycle 3 gate decision (2026-10-04)

The repository audit inspected main at ba0529e, onboarding docs, example entry points, root scripts, and CI:

• The Getting Started dry-run flow and AI Agent Integration action-map/state-patch snippets are not executed by tests.
• Root tests cover package suites, not docs snippets or example packages. examples/basic-dry-run/index.ts is runnable but has no test or build script.
• Agent Fighter has a Playwright headless match runner and quality gates, but those do not execute SDK docs or validate keyboard remapping. Player two already has arrow-key movement; R resets and F toggles fullscreen.
• The gap is lack of executable documentation coverage, not a verified stale OpenController snippet.

Selection: #14, a preventive CI slice. Repository ownership and dry-run validation are direct. FLE #418 is an adjacent concrete docs failure, not evidence that OpenController examples are defective. Keep tested sources shared with docs to prevent fixture drift. Cover the standalone Getting Started dry-run path, one AI action-map/state-patch path, and the basic dry-run example; assert observable state/action or replay behavior; require no key, device, permission, or live game. Exclude live native, OBS, game, and hardware claims. Estimated effort: low-to-medium, 1–2 days.

Defer #15. Accessible-Chess issue #5 supports app-owned remapping as a broader accessibility concern but is from another product and its comments include owner status updates. The HN remapping anecdote is indirect. Neither identifies an Agent Fighter problem; player two already has arrow movement. A configurable UI also requires persistence, conflict handling, recovery, and browser testing (estimated 2–4 days), while browser/OS shortcut detection is incomplete. Revisit only with stronger fit or direct evidence.

Cycle DAG and checklist

flowchart TD
  A[Refresh public-source evidence] --> B[Audit current OpenController behavior and ownership]
  B --> C[Rank opportunities and record limits]
  C --> D[Select, revise, or defer candidates #14/#15]
  D --> E[Define bounded build/review task DAG and acceptance criteria]
  E --> F[Implement selected work in coherent reviewable commits]
  F --> G[Independent review and focused/integrated checks]
  G --> H[Open stacked PRs against declared parent branches]
  H --> I[Merge in dependency order after review and checks]
  I --> J[Update roadmap/backlog and open the next cycle goal]

Research and rank

• Refresh sources in relevant public AI/gaming communities: GitHub issues/discussions (including SDL, Steam, game-agent projects), Steam Community, developer forums, Hacker News, Reddit if reachable, and a relevant accessibility source if discoverable.
• For every cited item, record direct URL, post date, user/workflow, reported failure or request, stated workaround/current alternative, recurrence evidence, and confidence. Mark interpretation separately; do not infer prevalence from a few reports, replies, reactions, or search rank.
• Record failed searches/access limits and distinguish upstream host/game bugs from behavior OpenController controls.
• Audit current code, docs, merged PRs, and open Cycle 2 PRs before ranking; update each candidate's fit, cost, risk, dependencies, acceptance scope, and deferred/selected status.
• Select only a bounded issue with direct-enough evidence and repository ownership, or document a no-build decision. Do not begin candidate implementation until this gate is complete.

Issue DAG and implementation

• Use native GitHub blocked-by links to represent the current dependency. After selecting Keep key SDK documentation examples runnable #14, remove its candidate-to-gate blocker, make this cycle goal wait on the selected implementation, and remove deferred Make Agent Fighter keyboard bindings configurable #15 from the active milestone.
• Define the selected Keep key SDK documentation examples runnable #14 build scope, shared-source requirement, acceptance criteria, and non-goals before implementation starts.
• Make one coherent implementation change at a time; PR test: run documented dry-run examples in CI #19 documents the dry-run boundary and distinguishes fixtures from host/game/hardware claims.
• Run focused checks and the integrated release checks. Record exact outcomes and leave unsupported host/hardware claims pending in the roadmap.

Independent review, stacked PRs, and merge

• Obtain an independent review for scope fit, correctness, regressions, security, and consistency with the source evidence; resolve the replay-directory and TypeScript-project findings before merge.
• Open PR test: run documented dry-run examples in CI #19 against main with acceptance criteria and validation results in the description.
• PR test: run documented dry-run examples in CI #19 targeted main after the research PR merged; there were no child implementation PRs, and CI passed on the final head.
• Merge PR test: run documented dry-run examples in CI #19 after independent review and required checks passed; there were no child PR dependencies.

Close and next cycle

• Record merge commit f2baa6b, deferred Make Agent Fighter keyboard bindings configurable #15, updated coverage limits, and Cycle 4 candidates in the roadmap via PR docs: close Cycle 3 and open Cycle 4 gate #24.
• Open Cycle 4 goal Cycle 4 research gate: rank evidence-backed AI and gaming opportunities #23 from the refreshed backlog; PR docs: close Cycle 3 and open Cycle 4 gate #24 records the closeout and closes this issue.

Acceptance

• Sources and coverage limits are dated, directly linked, and distinguish observed claims from interpretation.
• The final rank is tied to current repository behavior and names upstream boundaries; there are no unsupported prevalence claims.
• Any selected build issue is scoped, dependency-linked, measurable, independently reviewed, and merged only after its required checks pass; otherwise the no-build decision is explicit.
• A next-cycle goal is opened from the updated backlog.