AI and gaming improvement cycle 2: evidence-backed workflow improvements #5

Goal

Continue the evidence-led improvement cycle for AI and gaming workflows. This cycle should end with reviewed changes merged to main, an updated opportunity backlog, and a concrete next-cycle goal using the same workflow.

Status: complete. Research gate #7 and roadmap PR #11 are merged. Issues #8 and #9 closed after their independent implementations merged in PR #12 (067fc591fe34e7d09cfdfedcf8db48fcd6a7c0d1) and PR #13 (5f7e6fc7054654f220a6cb45a4c438ae0d9e4ae7). The combined main release check passed: 140 files linted, typecheck passed, 162 tests/0 failures/1,170 assertions, 9 build tasks, audit of 90 packages with no vulnerabilities, and packing of 7 packages. Roadmap closeout PR #17 is merged. Cycle 3 research gate #16 is open with research-gated candidates #14 and #15. GitHub Project V2 creation was blocked by missing Projects write permission on the connected integration, so Cycle 2 was tracked in milestone 1.

Cycle 2 selection (research gate #7; roadmap PR #11): the selected independent build slices are #8, which preserves command intent and structured diagnostics in replay errors, and #9, which verifies OpenController profile-to-report conformance. They improve behavior OpenController controls and do not claim to fix upstream FLE, SDL, or Steam failures. Exploratory physical input capture for OBS (#10) is deferred pending more firsthand evidence and macOS API/permission feasibility.

Existing evidence includes SDL's 8BitDo analog-trigger report, Betop motion-data report, GameSir calibration report, and Steam Deck discussions about missing rumble and controller ordering. These are device- and host-specific reports; research must distinguish upstream problems from issues this SDK can address.

Cycle DAG

flowchart TD
  A[Refresh public AI and gaming research] --> B[Log observations, workarounds, and evidence limits]
  B --> C[Rank opportunities and select a project-controllable painpoint]
  C --> D[Define scope, acceptance criteria, and task DAG]
  D --> E[Build in owned worktrees and commit coherent slices]
  E --> F[Independent review and integrated release checks]
  F --> G[Create stacked PRs with explicit bases]
  G --> H[Merge in dependency order after checks pass]
  H --> I[Update backlog and open the next cycle goal]

The graph above describes one cycle. After step I, start a new issue with the same stages; do not draw a back-edge that would make the DAG cyclic.

Checklist

Research and selection

• Refresh sources across Reddit (if reachable), Steam Community, relevant GitHub issues/discussions, AI developer forums, Hacker News, and other high-signal venues.
• Record source URL/date, user situation, reported pain, workaround, recurrence evidence, confidence, and coverage gaps. Separate direct evidence from interpretation.
• Reassess controller compatibility, lifecycle/reconnect recovery, accessibility/remapping, and any stronger new signals against value, evidence, project fit, cost, and risk.
• Select one bounded improvement OpenController can control, or defer this hypothesis and choose the stronger candidate.

Build and review

• Define a task DAG with dependencies, measurable acceptance criteria, and explicit non-goals.
• Use gpt-6-luna agents and isolated worktrees for independent research/build/review slices; assign file ownership and acceptance criteria to prevent conflicts.
• Fetch origin before work, at integration points, and before final validation. Rebase promptly when a parent branch moves.
• Build the selected slice, update user documentation, and keep this checklist current.
• Validate CI fixtures separately from real OS/game/hardware claims; include hardware or manual checks only when they can actually be performed and recorded.
• Have an independent reviewer inspect the combined changes for correctness, security, regressions, and fit to the source evidence.
• Run focused checks and the integrated release checks; report exact pass/fail/pending results.

Stacked PRs and merge

• Split by meaningful ownership and dependency; use coherent commits without artificial fragmentation.
• Open PRs against their real parent branches and include the user problem, evidence, acceptance criteria, and validation.
• After each parent merges, rebase and retarget child PRs to the updated main; rerun checks on the final stack.
• Merge PRs in dependency order only after required review and checks pass.
• Record merged commits, deferred work, source coverage limits, and the next ranked opportunity.
• Open the next cycle issue from that updated backlog so the process continues.

Acceptance

• The selected work is traceable to dated public evidence and to OpenController-controlled behavior.
• The implementation has a testable, bounded acceptance contract; no source sample is presented as prevalence.
• The reviewed PR stack is merged to main, and integrated release checks pass.
• This issue closes only after the next cycle goal has been opened from the refreshed backlog.