Run a recurring evidence-backed AI and gaming improvement cycle #55

Goal

Keep improving OpenController through a repeatable, evidence-backed cycle: research real AI, gaming, and accessibility workflows; verify which problems belong to this SDK; express selected work as a dependency DAG; then build, review, and merge small, stacked PRs with coordinated agent lanes.

Do not manufacture feature issues to fill a fleet. A well-supported no-build decision is a valid cycle outcome, and it does not end the recurring goal.

Recurring DAG checklist

For each cycle:

• Research current reports across relevant AI and gaming communities, developer forums, platform ecosystems, and project issues; preserve source links and dates.
• Record who is affected, situation, recurrence, workaround/counterevidence, confidence, fit, and access/coverage gaps. Keep individual anecdotes distinct from trends.
• Audit each signal against OpenController's source, supported package boundaries, existing/deferred issues, and testability; rank user value, evidence, fit, cost, and risk.
• Open a research gate and create scoped implementation issues only for supported SDK-owned work. State acceptance checks, package/file ownership, and required dependencies.
• Draw native issue dependencies as a DAG. Start independent lanes in parallel; wait for prerequisites before releasing blocked work.
• Use gpt-6-luna research, implementation, integration, and exact-head review agents. Give each implementation lane isolated worktree/branch ownership and a clear deliverable.
• Fetch often, rebase on the declared parent, resolve conflicts promptly, and keep commits coherent. Use stacked PRs only where dependencies require them.
• Run relevant checks, resolve review blockers, and merge authorized PRs in DAG order after exact-head validation and independent review.
• Close the cycle with merged PRs/commits, checks, deferred work, source limitations, and the next research gate. Keep this goal open.

flowchart TD
  R[Research reports] --> A[Audit SDK ownership and issue overlap]
  A --> D{Supported, testable candidate?}
  D -->|Yes| I[Create scoped issues and dependency DAG]
  I --> W[Parallel Luna worktrees]
  W --> P[Stacked PRs as dependencies require]
  P --> V[Checks and exact-head review]
  V --> M[Merge in dependency order]
  D -->|No| N[Record no-build decision]
  M --> C[Close cycle and open next research gate]
  N --> C
  C --> R

Completed work

• Cycles 7 and 8 completed their selected builds and closeouts; see the roadmap and gates Cycle 7 research gate: expand the evidence-backed parallel build backlog #37, Cycle 8 research gate: build an evidence-backed controller improvement DAG #57.
• Cycle 9 repository findings Align native adapter capabilities with configured bridge outputs #33 and Reject negative controller action durations #34 were implemented in PRs fix: align native bridge capability metadata #69 and fix(core): reject negative action durations #70. Both merged on 2026-10-05; milestone 8 is closed. PR fix: align native bridge capability metadata #69 aligned bridge metadata with serialized output; PR fix(core): reject negative action durations #70 rejects negative durations before runtime dispatch.
• Cycle 10 research gate Cycle 10 research gate: audit current AI, gaming, and accessibility pain points #71 completed three public-source lanes and a separate repository audit. Sources, dates, workarounds, confidence, coverage gaps, and fit are recorded in Cycle 10 research gate: audit current AI, gaming, and accessibility pain points #71.
• Cycle 10 decision: no new implementation issue was supported. Current reports concern desktop input backends, upstream drivers/engines, or individual device ergonomics; repository findings are already fixed or tracked. Existing Wait for native helper readiness acknowledgement #46 and Add an opt-in expiry lease for persistent controller state #60 remain deferred until their prerequisites/evidence change.
• Merge roadmap closeout PR docs(roadmap): close cycle 9 and record cycle 10 decision #72 at fb70a63ea962fca5ae06e2ab2815a84a6e01bc63; close gate Cycle 10 research gate: audit current AI, gaming, and accessibility pain points #71 and milestone 9.
• Start Cycle 11 with a fresh evidence sweep; create implementation issues only if it finds supported SDK-owned work.

Project tracking

The request to create a GitHub Project V2 and move this goal into it is blocked: the authenticated integration lacks Projects write permission (createProjectV2 returns Resource not accessible by integration). Track the active DAG through repository milestones, labels, and GitHub issue dependencies until that permission is available. Do not treat the issue as already added to a Project.