List view

• Foreman: trustworthy local-first harness

  Make the Foreman agentic-coding harness trustworthy enough to run all dev on local hardware: gate trustworthiness (envtest/e2e/scope guards), single-pass convergence reliability, coder-image/toolchain correctness, and per-model harness profiles. Theme 2 of the initiatives backlog.

  No due date

  •7/10 issues closed
  70% complete3 open7 closed

• Backlog (Q1 2027+)

  Deferred: real value, wrong moment, or blocked on infra that does not exist yet.

  No due date

  •0/5 issues closed
  0% complete5 open0 closed

• M4: Foundation + depth

  Always-on tail: testing/coverage, oMLX + turboquant depth, the mlx-server-vs-Ollama benchmark.

  No due date

  •4/12 issues closed
  33% complete8 open4 closed

• M3: Governed + heterogeneous breadth

  Make governance real (ModelRouter enforces budgets/SLO/audit, not theater) and extend runtime/accelerator reach (per-CR metal runtime, multi-model edge).

  Due by December 31, 2026

  •7/14 issues closed
  50% complete7 open7 closed

• M2: Sovereign enterprise (B200-ready)

  The enterprise production gate: multi-cluster federation, B200 validation, air-gapped model sources, SLO declaration (Pyrra), GPUQuota, supply-chain provenance, and multi-vendor GPU scheduling.

  Due by August 31, 2026

  •22/37 issues closed
  59% complete15 open22 closed