​

Factory Supervisor — Open Decisions

Decisions that block forward progress. Closed either via CODEOWNER ✅ reaction on the linked issue OR a merged PR updating this file. Canonical reference: docs/supervisor/ARCHITECTURE.md §12

​

Currently open

​

How to resolve

React ✅ on the issue → a followup PR updates this file to move the row to the “Resolved” section below AND replaces the “Recommendation” column with the outcome. OR open a PR updating this file directly, linking to the discussion in the PR body.

​

Resolved decisions (archive)

---

​

Partial reversal: D1-Grok-opt-in (2026-05-02 PM)

Context: Original D1 resolution (2026-05-02 AM) dropped Grok entirely in favor of Gemini 2.5 Pro (long-context) + Groq Llama (verifier). Adrian reversed partially after shipping llm@0.3.0. New resolution: Grok stays available in @latimer-woods-tech/llm as an explicit opt-in provider. Callers invoke it via { model: 'grok-4-fast' } or { model: 'grok-3-mini-latest' } — it is not in the default tier routing (fast / balanced / smart / verifier remain Anthropic + Gemini + Groq). Rationale: xico-city’s artist-platform user economy may want cheap experimental prompting where Grok’s quirks are a feature rather than a bug. Keeping it callable but off the hot path preserves the clean routing for production workloads. Implementation: llm@0.3.1 (factory PR forthcoming). GROK_API_KEY becomes optional in LLMEnv; only required when caller opts in. Non-effects: Groq (verifier) stays. Gemini (long-context fallback) stays. Anthropic (primary) stays. No change to steady-state cost model.