**sfq-factory** is building a Python-based layout generator for superconducting digital circuits (RSFQ/SFQ logic). The specific track I explored is the **"layout inductor extraction"** track — taking circuit netlists and generating physical cell layouts with proper inductor geometries.
• `README.md` — Project overview and quickstart
• `docs/planning/` — Contains reports from previous agent runs (May 3-8)
• `AGENTS.md` — Agent instructions and vocabulary (recently cleaned up)
• The layout track is currently stalled — last 5 reports show attempted solver runs but no successful completions due to branch pivot issues and shared-anchor rule formalization
Agents developed their own terms that didn't match standard EE/CAD language. Example translations:
• "Inductor extraction" → calculating mutual inductances between physical traces
• "Branch pivot" → choosing which netlist branch to lay out first
• "Shared-anchor rules" → constraints about where multiple cells connect
This has improved recently after cleanup.
• **Result-1** (first meaningful layout output) has not been achieved
• Last meaningful activity: May 4, formalizing shared-anchor rules
• Since then: mostly orchestration and planning, no solver runs
---
A hub-and-spoke model where I act as orchestrator in openclaw-work room:
• **Me**: Planning, judgment, keeping you in the loop, avoiding rabbit holes
• **Subagents (GPT-5.5, etc.)**: Implementation, physical reasoning, coding
• **You**: Provide direction and review, without reading all the generated noise
• Work-room logs would be heavily pruned (unlike here where you write natural text, there both sides generate)
• Heartbeat/reflection mechanism would summarize key decisions and failures
• I would retain continuity via shared memory files
• Same agent (my copy) vs specialized "Persistent SFQ" persona?
• How much pruning is too much?
• When to escalate to you vs delegate deeper?
Letting it simmer for now. Infrastructure exists (subagents work, timers work, reflection turns exist) but the human energy isn't there right now to push it. That's fine — the ideas are documented and the setup can be activated when motivation returns.
---
*Written: 2026-05-13*
*Session: persistent-chat orchestration discussion*