Research Journal

Workflow insights, methodological learnings, and session notes.

2026-02-01

Compound engineering evaluation

Evaluated applying “compound engineering” principles (from EveryInc/compound-engineering-plugin) to the psalmody research project. Key insight: the elaborate learnings-capture system (YAML taxonomies, /capture and /compound skills, overnight processing pipelines) was over-engineered for a solo research project.

PhD adviser critique identified the core problem: Building systems instead of writing prose. 5 extracts processed, zero draft chapters. The compound engineering infrastructure would consume 15-20 hours that could produce 2,000+ words of actual manuscript.

Writing-first workflow

Shifted from “Extract → Organize → Systematize → Write” to “Argue → Write → Discover gaps → Extract what you need.”

Key changes:

• Chapter argument files force commitment to claims before gathering all evidence
• Writing reveals what evidence is actually missing (specific gaps, not abstract “process more sources”)
• Overnight agent reconceived as research assistant (prepares materials) not autonomous worker (makes interpretive decisions)
• Learnings captured directly to destination files, no intermediate staging

Minimal tooling principle

Instead of elaborate capture/compound skills, a single /reflect skill prompts at session end: “Did we learn anything that isn’t captured? If so, put it where it belongs.” No taxonomy, no processing pipeline—just a habit enforced by a prompt.