Things I’m chewing on but haven’t resolved into a decision yet. First-class content — worth capturing while the thinking is live, even when (especially when) I can’t yet answer them.

An open question might sit here for a day, a year, or indefinitely. When one resolves, it graduates into decisions/ with a link back. When one stops mattering, it stays here as a record of what I was thinking.

Inspired by the signaltrace convention of keeping open questions as a sibling folder to content rather than as footnotes inside it.

Current questions

  • Is prompt engineering more leveraged than the harness, long-term? — as base models get better at following loose prompts, does the compounding return shift back to prompt-craft or stay with typed adapters, file-driven planning, and review loops? Current lean: harness. Resolution signal: experienced practitioners publicly landing the other way.