Autoresearch Loop
Seven skills span two repos. None of them chain. What breaks when the conductor is missing?
Pipe Dreams— Standard interfaces make compound loops buildable
Commissioning Protocol— L0-L4 maturity gates that the loop advances
Spec Contract— The engineering boundaries
Scorecard
| Dimension | Score | Evidence |
|---|---|---|
| Pain | 4/5 | 7 skills exist across 2 repos but run in isolation. Every step requires human trigger. Every North Star in priorities/index reads "Queryable: No". |
| Demand | 4/5 | All 7 active PRDs need queryable metrics. Karpathy's autoresearch validated the pattern at scale. The factory is blocked on activation, not new builds. |
| Edge | 5/5 | The skill chain already exists — pain-signal-extractor through proof-to-story. Others build from scratch. We wire what's built. |
| Trend | 5/5 | Autoresearch, Devin, overnight agents — the industry moves toward autonomous loops. Manual triggers are the gap. |
| Conversion | 3/5 | Infrastructure play. Revenue path is indirect: the loop produces proven-demand stories for marketing. Meta-capability, not product feature. |
| Composite | 1200 | Prepare |
Context
- Prompt Deck — The 5 Slide pitch
- Autoresearch Pattern — Karpathy's overnight loop
- Pain-to-Proof Loop — The commissioning pipeline this activates
- Agent Platform — PLAT-004 (Autonomous loop control) lives here
Questions
What compounds faster — building new skills or wiring existing ones into a loop?
- If every overnight session produced a structured experiment log, what would you learn by morning?
- Which dormant algorithm in libs/agency/ would benefit most from 100 automated iterations?
- When the loop runs, who edits the program.md — and what does that make you?