solo-scientist

The Force-Multiplier Playbook

How One Scientist + One LLM Can Match a Research Team

A structured protocol turns the LLM from a chatbot into a force multiplier.

DOI: 10.5281/zenodo.20154578 License: CC BY 4.0

---

The Problem

Modern science rewards big teams. The ATLAS collaboration has 3,000+ scientists. The average biomedical paper lists 6.5 authors. The solo scientist — once the default mode of discovery — is at a structural disadvantage, not because they lack ideas, but because they lack throughput: literature review, code prototyping, equation derivation, drafting.

The Shift

LLMs crossed a threshold. They can now synthesize literature, derive and verify equations, generate and run code, and draft technical prose — all in a single conversation with file access and code execution.

The Claim

A single researcher, following a structured five-phase protocol with an LLM, can reproduce the output of a small research team. Our preliminary self-experiments suggest speedups of $25\times$ to $90\times$ across two domains (theoretical physics, computational linguistics).

The Stack (4 components — that’s it)

Component	Why
LLM Interface	The brain
File I/O	Persistent state across turns
Code Execution	All quantitative claims verified
Git	Audit trail, reproducibility

No Docker. No API keys. No multi-agent frameworks. A single conversation thread with file access and code execution is the entire architecture.

The 5-Phase Protocol

Phase	What Happens	Who Leads
1. Define	Frame the question, set success criteria	Human
2. Delegate	Issue structured prompts	Human → LLM
3. Execute & Iterate	LLM produces; human reviews; repeat	LLM + Human
4. Verify	Code checks, limit tests, reader test, human review	LLM + Human
5. Synthesize	Assemble final output	LLM

The 5 Prompts

Five reusable prompt patterns cover the full pipeline: Literature Synthesis, Derivation with Reality Check, Code Prototyping, Section Drafting, Verification Audit. → See release/prompts/

Try It Today

1. Pick a research question — something you’d normally budget a week for.
2. Open an LLM conversation with file access and code execution.
3. Follow the five phases (total: ~5 hours of focused time).
4. Measure your speedup. Compare against how long it would have taken alone.
5. Report back. → See CONTRIBUTING.md

Read the Full Whitepaper

→ PLAYBOOK.md — complete methodology, case studies, verification protocol, limitations, and call to action.

---

The bottleneck to scientific productivity could shift from team size to human creativity and LLM-fluency. The solo scientist is back.