← Lab Handbook

Starting a Research Project

Starting a new project means putting structure around uncertainty. We separate the one-time initialization from the ongoing phases of research. Companion pages: Lab Expectations and Communication Rules.

Phase 1: Initialization

Before writing substantial code or math, set up the project infrastructure. This takes an afternoon and pays for itself many times over.

  1. Create the repository. Stamp a new repo from the lab’s project template using GitHub’s “Use this template” button rather than starting from an empty folder. It ships with the standard layout for R/Python code, simulation studies, and writing — see Lab Infrastructure for what goes where.
  2. Create the Overleaf project. Establish the LaTeX document that will hold the mathematical formulations and causal estimands from day one. Math scattered across notebooks and whiteboard photos does not survive contact with a manuscript deadline.
  3. Stamp the landing page. Every project has a central working document that serves as the single source of truth. Make a copy of the landing-page template and link it from the repository README.

The Project Landing Page

  • Top-level goal: one crisp sentence.
  • DRI (Directly Responsible Individual): lead PhD student or postdoc — the person who owns and drives the project; Lab Expectations describes the role.
  • PI / high-level strategy: Walter Dempsey.
  • Primary communication: the project-specific Slack channel (e.g., #project-analysis).
  • Code repository: link to GitHub.
  • Methodology working doc: link to the Overleaf project holding formulations and estimand definitions.
  • Grant alignment: link to the relevant planning docs (e.g., TrainMHealth T32, R01 PAR-25-144).

Roadmap & milestones (probabilistic)

Research is unpredictable, so we frame milestones as probabilistic targets rather than delivery dates — a framing taken from Jacob Steinhardt’s Research as a Stochastic Decision Process.

  • Milestone 1: [deliverable] — target: YYYY-MM-DD (X% confidence)
  • Milestone 2: [deliverable] — target: YYYY-MM-DD (X% confidence)

Top open questions & risks

A ranked list of the biggest uncertainties that need to be de-risked — resolved cheaply and early, before they can quietly sink the project (Phase 2 below). This list, more than the milestones, determines what we work on next.

Phase 2: The research loop

  1. Exploration & de-risking. Attack the highest-risk uncertainties first — the steps most likely to kill the idea (e.g., identifying the boundaries where a sequential decision-making model breaks down). They are the cheapest place to learn you are wrong.
  2. Execution. Cycle through observe–orient–decide–act: push code, review simulations, refine the approach, repeat. Progress shows up in your weekly updates as things learned, not hours spent.
  3. Write-up & polish. Translate the working documents into a publication-ready manuscript. If the landing page and Overleaf doc were maintained throughout, this phase is translation, not archaeology.

© 2026