A researcher says your README is a context bill

Reproducible research has always been a hard sell. Clean commits, decision records and repository structure cost time up front, and the benefit falls to some other researcher, later, maybe never. Lorena Barba's new position paper flips that math for anyone running coding agents against their code.

Her argument is that tests, commit histories, repo structure, instructions and decision records aren't separate from reproducibility practice. They're the context a coding agent reads before it can do anything useful. Miss that context, or get it wrong, and the agent burns tokens rebuilding it from scratch, or guesses. Get it right, and the agent uses it in the same session, per Barba's position paper.

That reverses the usual pitch. Reproducibility work has always asked researchers to pay up front so someone else benefits down the line. Barba's version puts the payoff in the same session as the work, for whoever is running the agent.

She doesn't let the researcher off the hook, though. Agents can maintain tests, docs and changelogs more cheaply than a person typing them by hand, but verifying what those artifacts claim, and standing behind the judgments they encode, stays the researcher's job. An agent can write a decision record. It can't decide whether the decision was right.

If you've treated documentation and test coverage as a chore that pays off eventually, it costs you on every agent session against an undocumented repo. The README you never wrote used to be a someday problem. It's a per-session tax instead.