Imagine a parallel world: your experiment is still running, the deadline for the first draft of your paper arrives, and you decide to write the Results chapter first.
The first ending: the numbers aren’t out yet, so you plug in "expected values" and plan to replace them when the experiment ends. Later, the experiment cycle is extended, you switch to two sets of parameters, and you forget to update the old numbers in the Results chapter—they end up in the submitted version.
The second ending: you ask AI to "help generate a plausible-looking results analysis." The text is smooth, but there’s no data to support it, it’s full of empty strong assertions, and once you run an AI-flavor removal report, this entire chapter lights up red.
The third ending: you’re the most diligent and write every analysis you can think of. On the day the real experimental data finally comes back, you find that the actual results don’t match the structure you inferred—and two chapters have to be torn down and rewritten.
Three parallel worlds point to the same root cause of the error:The Results chapter was placed at a time that wasn’t its own.
Correct timing: data_needed guards the gate for you
In the writing blueprint of muchsci Paper Sailor, there is a step like this: when generating the outline, mark chapter by chapter which chaptersdo not depend on data—literature review and theoretical framework can be written first; and which chaptersmust have real results—mark them data_needed and wait for the experiment.
After marking, your writing order automatically becomes sensible: write as much as possible of what can be written now; for chapters waiting for data, keep the structure and leave placeholders—and placeholders must never be automatically filled with numbers, a rule hard-coded into the muchsci product. When the data is in place, just fill it back in; no rework.
Gap action list: the basis for experiment scheduling
data_needed markers gathered together are an action list: Chapter 4 needs the results of two A/B control experiments, status "waiting for experiment"; Chapter 3 has sufficient literature support and is complete; Chapter 5 needs questionnaire collection and has been scheduled.
The value of this list lies beyond writing—it gives experiment scheduling a basis: which data are on the critical path, and which can be collected in parallel as supplements, are clear at a glance. Exposing gaps first is always better than exposing them the night before submission.
In the real world, there’s no parallel universe for trial and error, but you can let muchsci’s mechanisms help you avoid all the bad endings. Make each chapter’s data dependencies crystal clear, and the two tracks of writing and experimentation can run in parallel without conflict—this is the most cost-effective kind of order in paper writing.