preregister-paper

Preregister the paper

You write the paper before the experiments, from the hypothesis and experimental design alone. The commit that adds this paper is the preregistration record: every claim, criterion and expected result is frozen in git history before any result exists, so nothing can be quietly rewritten to fit the data later. Runs are expected to descend from this commit.

main.tex lives in .research/latex/{template}/, where {template} is one of the bundled template directories (iclr2024, mdpi, agents4science_2025) — the name the verification tools address the paper by. The directory is only a slot: use the bundled style or your own preamble, keeping the usual structure (title, abstract, numbered sections, figures, bibliography). Write in the user's working language; a CJK paper needs a LuaTeX preamble (luatexja-fontspec — pdflatex silently drops non-Latin text from the PDF). This skill only changes when the paper is written and how Results are stated.

Preconditions

• A local clone of the experiment repository.
• Hypothesis and experimental design exist — in .research/research_history.json, or supplied by the user. The design must fix the run ids (e.g. proposed, baseline) and the metrics; if it does not, settle those with the user first, because the placeholders below are named after them.

Steps

1. Create the canonical record with preregister_record. This writes .research/record.json — the machine-readable original the whole verification system keys on. The record is a tree, read as "to support this hypothesis, these claims; to verify this claim, these designs; a design is these runs":

   hypotheses: [{
     "id": "h1", "statement": "the hypothesis, in prose",
     "claims": [{
       "id": "c1", "statement": "one assertive sentence",
       "designs": [{
         "id": "d1", "summary": "...",
         "runs": [{"run_id": "proposed-...", "description": "...",
                   "params": {"mode": "full"}}]
       }]
     }],
     "tables": [...], "notes": [...]
   }]