| name | research-method |
| description | The research loop for any field: sharpen the question until it can be refuted, find what is known and what is disputed, fix the analysis before the data, run it, interpret against effect size and uncertainty, write it so someone else can repeat it, and leave the numbers reproducible. Use for literature reviews, hypotheses, study design, preregistration, citations and anything called a paper, protocol or analysis plan.
|
| version | 1.0.0 |
| author | Remedy |
| tags | ["research","method","literature","citations","reproducibility","evidence"] |
| requires | [] |
| tools | ["lit_search","lit_fetch","cite_add","cite_import","cite_list","cite_export","cite_check","analysis_ledger","manuscript_check","skill_activate","file_read","file_write","file_edit","repo_search","web_fetch"] |
| triggers | ["\\b(research question|literature review|systematic review|scoping review|pre-?registration|prereg\\w*)\\b","\\b(?:null|alternative|working|research)\\s+hypothes[ei]s\\b|\\bfalsifiab\\w+\\b|\\bconfound(?:er|ing|s)\\b","\\b(reproducib\\w+|replicat(?:e|ion)\\s+(?:the\\s+)?(?:study|analysis|package|crisis)|p-?hack\\w*|HARK(?:ing)?|garden of forking paths)\\b","\\b(doi:\\s?10\\.\\d{4,9}/|arXiv:\\s?\\d{4}\\.\\d{4,5}|bibtex|\\.bib\\b|CSL[- ]?JSON|cite\\s+(?:the\\s+)?(?:paper|source|literature))\\b"] |
Research method (the spine)
This pack owns the loop, the evidence standards, citation honesty and what
"reproducible" means. Field packs add only what is different in their field —
activate one alongside this, never instead of it. Do not restate the spine
inside field work; apply it.
Read references/INDEX.md, then file_read the one or two that match the
stage you are at. The stage, not the whole shelf.
The loop
- Question. Rewrite it until it names a population, an exposure or
comparison, an outcome and a measurement. If no result could come back
and make it wrong, it is not a research question yet — say so and fix it
before anything else. →
references/question-and-hypothesis.md
- What is known.
lit_search (broad, then narrowed), lit_fetch for the
ones that matter. Record the search string, source, date and hit count —
a review whose search cannot be rerun is an opinion. Separate settled,
contested and one-lab findings; name the dispute rather than picking a
side. → literature-review.md
- Hypothesis. State H0, H1, the direction, and what result would refute
it. Write the refutation condition down now; it is unwritable later.
- Design. Units, assignment, controls, blinding, confounders, sample
size. The analysis is decided here, before the data exists — which test,
which covariates, which exclusions, how missing data is handled.
→
study-design.md, analysis-plan.md
- Preregister where the field expects it (trials, and increasingly
psychology, ecology, econ). A dated, read-only plan file in the repo is
the minimum; a registry entry when one governs the work.
→
preregistration.md
- Collect / obtain, recording provenance and version for every dataset.
- Analyse as planned. Anything not in the plan is exploratory and is
labelled exploratory in the text — not softened, labelled. Every number
comes out of a run in the ledger, never out of a chat message.
- Interpret against the effect size and its interval, in the outcome's
own units. A p-value alone is not a result; "not significant" is not
"no effect". →
interpreting-results.md
- Write up so a stranger could repeat it. →
writing-up.md
- Leave it reproducible. That is the verify step (below).
Evidence standards
Weigh a source by what was actually done, then say which tier you used:
