preregister

name	preregister
description	Write and timestamp a preregistration before the data is analysed: hypotheses stated as directional predictions, the design, the sampling and stopping rule, the analysis plan with the exact tests, what counts as a confirmation and what would falsify it, and which analyses are exploratory. Attested through the hash-chained provenance ledger so the plan cannot be quietly revised after the results are seen. Use for empirical, RCT and observational studies, before /experiment.
disable-model-invocation	true
argument-hint	[--attest] [--osf]

/preregister — The Plan, Written Before the Results

A preregistration is not paperwork. It is the difference between a hypothesis and a description of what happened to be significant, and it is the one document that has to exist before the analysis, which is exactly why it is the one that gets written afterwards.

Gate

1. PROJECT.md complete (topic, RQs, type).

2. Check analysis/runs.jsonl. If runs already exist, say so plainly:

   Bu projede zaten N koşu kayıtlı. Sonuçları gördükten sonra yazılan bir plan ön kayıt değildir. İki dürüst seçenek var: (a) bunu analiz planı olarak yazıp ön kayıt demeyin, (b) yalnız bundan sonraki aşama için ön kayıt yapın ve neyin daha önce çalıştırıldığını açıkça yazın.

   Do not offer a third option, and do not backdate anything.

What goes in it

Write analysis/preregistration.md, section by section, with the author — each of these is a decision, not a field to fill:

# Preregistration — <title>
Date: <today>   ·   Status: before data analysis

## 1. Question and hypotheses
H1 (directional): <X increases Y> — not "X and Y are related". A non-directional
hypothesis cannot be wrong in an interesting way.

## 2. Design
<observational / experimental / RCT>, units, conditions, blinding, randomisation.

## 3. Sampling and stopping rule
Target N and how it was chosen (power analysis: effect size, alpha, power).
When collection stops — a date, an N, or a criterion decided NOW. "We will
collect until the effect is significant" is the failure this section prevents.

## 4. Variables
Outcome(s), predictor(s), covariates, and how each is measured/derived.
Exclusion rules for records, stated before seeing them.

## 5. Analysis plan
The exact test for each hypothesis, the model, how missing data is handled,
how multiple comparisons are corrected, and the software.

## 6. What would falsify H1
The result that would make you abandon it. If nothing would, H1 is not a
hypothesis.

## 7. Exploratory
Everything else you intend to look at, named as exploratory in advance — which
is what makes it publishable as exploratory rather than dressed up as confirmatory.

## 8. Deviations
(empty at registration; every later change is appended with its date and reason)

Dispatch methodology-advisor for sections 2-5 — power, design and the choice of test are its subject, and getting them wrong here is expensive later.

Attestation

python scripts/provenance.py --record preregistration --path analysis/preregistration.md

Appends a record carrying the SHA-256 of the file and the hash of the record before it. Rewriting it later breaks the chain, and provenance.py --verify names where.

State the limit, every time, because it is exactly the kind of claim that grows in the retelling: this proves the file was not rewritten afterwards, not that it was written before the data existed. The ledger has no key and no external timestamp; anyone with the repository can append. For a preregistration that a reader must be able to trust without trusting you, register it on OSF (osf.io/prereg) or AsPredicted, where the timestamp is a third party's — and put that registration's URL in section 0 of this file.

Deviations, later

When the analysis departs from the plan — and it usually does, legitimately — append to section 8 with the date and the reason, then re-attest. A deviation recorded is a methods sentence; a deviation not recorded is what a reviewer finds when the plan and the paper disagree.

Report

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
📋 Ön kayıt yazıldı
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Dosya:  analysis/preregistration.md (8 bölüm, hepsi dolu)
Zincir: provenance.jsonl kaydı #12 — SHA-256 kayıtlı
Sınır:  bu kayıt dosyanın SONRADAN DEĞİŞTİRİLMEDİĞİNİ gösterir;
        veriden ÖNCE yazıldığını değil. Üçüncü taraf zaman damgası için OSF.

Sonraki: /experiment (plan ile koşu kaydı birbirine bakar)

Reference

• Subagent: methodology-advisor (design, power, choice of test)
• Scripts: python scripts/provenance.py --record preregistration
• Writes: analysis/preregistration.md
• Related: /experiment (runs the plan), /replication-package, reporting-guidelines skill (CONSORT/STROBE items overlap with sections 2-5)

---

A plan written after the results is not a plan. The kit can prove a file was not edited; only a third party can prove when it was written.