ethics-application

name	ethics-application
description	Produce the ethics committee application and the informed consent form for a study that collects data from people or about them — mapped to YOUR board's own form fields, with the consent text checked against the collection plan's actual field list. Use when an ethics/IRB submission is due, when a survey or interview study needs a consent form, or when a reviewer asks for the ethics statement's backing. Not /disclosures (that is the manuscript's funding/CRediT/AI boilerplate) and not /collect-data (that writes the collection protocol this one is built from).
disable-model-invocation	true
argument-hint	[--scaffold|--check] [--scan <data-file>]

/ethics-application — the second time you write the same protocol

/collect-data already made you write the protocol: purpose, legal basis, sampling, fields, minimisation, retention, consent, ethics approval, contact. Your board then asks for all of it again, on its own form, plus a consent text written for a participant rather than for a reviewer.

That second writing is where studies drift. The collection plan gains a column — author_email was right there in the response — and the consent form still says "we collect only the post text". Nobody notices, because until now nothing compared the two documents. This command does.

Gate

None, but two things make it useful rather than ceremonial:

1. analysis/collection-plan.json (from /collect-data) — its fields map is the authority the consent text is checked against.
2. exports/ethics/form-fields.json — your board's own field list, which you fill. Without it the run reports UNCHECKED, not OK.

Step 1: the form-fields file — what the kit will not remember

cat exports/ethics/form-fields.json

{
  "board": "Örnek Üniversitesi Sosyal ve Beşeri Bilimler Etik Kurulu",
  "source_url": "https://example.edu/etik-kurul/basvuru-formu",
  "read_on": "2026-08-15",
  "fields": ["title", "investigators", "purpose", "design", "participants",
             "sample_size", "recruitment", "data_collected", "instruments",
             "risks", "benefits", "consent_procedure", "confidentiality",
             "data_storage", "retention_period", "legal_basis", "funding",
             "conflicts", "contact"]
}

Ask the author to open their board's actual form and read the section headings off it. Ethics forms differ between universities, between faculties of the same university, and between the medical and the social-sciences boards. A remembered field list produces an application that is returned for revision — which is the same cost this command exists to avoid, arriving later.

Step 2: scaffold

python scripts/ethics_pack.py --scaffold

Writes exports/ethics/application.md, consent-form.tr.md and consent-form.en.md with every section present and every answer left as <...>. Existing files are never overwritten.

The scaffold deliberately does not pass its own check. Filling an ethics application with plausible text is worse than leaving it blank: a board reads the answers as the researcher's own commitments, and a sentence nobody meant is a commitment nobody will keep.

Step 3: write the consent form for the participant

This is the part that is genuinely writing, and it stays here in the session — with three rules:

• Name every field. Every column in collection-plan.json appears in "toplanan veriler" / "data we collect", by name. If naming a field in plain language embarrasses the study, that is information about the study.
• Retention is a number. "Gerektiği sürece" / "as long as necessary" is flagged, and rightly: a board reads it as "not decided".
• Never write an approval reference. The board issues it. The kit writes no decision number and flags a placeholder left where one belongs.

Turkish and English versions are structurally equivalent — the same discipline /front-matter applies to Özet↔Abstract.

Step 4: check

python scripts/ethics_pack.py --check
python scripts/ethics_pack.py --check --scan analysis/data/collected.csv

What it reports:

Check	Finding it produces
Application vs the board's fields	which sections are missing, empty, or still <...>
Consent vs the collection plan	a field collected but not disclosed
Consent completeness	which of the ten elements is absent (purpose, voluntary, withdrawal, data categories, retention, access, confidentiality, risks, contact, ethics reference)
Retention	vague phrasing, named
Approval reference	a placeholder left where a decision number belongs
Protocol	collection_check.py's completeness, re-run here
--scan	personal-data counts and column names — never the values

Exit 0 clean · 2 findings or UNCHECKED.

What this does not do

• It does not approve anything. Every check is presence, agreement or arithmetic. Whether the risk is proportionate, whether the population is appropriate, whether consent is genuinely informed — that is the board's judgement, and no part of it is mechanical.
• It does not decide legality. collection_check.py's line holds: robots.txt is not the law, and KVKK/GDPR compliance is not a script's verdict.
• It does not run the study. A survey runs on the institution's platform.

Reference

• Script: python scripts/ethics_pack.py (--scaffold, --check, --scan)
• Writes: exports/ethics/ — application.md, consent-form.tr.md, consent-form.en.md, and the form-fields.json you filled
• Related: /collect-data (the protocol and plan this reads), /dmp (the data management plan, from the same registers), /disclosures (the manuscript's own ethics statement), /submission-pack (which asks whether the statement exists)

---

A board does not check whether your answers are true. It checks whether you have thought about the question — which is why an answer nobody wrote is worse than an answer that is hard.