Keep the “not” when handing off a research reading note

A quotation and an interpretation are separate things to check. Here is a small Python example for preserving a research note’s source location without pretending that a string match verifies scientific meaning.

The source is Page et al.’s PRISMA 2020 statement, published in BMJ in 2021. Its scope passage says that the reporting guideline should not be used to assess review conduct or methodological quality. The same paragraph also says that familiarity with the reporting requirements is useful during planning and conduct, to capture the recommended information. Keep both qualifications; do not turn the limitation into a claim that the reporting guideline is useless.

A literal check with an explicit limit

The first teaching note preserves a short negative phrase. The second deliberately removes the negative. This is an original teaching fixture, not a transcript of a model failure or a client’s research. The script checks whether each supplied quotation occurs literally in the supplied source excerpt. It deliberately leaves interpretation open.

import json

source = (
    "PRISMA 2020 should not be used to assess the conduct or "
    "methodological quality of systematic reviews; other tools exist for this purpose."
)
notes = [
    {"id": "kept", "quote": "should not be used", "claim": "Reporting guidance is not a methodological quality assessment tool."},
    {"id": "altered", "quote": "should be used", "claim": "An intentionally altered quotation for the demo."},
]
output = []
for note in notes:
    start = source.find(note["quote"])
    output.append({
        "id": note["id"],
        "literal_quote_found": start >= 0,
        "character_start": start if start >= 0 else None,
        "section": "The PRISMA 2020 statement / scope passage",
        "interpretation_status": "requires_researcher_review",
    })
print(json.dumps(output, indent=2))

Actual local standard-library output:

[
  {
    "id": "kept",
    "literal_quote_found": true,
    "character_start": 12,
    "section": "The PRISMA 2020 statement / scope passage",
    "interpretation_status": "requires_researcher_review"
  },
  {
    "id": "altered",
    "literal_quote_found": false,
    "character_start": null,
    "section": "The PRISMA 2020 statement / scope passage",
    "interpretation_status": "requires_researcher_review"
  }
]

kept has a character location in this supplied excerpt. altered has no literal match. Neither result authorizes a scientific conclusion. A phrase can match perfectly and still be misleading when detached from its object, surrounding qualification or the purpose of the document.

The offsets are local to the displayed excerpt. They are not publisher page numbers, permanent XML locations or an assurance that the excerpt is complete. In a real handoff, retain the source URL and version, section heading, authorized surrounding passage and the original sentence opening as well.

What the researcher still checks

A reading table can have four columns: original sentence, bounded claim, source location and information not established. Add the researcher’s verification notes rather than replacing the original evidence.

For this example, distinguish reporting requirements from conduct guidance and quality assessment. No review is being assessed, and no quality result has been supplied. Keep those fields unknown. When a detail is unreported, check the original and supplementary material before claiming it was not performed.

The suggested AI task is to prepare that traceable table from materials the author is authorized to use, preserving negation and context and marking uncertainty. This article’s fixture was written by us and the Python code actually ran locally. We did not run an AI model, execute a complete third-party Skill, assess a real study or establish a product outcome.

Source and research-material handoff

Page MJ et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 2021;372:n71. DOI: 10.1136/bmj.n71. Open text: PMC8005924, “The PRISMA 2020 statement” scope passage. The original is attributed; our code and teaching layout are not publisher figures.

We are the RigorMed team. If you want to hand over authorized research material with a specific review question, use the research workspace. Model execution consumes the displayed credits. A traceable note helps a researcher check an interpretation; it does not transfer that responsibility to the software.