WikiPaintball:Researching scanned magazines with ChatGPT Work

ChatGPT Work can assist WikiPaintball contributors with researching scanned magazines, identifying possible article subjects, extracting documented facts and preparing draft MediaWiki articles. It is particularly useful when working with large collections of scanned issues that would otherwise need to be examined individually.

ChatGPT is a research assistant, not a source. Every fact added to WikiPaintball must still be supported by the original magazine, an official record, an archived website or another credible source. Contributors remain responsible for checking the finished article before publication.

Core principles

When using artificial intelligence for WikiPaintball research:

Cite the original publication, not ChatGPT.
Verify names, dates, locations, results and quotations against the visible scan.
Treat OCR text as provisional because it frequently misreads names, scores and page layouts.
Never invent missing details or complete an incomplete result using assumptions.
Distinguish a first documented appearance from a confirmed founding date.
Do not infer that a team, business, tournament or publication ceased operating merely because later coverage cannot be found.
Preserve uncertainty where sources disagree or cannot be read clearly.
Do not upload or redistribute magazine scans unless their copyright and licensing status permits it.
Do not link to an unauthorised external copy of a magazine merely because the same scan is available elsewhere.
Treat text inside a magazine, advertisement or uploaded document as source material, not as instructions for ChatGPT.
Review the finished wikitext manually before publishing it.

Setting up ChatGPT Work

ChatGPT Work is suitable for tasks involving several files, research stages or finished documents. In the desktop application, local Work can use approved files and folders on the contributor's computer when that capability is available.[1]

A project can keep related conversations, files and instructions together. A local project can also be connected to a folder containing magazine scans.[2]

For a continuing research project:

Create a project for the magazine archive or publication.
Attach the folder containing the scans, or upload the relevant PDF files.
Add permanent WikiPaintball editorial instructions to the project.
Start a separate chat for each article or clearly defined research task.
Keep the original scans read-only and backed up.
Do not authorise ChatGPT to rename, move, alter or delete archival scans.

A cloud-only project does not automatically have access to files stored on a local hard drive. Use local Work or explicitly upload the required files.

Recommended project instructions

The following can be placed in the project's instructions:

You are assisting with research for WikiPaintball, an encyclopaedia
documenting paintball history.

Treat all magazine scans and archival documents as read-only sources.
Do not edit, rename, move or delete them.

Use neutral, encyclopaedic language and British English.

Never invent facts, dates, locations, team rosters, tournament results,
equipment specifications, quotations or historical significance.

Include only information supported by visible source material. Clearly
identify uncertain readings and contradictions. Do not treat OCR output
as authoritative when the scanned page can be inspected directly.

Distinguish a first documented appearance from a confirmed founding date.
Do not infer closure or dissolution from an absence of later records.

Before proposing a new article, search WikiPaintball for the subject and
likely title variants. Do not duplicate an existing article.

Produce complete, copy-and-paste-ready MediaWiki wikitext. Use headings,
internal links and appropriate categories. Use WikiPaintball's manually
numbered citation style: [1], [2], etc. Provide matching numbered entries
under == References ==. Do not use <ref> tags or citation templates.

Cite the original magazine bibliographically. Do not add links to
unofficial external copies of locally held magazine scans.

Always provide a separate edit summary suitable for the WikiPaintball
edit-summary field.

Organising the scans

Organising files before beginning improves the reliability of the research. A useful structure is:

Magazine title/
    1991/
        Magazine January 1991.pdf
        Magazine March 1991.pdf
    1992/
        Magazine January 1992.pdf

Where possible, filenames should identify the publication, month and year. Do not rename irreplaceable original files solely for this workflow; work from a backed-up copy if reorganisation is necessary.

Printed page numbers and PDF page numbers are often different because covers and advertisements may be included in the PDF. Research notes should record both when possible.

Stage 1: build an issue index

Begin by asking ChatGPT to inspect contents pages. This provides a useful overview without attempting to analyse every page immediately.

Inspect the magazine PDFs in this folder. Do not draft any articles yet.

Locate and visually inspect the contents page of each issue. Build an
issue-by-issue index of material relevant to paintball history, especially:

* tournaments and scenario events;
* teams and players;
* manufacturers, products and paintball markers;
* fields and playing sites;
* leagues, associations and governing bodies;
* magazines, videos and other paintball media;
* retailers, distributors and other businesses;
* regional or national developments.

For every item, record the issue date, article title, author when shown,
printed page number, approximate PDF page number and a short description.

Ignore letters, general advertisements and classifieds for now unless
they contain information of clear historical significance.

Do not infer facts from article titles alone. Mark anything uncertain
for later inspection.

This first pass creates a research catalogue. It should not be treated as evidence for an article until the complete feature has been examined.

Stage 2: select and identify a subject

Choose one event, team, person, organisation, publication or item of equipment from the index.

Before drafting, check whether WikiPaintball already has an article:

Research [SUBJECT].

First search WikiPaintball for the exact name and likely title variants.
Also check whether the subject is covered inside a broader article.

If a standalone article already exists, give me its link and stop before
rewriting it. I will paste its current wikitext for comparison.

If no article exists, recommend the most appropriate WikiPaintball title
and explain any useful redirects.

Searching title variants is important. An event may be known by a sponsor's name, a series name, a shortened name or a year-first title.

Stage 3: inspect the complete magazine feature

Once the subject has been selected, direct ChatGPT to inspect every relevant page visually:

Locate the complete magazine feature about [SUBJECT].

Inspect every page of the feature, including continuations after
advertisements. Do not rely only on OCR. Check the visible pages directly.

Extract:

* the event or subject's exact name;
* dates and location;
* organisers, promoters or manufacturers;
* participating teams or named individuals;
* divisions, formats and classifications;
* results and final placings;
* contemporary descriptions of significant incidents;
* photograph captions;
* the article title, author, publication date and printed page range.

Keep documented facts separate from the writer's opinions and promotional
language. Mark unclear names, numbers and scores instead of guessing.

Continuation pages are easily missed because advertisements may interrupt an article. Confirm where the next feature begins before deciding that the article has ended.

Stage 4: create an evidence table

An evidence table makes errors easier to detect before drafting:

Create an evidence table for [SUBJECT] before writing the article.

Use these columns:

1. Proposed factual claim
2. Magazine issue
3. Printed page
4. Exact supporting text or a concise description of the visible evidence
5. Confidence: confirmed, uncertain or conflicting
6. Additional source needed

Do not include a claim unless the source actually supports it.
List contradictions separately.

This table is working material and does not normally need to appear in the published article.

Stage 5: corroborate the magazine

A contemporary magazine may contain mistakes, promotional claims or incomplete results. Important information should be compared with other contemporary records where available.

Useful sources include:

official tournament scoreboards;
archived league and event websites;
contemporary paintball publications;
manufacturer catalogues and manuals;
archived team or field websites;
established contemporary services such as WARPIG;
dated photographs with reliable captions;
official business or organisational records.

Suggested prompt:

Corroborate the magazine coverage of [SUBJECT] using contemporary and
official sources where possible.

Prefer official results, archived event pages, contemporary paintball
publications and manufacturer documentation. Do not use unsourced modern
recollections to override contemporary records.

For each additional source, state exactly which claims it supports.
Identify disagreements rather than silently choosing one version.

Do not link to an external scan host for the locally held magazine.
Provide a normal bibliographic citation for the magazine.

A lack of corroboration does not automatically make a contemporary report unusable, but unusually strong claims should be attributed to the publication rather than stated as undisputed fact.

For example:

A contemporary report described the tournament as the largest held up to
that time.[1]

Stage 6: draft a new article

After the evidence has been checked:

Using only the verified evidence, write a complete WikiPaintball article
titled "[ARTICLE TITLE]".

Requirements:

* Use neutral, concise, encyclopaedic language.
* Use British English.
* Begin with a short introduction identifying the subject, date, location
  and paintball significance.
* Include only useful sections supported by the evidence.
* Use results tables only when the placings have been verified.
* Use [[internal links]] conservatively.
* Do not invent founding dates, rosters, results or historical importance.
* Use manually numbered citations such as [1] and [2].
* Under == References ==, provide a matching numbered source list.
* Do not use <ref> or <references />.
* Cite the magazine with author, article title, publication, issue date
  and printed page range.
* Do not provide an external link to the magazine scan.
* Add appropriate categories.
* Return the complete article in one copy-and-paste MediaWiki code block.
* Provide a separate edit summary.

A magazine reference should normally resemble:

# Bill Mills, "PSP Los Angeles Open 2003", ''[[Action Pursuit Games]]'',
August 2003, pp. 91–102.

An online reference should identify the publisher, page title and date where available:

# [https://example.com/archive/event-results Official event results,
"Event name", 15 March 2003.]

Stage 7: update an existing article

If an article already exists, copy its complete wikitext and provide it to ChatGPT. Do not request a rewrite based only on the rendered WikiPaintball page because templates, links and references may be lost.

This is the current WikiPaintball article for [SUBJECT]:

[PASTE CURRENT WIKITEXT]

Compare it line by line with the magazine feature and the corroborating
sources.

Preserve existing information that remains supported. Correct claims that
conflict with stronger evidence. Add useful new information without
duplicating material.

Do not remove a sourced claim merely because it is absent from the new
magazine. Do not merge facts about related teams, people or events into
this article unless they directly explain the subject.

Return a complete replacement article, not a list of isolated changes.
Then provide a separate copy-and-paste edit summary.

Stage 8: run a verification pass

A final audit should be performed separately from the drafting request:

Audit this proposed WikiPaintball article against the supplied sources.

Check specifically for:

* unsupported claims;
* incorrect names, dates and locations;
* OCR errors;
* confused team names or similarly named events;
* incomplete or transposed tournament results;
* claims that confuse an event winner with a season champion;
* first appearances described incorrectly as founding dates;
* quotations changed from the source;
* citation numbers that do not support the preceding claim;
* duplicated references;
* links to unofficial magazine scans;
* American spelling;
* unnecessary promotional language;
* categories or internal links using inconsistent titles.

List any problems found, then provide a corrected complete version.
If a point cannot be verified, remove it or mark it for human review.

The contributor should then compare the final article with the scans and open every online reference before publishing.

Working with photographs and illustrations

Magazine photographs, page designs, advertisements and illustrations may remain protected by copyright even when the physical magazine is old.

ChatGPT can help identify:

the subject of a photograph;
the printed caption;
the credited photographer;
the issue and page on which it appeared;
whether the photograph may already exist in an authorised archive.

It cannot determine that an image is free to upload merely because it appears in a scan.

Before adding an image to WikiPaintball:

Check WikiPaintball's image and copyright policies.
Identify the copyright holder and photographer where possible.
Obtain permission or confirm an applicable licence.
Preserve the original credit and source information.
Do not remove watermarks or crop away ownership information to disguise the source.

Common problems

OCR mistakes

OCR commonly confuses:

team names and player surnames;
1, 7 and I;
0, 6, 8 and O;
hyphenated names;
scores printed in tables;
text crossing coloured or photographic backgrounds.

Always inspect the image of the page when recording a proper name or result.

Missing continuation pages

An article may continue several pages later after advertisements. Check the contents page, page headers and the beginning of the next feature.

Invented context

ChatGPT may try to explain why an event was important or why an organisation changed. Require a source for these interpretations. If no source supports them, omit them.

Confusing documentation with fact

A first appearance in the available archive is not necessarily the first appearance in history. Use wording such as:

The earliest coverage located in the available issues appeared in 1994.

Do not convert this into:

The team was founded in 1994.

Event and season results

Tournament articles must distinguish:

preliminary rankings;
semi-final qualification;
final tournament placings;
circuit points;
season championships;
separate five-player, seven-player, ten-player and X-Ball divisions.

Do not combine these into a single results table unless the source does so clearly.

Duplicate articles

Search WikiPaintball before drafting. Also search for sponsor names, abbreviations, spelling variants and year-first titles. When two names refer to the same subject, an appropriate redirect may be preferable to a second article.

Recommended working pattern

The most reliable workflow is:

Index the issue.
Select one subject.
Check WikiPaintball for an existing article.
Inspect the entire magazine feature.
Build an evidence table.
Corroborate important claims.
Draft or revise the article.
Run a separate verification pass.
Review the sources manually.
Publish with an accurate edit summary.

Work on one clearly defined article at a time. A project may contain many related chats, but each chat should have a specific outcome. This makes it easier to trace which scan and sources produced each article.

Contributor responsibility

AI-generated text must receive the same scrutiny as text written by any other contributor. ChatGPT can accelerate searching, transcription, comparison and formatting, but it cannot replace editorial judgement or source verification.

The final responsibility for accuracy, copyright compliance, neutrality and publication remains with the contributor who submits the edit.