Wikipedia Publisher

Use a conservative workflow. Prefer publishable drafts over impressive drafts.

Default workflow

1. Collect evidence first

   • Gather independent secondary coverage before writing.
   • Separate independent sources from primary/company sources.
   • Do not treat the subject's own site as proof of notability.
   • Run scripts/citation_fetch_enricher.py first when URLs are bare, weak, or missing metadata.
   • Run scripts/source_hygiene.py and scripts/notability_score.py early.

2. Decide the safest target

   • Prefer User: sandbox or Draft: namespace for new articles.
   • Use mainspace only when the article is already established or the user explicitly wants that path and the sourcing is strong.
   • For conflict-of-interest situations, prefer sandbox + disclosure-minded language.

3. Write plain, non-promotional prose

   • State what the subject is, why it received coverage, and what reliable sources say.
   • Attribute claims when needed: "According to X..." or "Coverage in Y described..."
   • Avoid mission-language, puffery, rankings, and unverified superlatives.
   • If a paragraph smells like PR, run scripts/coi_rewrite.py and then edit manually.

4. Run a risk pass before publishing

   • Check tone with scripts/coi_tone_lint.py.
   • Check citation/source mix with scripts/source_hygiene.py.
   • Check likely rejection risk with scripts/deletion_risk_check.py.
   • Use scripts/citation_normalizer.py to clean weak references.
   • Remove unsupported claims, crowded lists, and anything sourced only to the subject.
   • Make sure the lead can stand on independent sources alone.

5. Publish conservatively

   • Save to sandbox/draft first unless there is a good reason not to.
   • Use scripts/sandbox_publish.py --dry-run before live save.
   • Use a factual edit summary.
   • Verify the live page after save.

Writing rules

• Prefer independent news/features/books over press releases, directories, or the company site.
• Keep claims proportional to source strength.
• Use primary sources for routine facts only: founding date, location, official name, contact details, basic product list.
• Avoid these unless clearly and independently sourced:
  • "leading", "world-class", "premier", "innovative", "trusted"
  • "official" except when it is structurally true and relevant
  • unsourced rankings, market leadership, customer counts, geographic scale, or awards
• Do not pad new articles with trivia or exhaustive project catalogs.
• When in doubt, shorten.

Article shape for company/org drafts

Use this default structure unless the sourcing suggests otherwise:

• Lead
• History
• Operations or Products/Services
• Reception / Coverage / Major projects (only if independently notable)
• See also

Skip sections that cannot be supported cleanly.

Sandbox-first publishing pattern

When the user wants a new page:

1. Draft locally in .wiki or .md.
2. Convert to clean MediaWiki markup if needed.
3. Save to User:<name>/<topic> or Draft:<topic>.
4. Fetch the saved page and inspect the rendered result.
5. Only then discuss moving toward mainspace.

Wikidata pattern

For Wikidata work:

• confirm an item does not already exist
• collect minimally sufficient identifiers and statements
• prefer low-controversy statements first: instance of, country, official website, inception, founders
• cite every claim you can
• avoid copying marketing copy into descriptions or aliases
• use scripts/wikidata_helper.py for a clean starter bundle

Red flags that should slow you down

Read references/red-flags.md when any of these appear:

• only primary or affiliated sources
• article reads like PR
• many claims use the same weak source repeatedly
• the user wants direct mainspace publication for their own company/client
• notability seems to rest on routine business coverage or local mentions

Helpful resources

• scripts/source_hygiene.py — classify citations as independent, primary, press-wire, directory, or review-needed
• scripts/notability_score.py — rough article-viability scoring from the citation mix
• scripts/coi_tone_lint.py — lightweight tone/source risk checker for drafts
• scripts/deletion_risk_check.py — heuristic rejection/deletion-risk checker
• scripts/citation_fetch_enricher.py — fetch live metadata from URLs and suggest richer cite templates
• scripts/citation_normalizer.py — suggest cleaner cite templates and missing metadata
• scripts/coi_rewrite.py — rewrite PR-ish text into more neutral attributed prose
• scripts/sandbox_publish.py — sandbox/draft publishing helper for MediaWiki API
• scripts/wikidata_helper.py — starter statement generator for Wikidata items
• references/red-flags.md — quick safety checklist for companies, people, and organizations
• references/workflow.md — recommended script order and interpretation
• references/citation-guidelines.md — citation cleanup guidance

Output preferences

When showing a proposed edit:

• start with the target page title
• show the lead first
• summarize major sourcing assumptions
• call out any unsupported or COI-sensitive areas plainly
• recommend sandbox/draft vs mainspace explicitly