Add factual-reliability tell and broaden the dash rule

[karotkriss]

What this does

Adds one well-sourced AI tell, sharpens two existing ones, and broadens the dash rule. Each addition is backed by a real, citable source (listed at the bottom).

Factual reliability (new section 34)

Adds a "Factual Reliability" group with a new pattern, Fabricated and unverifiable citations. Invented but plausible references are among the most consequential documented failure modes of LLM writing, and the skill did not cover them. The pattern tells the editor to confirm a source exists and supports the claim, then keep it, or cut the claim.

Dashes as punctuation (section 14, broadened)

The rule now targets the dash look as punctuation rather than one glyph: em dash, en dash, and the spaced hyphen standing in for one. Swapping an em dash for a spaced hyphen relocates the tell instead of removing it. The section adds a rephrase ladder (period, semicolon, comma, parentheses, colon) and explicitly protects real word hyphens (well-known, names, URLs, identifiers).

Section 7 (AI vocabulary)

Replaces an unsourced "appears more in post-2023 text" assertion with grounding in the Kobak et al. excess-vocabulary study, which measured this register surging across 15 million PubMed abstracts.

Section 20 (chatbot artifacts)

Adds the leaked-assistant phrase family ("As an AI language model...", "as a large language model", "I'm sorry, but I cannot").

Consistency fix

The skill's own instructional prose still had three em dashes in it, which is a bad look for a skill about killing dashes. This may influence the AI as well: a skill that models the very pattern it is asking the model to remove can reinforce it. Those three are rephrased out. The only dash glyphs left in the file are intentional, either naming the character in the rule or sitting inside the deliberately bad "before" examples.

Also syncs the README pattern index from 33 to 34.

References

• Walters and Wilder, "Fabrication and errors in the bibliographic citations generated by ChatGPT," Scientific Reports (2023). https://www.nature.com/articles/s41598-023-41032-5
• Mata v. Avianca, Inc. (S.D.N.Y. 2023), the sanctioned ChatGPT fake-cases filing. https://en.wikipedia.org/wiki/Mata_v._Avianca,_Inc.
• Kobak et al., "Delving into LLM-assisted writing in biomedical publications through excess vocabulary," Science Advances (2025), arXiv:2406.07016. https://arxiv.org/abs/2406.07016
• Wikipedia, "Signs of AI writing" (WikiProject AI Cleanup). https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing
• Josh Bernoff, "The em dash is a bit of a jerk, replace it whenever possible." https://bernoff.com/blog/the-em-dash-is-a-bit-of-a-jerk-replace-it-whenever-possible
• Grammarly, "What is an em dash?" https://www.grammarly.com/blog/punctuation-capitalization/what-is-an-em-dash/
• Wordvice, "When to use a semicolon vs colon, comma, and dash." https://blog.wordvice.com/when-to-use-commas-colons-semicolons-and-dashes/

[karotkriss]

@blader I made some tweaks I think are useful based on some of my university research and private works.
Let me know if you have any questions or if anything doesn't align with you want for this project, I notice you don't have a contribution doc for guidelines so just checking with you first.