The /factcheck Command — Verify Claims Before They Go Live

Publishing a factual error is not just embarrassing — it erodes trust and requires a correction. /factcheck is designed for one thing: finding claims in a draft that can be verified and checking each of them before the piece is published.

The command takes a file path as its argument and routes to the factcheck skill, which runs a structured verification pipeline. A timestamped backup of the original is saved before any changes are made.

Claim extraction

The first step is reading the document and building a table of every verifiable claim. Each claim is tagged by type — statistic, date, name, technical specification, scientific fact, legal statement, or attributed quote — and assigned an importance level. A claim at the center of the article's argument is marked as critical; background context is marked as medium or low. Importance determines how many independent sources are required: critical claims need three or more, medium claims need one.

This classification step happens before any web search. Its purpose is to allocate verification effort where it matters most — not to spend equal time on every sentence.

Verification method

Claims are processed in order of importance, highest first. For each claim, a specific search query is formulated — including names, dates, and product versions. Vague queries are not used. Sources are ranked by authority: official product pages and developer documentation rank above press releases, which rank above news outlets. Social media posts are not treated as authoritative sources for any claim type.

Each source is fetched and read, not just cited from a search snippet. For critical and high-importance claims, a second independent source is required. Each claim receives a verdict: ACCURATE, INACCURATE, OUTDATED, MISLEADING, UNVERIFIABLE, or NEEDS_CONTEXT.

What changes and what stays

The correction policy is narrow by design. Inaccurate claims are corrected to match what the source says. Outdated claims are updated. Claims that are misleading without additional context get a clarification added. Everything else — structure, tone, the author's word choices, the argument — is left alone. The output should be the same article, more accurate.

After corrections are applied, an editorial report lists every change by category, each with the source that drove it. Any correction that alters meaning is flagged for the author's attention before it is applied to the original file.

When to use it

/factcheck is the targeted, single-purpose version of the verification phase that also runs inside /dr-edit. Use it when a draft needs fact verification but not the full editorial pipeline — no AI pattern scanning, no structural review. It is also useful as a standalone check after an article has been through editing but before final publication.

For the full editorial pipeline including AI pattern removal and structural polish, see /dr-edit. For the companion command that removes AI writing patterns without touching facts, see /humanize. Read what Datarim is for context on where these commands sit in the broader system.