---
name: paper-trail
description: Trace a claim back through its citation chain to the primary source and rate each hop. Use when a number, quote, or "study shows" claim matters - a decision, a publication, a pitch - and you need to know whether the chain underneath it is solid or goes soft before it reaches anything primary.
---

# Paper Trail

## When to use

- A load-bearing claim rests on "a study", "reports say", or a
  statistic with no visible origin
- Before citing a claim onward in something the user will publish

Do NOT use for claims nobody is relying on. Tracing is work; spend it
where being wrong would cost something.

## Walk the chain

1. Start from where the claim was found. What exactly does it cite?
2. Follow each hop: article to article, article to press release,
   press release to paper, paper to dataset. Record each hop with
   its URL and what it actually says.
3. Stop at the primary source - the original study, dataset, or
   direct statement - or at the point the chain dead-ends.
4. A dead end is a finding: "cited everywhere, sourced nowhere"
   happens more than anyone expects.

## Rate each hop

- **Primary**: original study, dataset, transcript, direct quote.
  Flag when the primary source IS the claimant - a vendor press
  release, an author's own blog, a company reporting its own numbers.
  That is primary by provenance and self-interested by credibility;
  the report must say both.
- **Secondary**: reports on the primary source accurately, with a
  working link to it - and "working" means you opened it: confirm the
  link lands on the claimed primary source, not a homepage or a dead
  page. Link rot is a finding, not a pass.
- **Soft**: blog-of-a-blog, no link, misquoted numbers, or a claim
  that mutated between hops (300 becomes "hundreds" becomes "most").
- **Broken**: the cited source says something different, or nothing
  at all.

Watch for the classic mutations: sample size shrinking, correlation
becoming causation, "in one small study" becoming "proven", and dates
quietly moving.

## Output format

The chain as a list, each hop rated, with the exact sentence at each
step that carries (or fails to carry) the claim. Then the verdict:
what the claim can honestly support - the strong version, the weak
version, or nothing.

## Hard boundaries

- Never upgrade a claim beyond what the primary source supports.
- Never treat a popular claim as a verified one; repetition is not
  evidence.
- If the primary source is paywalled or gone, say the trail is
  unverifiable - do not launder it through a summary of a summary.
