# Source Grading

Grade every source before citing a claim from it, using this scale. Record the grade
next to the source in the source list (Step 3 of `SKILL.md`).

## Grades

- **A — Primary, authoritative.** Original data/docs/announcement from the entity the
  claim is about (official documentation, a company's own filing/announcement, the
  actual research paper, a firsthand account/interview with a direct participant).
- **B — Primary, independent.** Firsthand reporting/analysis by an independent party
  with direct access (a journalist who obtained documents, an independent benchmark
  someone ran themselves, a practitioner's own writeup of their own experience).
- **C — Secondary, reputable.** A summary/analysis by a source with a track record of
  accuracy, citing its own sources (a reputable publication citing the primary source,
  a well-regarded technical blog that shows its work).
- **D — Secondary, unclear provenance.** Summarizes or claims something without
  clearly citing where it came from; may still be useful as a lead to find the
  primary source, but don't cite its claims directly without corroboration.
- **F — Unreliable.** Anonymous/unverifiable claims, content with an obvious
  undisclosed conflict of interest relevant to the claim (e.g. a vendor's marketing
  page as the sole source for a competitive comparison), or a source with a known
  track record of inaccuracy on this topic.

## Rules of use

- A-grade and B-grade sources can support a claim on their own. C-grade sources are
  strongest when corroborated by at least one other independent source. D-grade
  sources should not be the sole support for a claim in the final brief — use them to
  find a better source, or explicitly flag the claim as weakly-sourced if none exists.
- F-grade sources are not cited as support for a claim, ever — they can only be
  mentioned if the brief is specifically discussing misinformation/the existence of a
  claim, clearly labeled as such.
- Grade the *source*, not the claim's plausibility — a source doesn't get a better
  grade because you happen to agree with what it says. This is exactly the bias the
  disconfirming-evidence search in `SKILL.md` Step 2 is designed to counteract.
- Recency matters independently of grade: an A-grade source from 3 years ago on a
  fast-moving topic (pricing, product features, personnel) needs a freshness check
  even though its grade doesn't change.
- Note conflicts of interest even on otherwise well-graded sources (e.g. a company
  blog post is A-grade for "what the company announced" but should be flagged, not
  treated as neutral, for "is this product actually good").