text
| 1 | # Source Grading |
| 2 | |
| 3 | Grade every source before citing a claim from it, using this scale. Record the grade |
| 4 | next to the source in the source list (Step 3 of `SKILL.md`). |
| 5 | |
| 6 | ## Grades |
| 7 | |
| 8 | - **A — Primary, authoritative.** Original data/docs/announcement from the entity the |
| 9 | claim is about (official documentation, a company's own filing/announcement, the |
| 10 | actual research paper, a firsthand account/interview with a direct participant). |
| 11 | - **B — Primary, independent.** Firsthand reporting/analysis by an independent party |
| 12 | with direct access (a journalist who obtained documents, an independent benchmark |
| 13 | someone ran themselves, a practitioner's own writeup of their own experience). |
| 14 | - **C — Secondary, reputable.** A summary/analysis by a source with a track record of |
| 15 | accuracy, citing its own sources (a reputable publication citing the primary source, |
| 16 | a well-regarded technical blog that shows its work). |
| 17 | - **D — Secondary, unclear provenance.** Summarizes or claims something without |
| 18 | clearly citing where it came from; may still be useful as a lead to find the |
| 19 | primary source, but don't cite its claims directly without corroboration. |
| 20 | - **F — Unreliable.** Anonymous/unverifiable claims, content with an obvious |
| 21 | undisclosed conflict of interest relevant to the claim (e.g. a vendor's marketing |
| 22 | page as the sole source for a competitive comparison), or a source with a known |
| 23 | track record of inaccuracy on this topic. |
| 24 | |
| 25 | ## Rules of use |
| 26 | |
| 27 | - A-grade and B-grade sources can support a claim on their own. C-grade sources are |
| 28 | strongest when corroborated by at least one other independent source. D-grade |
| 29 | sources should not be the sole support for a claim in the final brief — use them to |
| 30 | find a better source, or explicitly flag the claim as weakly-sourced if none exists. |
| 31 | - F-grade sources are not cited as support for a claim, ever — they can only be |
| 32 | mentioned if the brief is specifically discussing misinformation/the existence of a |
| 33 | claim, clearly labeled as such. |
| 34 | - Grade the *source*, not the claim's plausibility — a source doesn't get a better |
| 35 | grade because you happen to agree with what it says. This is exactly the bias the |
| 36 | disconfirming-evidence search in `SKILL.md` Step 2 is designed to counteract. |
| 37 | - Recency matters independently of grade: an A-grade source from 3 years ago on a |
| 38 | fast-moving topic (pricing, product features, personnel) needs a freshness check |
| 39 | even though its grade doesn't change. |
| 40 | - Note conflicts of interest even on otherwise well-graded sources (e.g. a company |
| 41 | blog post is A-grade for "what the company announced" but should be flagged, not |
| 42 | treated as neutral, for "is this product actually good"). |
| 43 |