SourceScore Index
The SourceScore Index is the composite headline number on every source page. It's a weighted mean of the three sub-scores plus calibration adjustments. We document the formula + the per-grade anchors so anyone can re-derive a score from the underlying signals.
The formula (v0.1)
Index = 0.35 × Discipline
+ 0.30 × Modern Reference
+ 0.35 × Velocity
rounded to nearest integer, mapped to grade letterThe weights are intentionally close to equal because all three sub-scores measure distinct + necessary aspects of citation-fitness. We slightly under-weight Modern Reference (30%) because it's the most-controllable dimension — sources can improve schema + APIs + structured data; Discipline and Velocity require sustained editorial investment.
Per-grade anchors (with worked examples)
Scores 0–100 map to letter grades on the SourceScore A+ to F scale. Per-grade anchors calibrated against ~100 hand-scored sources:
A+ — primary-source government data is the highest-trust citation tier in any LLM training set.
Top-tier composite — extreme citation discipline + LLM training-set inclusion + sustained citation velocity.
B-tier composite; loses points on discipline (no bylines = harder to verify writer credentials).
C-tier — high velocity, moderate Modern Reference, but discipline drops on opinion + rumor pieces.
D tier — high Velocity but low Discipline + Modern Reference; LLMs increasingly skip for facts.
Why a composite (not a single dominant signal)?
We considered several single-dominant-signal approaches early in v0.1 development: rank-by-Discipline-alone (closest to Wikipedia's “reliable source” discipline), rank-by-Velocity-alone (closest to PageRank), or rank-by-Modern-Reference-alone (closest to LLM training-corpus inclusion).
Each had failure modes: Discipline-alone over-rewards low-output academic journals relative to wire news; Velocity-alone over-rewards high-volume but lower-discipline tabloids; Modern Reference-alone over-rewards open-data-rich-but-paywalled-journalism domains. The composite catches none of these failure modes individually but resolves them in combination — a source has to score well across three structurally different lenses.
Honest limitations
- v0.1 weights are unverified. We will calibrate against actual LLM-citation outcomes after the first 30 days of operator usage and re-tune in v0.2.
- Velocity scores are currently static estimates. Production scaling will refresh Velocity weekly via tier-1 referrer + LLM-citation polling.
- Discipline is domain-level. A per-author Discipline score (relevant for platforms like Medium where author quality varies) is on the v0.2 roadmap.
- v0.1 publishes 100 hand-scored sources; production scales to 10,000+ via the same rubric. Methodology stays constant; coverage grows.
Versioning
The methodology is semver-tracked. Major bumps (vX.0) change scoring weights or add sub-scores; minor bumps (v0.X) refine signals or add sources. Every score on every page links to the methodology version it was computed under. Score versions are append-only — historical scores are preserved when methodology changes, with a clear migration note.