Scoring Methodology
How we calculate demographic scores — the formulas, the source hierarchy, and the limits of what the data can say.
How we treat coverage
WhatSphere does not outsource research to outlets that are telling a story. Coverage is evidence that a claim was made. Scores require the underlying record.
Outlets that publish a finished moral — a victim, a villain, a cultural verdict — are doing advocacy or entertainment of their own. That work can be quoted. It cannot be the spine of a score. We rebuild from the production and the source.
- Primary. The finished work; billed credits; the source novel, comic, game, or script where an adaptation exists; dated public statements by named creators with a URL.
- Production record. Studio and guild credits, ratings boards, official synopses used only as metadata, documented consultancy or advocacy-organization involvement.
- Claim, not fact. News features, recaps, campaign interviews, and critic essays. Logged when they exist so a claim can be checked. Never used as the sole basis of a score.
How It Works
Every title is placed on gender and age axes that form four quadrants. Additional dimensions cover racial audience, faithfulness (adaptations only), and advocacy classification. A score of 50 is the baseline. No score is inherently good or bad.
Data Sources
- TMDB / IGDB / Jikan / AniList / YouTube — credits and metadata only.
- Primary record — the finished work and, for adaptations, the source text.
- Attributed statements — dated public comments by named creators with a URL.
- Claim, not fact — news features, recaps, and campaign interviews. Logged when they exist. Never the sole basis of a score.
Methodology Changelog
Current version: v1.3, last reviewed September 7, 2026.
- v1.3 (September 7, 2026) — Published the source hierarchy: primary records and attributed statements outrank press narrative. Mainstream coverage is treated as a claim to verify, not as research.
- v1.2 (May 31, 2026) — Published sourcing, limitations and bias disclosures; added the versioned changelog and confidence/sampling explanation; introduced last-verified stamps tied to the verification pipeline.
- v1.1 (March 15, 2026) — Added racial-audience sigmoid deviation model and the four-tier advocacy classification; introduced data-confidence scoring and the preliminary flag below 0.50 confidence.
- v1.0 (February 20, 2026) — Initial scoring methodology: gender and age axes from weighted factors, four quadrant scores, and adaptation faithfulness.
Technical axis weights, stretch formulas, advocacy tiers, confidence floors, and known biases remain in this page's prior long-form sections on production; the source hierarchy above is the v1.3 addition. Submit a correction →