Three-Alignment Classification (Not Left/Right)

We classify assets by geopolitical alignment, not domestic partisan lean:

• Western: NATO/EU/US-aligned media ecosystems (e.g., BBC, NYT, DW, France24)
• Non-Aligned: Global South, BRICS, neutral/alternative perspectives (e.g., Al Jazeera, RT en Español, Global Times, Telesur, The Hindu)
• Adversarial: State media from regimes actively opposing Western bloc (e.g., RT, Press TV, CGTN, Sputnik, Tasnim)

Classification uses: ownership structure, state funding %, editorial control, geopolitical stance consistency across 100+ test intercepts.

Lean Rating (LEFT / CENTER / RIGHT)

Within each alignment, we apply a traditional Left/Center/Right spectrum based on:

• Economic policy framing (regulation vs. markets)
• Social policy framing (progressive vs. traditional)
• Foreign intervention stance (interventionist vs. restraint)
• Keyword frequency analysis across 500+ tagged articles per asset

Rating is alignment-relative: a "Left" asset in the Western alignment differs from "Left" in the Adversarial alignment.

Vetting Rating (HIGH / MIXED / LOW)

Based on:

• Correction/retraction rate (public corrections per 100 articles)
• Fact-check aggregator scores (IFCN signatories, Snopes, PolitiFact, Africa Check, etc.)
• Primary source citation rate (links to docs, data, official statements)
• Anonymous sourcing frequency

Thresholds: HIGH = <5% correction rate, >80% primary citations. LOW = >15% correction rate, <40% primary citations.

Black Site / Silent Sector Detection

An intercept is flagged as a Black Site when one alignment has <20% representation among assets covering it, AND total assets ≥ 3.

Formula: gap_score = (1 - alignment_share) * log(total_assets). Higher = more significant gap.

Radar Intensity

Logarithmic scale: intensity = log1p(signal_count) / log1p(max_country_count). Prevents superpowers from drowning out smaller theaters.

No Algorithmic Personalization by Default

"Dead Drop" feed is opt-in only. Monitored frequencies stored in localStorage. No server-side profiling. No tracking pixels in Numbers Station.

Errata Policy

All corrections logged publicly at Errata. Each entry: original claim, correction, date, source intercept link.

Data Sources & Refresh

RSS/Atom feeds from 100+ assets, polled every 4 hours. Clustering via embedding similarity (threshold 0.78). Alignment labels assigned at asset level, not per-intercept.