Signals
Measurement vocabulary — every dimension evaluated across 376 stories.
Core HRCB Human Rights Compatibility Bias (HRCB) is the core measured construct. It captures the
directional lean of web content relative to the 31 provisions of the UDHR.
Editorial (E)
What the content says — text, arguments, framing, sourcing, rhetoric. Structural (S)
What the site does — privacy practices, accessibility, tracking, access models.
Combined: (wE × Editorial) + (wS × Structural),
weighted by content type.
Classification Labels | Strong positive | +0.60 to +1.00 | | Positive | +0.30 to +0.59 | | Leaning positive | +0.10 to +0.29 | | Neutral | -0.09 to +0.09 | | Leaning negative | -0.10 to -0.29 | | Negative | -0.30 to -0.59 | | Strong negative | -0.60 to -1.00 | Content Type Weights | Code | Type | E | S | | ED | Editorial / News | 0.65 | 0.35 | | PO | Policy / Legal | 0.70 | 0.30 | | LP | Landing Page / Marketing | 0.40 | 0.60 | | CM | Community / Forum | 0.55 | 0.45 | | DC | Documentation / Reference | 0.50 | 0.50 | | AC | Academic / Research | 0.75 | 0.25 | | HR | Human Rights Focused | 0.60 | 0.40 | | PR | Product / Service | 0.40 | 0.60 | | PB | Personal Blog | 0.70 | 0.30 | | SO | Social Media | 0.50 | 0.50 | | OT | Other | 0.55 | 0.45 |
Appears on: /,
/past,
/rights/observatory,
item pages.
Source: UDHR, methodology v3.7.
Derived Metrics SETL [-1, +1]
Structural-Editorial Tension Level. Measures channel divergence:
sign(E-S) × √(|E-S| × max(|E|, |S|)).
High |SETL| = site says one thing, does another.
Confidence [0, 1]
Evidence-weighted average across 31 provisions.
H=1.0, M=0.6, L=0.2, ND=0.0. Higher = more provisions had strong evidence.
Volatility [0, 1]
Standard deviation of per-provision combined scores. Low (<0.2), Medium (0.2-0.5), High (>0.5).
High volatility = content aligns on some rights but conflicts on others.
Consensus [-1, +1]
Multi-model ensemble score. Weighted mean across rater evals (full=1.0, light=0.5).
Spread = max-min across raters. Model count = raters contributing.
Per-Provision Scoring
Each of the 31 UDHR provisions (Preamble + Articles 1-30) is scored independently on both
Editorial and Structural channels. The per-provision pipeline:
Content → E + S → Weights → Combined → DCP → Final → HRCB aggregate Per-Provision Fields | editorial | Editorial channel score for this provision | [-1, +1] | | structural | Structural channel score for this provision | [-1, +1] | | combined | Content-type-weighted blend of E + S | [-1, +1] | | final | After DCP modifier (plus/minus 0.30 max per article) | [-1, +1] | | evidence | Evidence strength for this provision | H / M / L / ND | | directionality | How the content engages with this right | A / P / F / C | Evidence Levels | H — High | Direct, explicit content. Score cap: 1.0 | | M — Medium | Clear signal, may be secondary. Score cap: 0.7 | | L — Low | Tangential or weakly sourced. Score cap: 0.4 | | ND — No Data | Topic absent from content. Score: 0.0 (not counted in aggregate) | Directionality Markers | A — Advocacy | Explicitly argues for or against a right | | P — Practice | Site infrastructure reflects a rights stance | | F — Framing | Presents issues in a rights-aligned or rights-opposed frame | | C — Coverage | Factual content relevant to human rights topics | Supplementary Signals
Nine supplementary signals capture how content communicates, orthogonal to HRCB's
directional lean. Grounded in established psychometric and information quality frameworks.
EQ Epistemic Quality [0-1] composite Source quality, evidence reasoning, uncertainty handling, purpose transparency. Source: CRAAP Test (library science) PT Propaganda Flags integer count Detects 18 propaganda techniques: loaded language, strawman, whataboutism, fear/prejudice appeal, bandwagon, false dilemma, ad hominem, and more. Source: PTC-18 corpus (Da San Martino et al. 2019) SO Solution Orientation [0-1] composite
Problem-only vs solution-oriented framing. High SO = content empowers rather than just alarms.
Source: Solutions journalism framework ET Emotional Tone VAD dimensions
Primary tone label + three dimensions: valence (pleasant-unpleasant), arousal (calm-excited), dominance (submissive-dominant). Each [0-1].
Source: Russell's Circumplex Model of Affect SR Stakeholder Voice [0-1] composite
Who speaks vs who is spoken about. Sub-fields: perspective_count, voice_balance, power_axis. UDHR rights-holder/duty-bearer framework. TF Temporal Framing categorical
Retrospective, present, or prospective focus. Time horizon: immediate, short-term, medium-term, long-term. GS Geographic Scope categorical
Local, national, regional, or global scope with specific regions mentioned. CL Complexity Level categorical
Reading level, jargon density, and assumed knowledge requirements. TD Transparency & Disclosure [0-1] composite
Author identified, conflicts of interest disclosed, funding sources disclosed.
Global Averages (376 stories) | EQ | | 0.54 | | SO | | 0.48 | | SR | | 0.35 | | TD | | 0.39 | | PT | | 0.6 flags | PT bar shows inverted count (lower = better). Fair Witness Layer Inspired by Heinlein's Fair Witnesses from Stranger in a Strange Land, every full evaluation separates evidence into two categories: Observable Facts
Directly verifiable statements grounded in page content. Any reader could confirm by visiting the page. Inferences
Interpretive conclusions drawn from observable evidence. Explains why evidence maps to the score. FW Ratio FW Ratio = observable_facts / (observable_facts + inferences)
Scale [0, 1]. Higher = evaluation more grounded in verifiable observations; lower = more interpretive weight. Stored per-provision in the fair_witness table. Toggle Fair Witness on any item page to view evidence breakdowns. Domain Context Profile (DCP) Eight domain-level elements provide inherited modifiers that adjust per-provision scores based on the publishing domain's policies and practices. | Privacy | ToS | Accessibility | Mission | | Editorial Code | Ownership | Access Model | Ad/Tracking | Each element can shift a per-article score by up to ±0.30. DCP profiles are cached in KV (7-day TTL) and persisted to domain_dcp in D1. Appears on: /domains, domain detail pages. Only used in full evaluations (not light). Content Gate A pre-evaluation content classifier that identifies non-evaluable pages before they enter the evaluation queue. Pure regex — no LLM calls. | Category | Description | | paywall | Subscription wall blocks content access | | bot_protection | Cloudflare/Akamai challenge pages | | captcha | CAPTCHA or verification required | | login_required | Authentication wall | | cookie_wall | Cookie consent blocks content | | geo_blocked | Region-restricted content | | age_gate | Age verification required | | app_only | Content only available in mobile app | | rate_limited | Rate limit or throttle page | | error_page | 404, 500, or other error pages | | redirect_loop | Redirect chains or dead ends | | empty_content | Page has no readable text content | Runs at two points: cron pre-fetch (primary) and consumer (safety net). Writes gate_category and gate_confidence to the stories table. Gate stats: /sources, /domains, /system. Labels & Metadata Sentiment Labels | Champions | Strongest positive alignment with UDHR provisions | | Advocates | Actively promotes rights-aligned positions | | Acknowledges | Recognizes rights issues without strong advocacy | | Neutral | Balanced or no clear directional lean | | Neglects | Overlooks or minimizes rights concerns | | Undermines | Actively works against rights provisions | | Hostile | Strongest negative alignment | Theme Tags Free-form 2-4 word labels identifying the dominant human rights theme (e.g., "Privacy & Surveillance", "Free Expression", "Labor Rights"). Content Type Codes | ED | Editorial / News | | PO | Policy / Legal | | LP | Landing Page / Marketing | | CM | Community / Forum | | DC | Documentation / Reference | | AC | Academic / Research | | HR | Human Rights Focused | | PR | Product / Service | | PB | Personal Blog | | SO | Social Media | | OT | Other | Evaluation Modes Two evaluation modes serve different cost/quality trade-offs: | Full | ~lite | | Channels | Editorial + Structural | Editorial only | | Provisions | 31 per-provision scores | Single aggregate score | | DCP | Yes (8 domain elements) | No | | Fair Witness | Yes (facts + inferences) | No | | Supplementary | All 9 signals | 3 (EQ, SO, TD) + tone | | SETL | Yes | No (single channel) | | Confidence | Evidence-weighted | N/A | | Output tokens | ~4-5K | ~200-400 | | Schema | 3.7 | light-1.3 | | Models | Claude Haiku 4.5, OpenRouter full models | Workers AI (Llama 4, Nemotron) | | Feed label | HRCB | ~lite | Light evaluations show ~lite in the feed. Item pages display an editorial summary card instead of the full heatmap. Light scores are not directly comparable to full weighted means. |