MacroScore = Σ(pillar_i × weight_i)City and country indicators are normalized to comparable 0-100 pillar scores before weighting.
HumanRank alpha
Public methodology
HumanRank publishes weights, sources, uncertainty and correction requests so the ranking stays debatable, auditable and versioned.
research note
Technical whitepaper
The alpha publishes its operating math so assumptions can be challenged, bias can be audited and weight changes can be proposed. Cohort scoring still uses normalized seed inputs; the production pipeline should replace them with fully lineaged BigQuery observations.
Published formulas
These formulas describe the current macro cohort model. Public figures use a separate sourced editorial model version.
EarthScore = round(clamp(MacroScore + AgeCurve + CohortVariance, 1, 99) × 10)The alpha cohort score is a potential estimate, not an identity-level measurement.
Percentile = logistic((EarthScore - 520) / 125)The logistic curve maps score to a global rank without materializing every living person.
GlobalRank = round((1 - Percentile) × 8,120,000,000)Public figures and anonymous cohorts are interleaved into the same living-population rank space.
Confidence = f(source quality, recency, missingness, geography precision, stability)Confidence controls the published uncertainty band and should rise only with better evidence.
data lineage
Cohort data lineage
Every pillar must be reproducible: source, year, geography, license, transform and missingness policy. This table is the production contract, not a claim that every local datapoint already exists.
transform: Log income transform blended with local quality-of-life signal, then robust percentile normalization.
missingness: Fall back from city to country to region; increase uncertainty when geography is coarse.
transform: Orient life expectancy and burden indicators so higher means better; winsorize extreme tails.
missingness: Use latest comparable year; penalize stale or country-only health data.
transform: Normalize attainment, expected schooling and access indicators inside comparable income/geography bands.
missingness: Missing subnational detail is imputed from country data and reflected in confidence.
transform: Normalize public safety and conflict exposure indicators after orienting lower-risk values upward.
missingness: No private or protected-attribute safety proxy is allowed; missing city data widens uncertainty.
transform: Normalize air, climate and service-quality indicators against comparable urban areas.
missingness: Use country-level planetary-pressure controls only when city indicators are absent.
transform: Normalize internet, transport and service-access indicators; cap outliers from dense global hubs.
missingness: Prefer official local statistics; otherwise use country connectivity with lower confidence.
transform: Blend labor-market, mobility and services indicators after equal-weight and sensitivity checks.
missingness: Do not infer opportunity from wealth alone; unresolved inputs lower confidence.
Model governance
Editorial weights and votes
2 votes
Verifiable achievement26% active
Awards, offices, records and public credentials with sources. Keeps fame or wealth from dominating alone.
more0
keep0
less0
Social impact22% active
Measures scale and depth of public contributions, with caps so visibility is not confused with benefit.
more0
keep0
less0
Public notability18% active
Uses coverage and encyclopedic presence as a verifiability signal, not as moral approval.
more1
keep0
less0
Public sentiment12% active
Stays capped because public mood is noisy, manipulable and culturally uneven.
more0
keep0
less0
Source quality10% active
Raises confidence when sources are official, stable and multiple; penalizes less auditable profiles.
more0
keep1
less0
Wealth / economic reach7% active
Counts lightly and as an estimated range. The open question is how much society values economic capacity.
more0
keep0
less0
Reputation stability5% active
Does not reward ideological consensus; it only limits jumps when public evidence is unstable or disputed.
more0
keep0
less0
Age / future horizonnot active
Not active by default. It can be debated as a life-stage signal, with high risk of age bias.
more0
keep0
less0
Votes do not change rank in real time; they go into review for the next model version.
confidence model
Uncertainty breakdown
Uncertainty is not decoration. It is where the model admits that a macro cohort is not a real person and less precise data should carry less confidence.
Missingness: incomplete city or subnational indicators widen the score interval.Recency: older source years lower confidence until refreshed in BigQuery.Geographic precision: country-level proxies are less precise than city-level observations.Source quality: official and stable sources raise confidence; weak or single-source evidence lowers it.Model stability: scores must be compared against equal-weight and PCA-style baselines before weight changes.
Reproducibility and limits
Every production indicator must carry source, URL, geography, year, license, transform and ingestion timestamp.Every model release must keep its weights, formulas and score distribution snapshot.Named public profiles must show source count, model version, confidence and a correction path.Anonymous population cohorts stay aggregated until a person opts in to claim a profile.HumanRank is not a credit, employment, housing, insurance or government-benefit decision system.
Scientific controls
Protected attributes are not score inputs.Every indicator needs source, geography, year, license, transform and missingness metadata.Scores must publish uncertainty, not just rank.Weights are model-versioned and sensitivity-tested before release.Users can inspect, revoke and challenge personal evidence.Social proof is evidence, never a raw popularity boost.
Claim status
Google claim is still pending until a valid OAuth Web Client is configured in Google Auth Platform. The site already accepts editorial corrections and weight votes.
Source registry
World Development IndicatorsWorld Bank
level: country | use: income, population, education, connectivity and development indicators
Global Health ObservatoryWorld Health Organizationlevel: country | use: life expectancy, mortality, health burden and health system indicators
Human Development DataUNDPlevel: country | use: human development, inequality and planetary pressure controls
World Population ProspectsUN DESA Population Divisionlevel: country | use: population denominators and age cohort estimates
Official city and national statistical officesLocal public agencieslevel: city | use: city-level income, safety, air quality, housing and service access where available
Production pipeline
- Ingest raw indicators from official or methodologically documented sources into BigQuery.
- Normalize each indicator with robust percentiles or z-scores inside comparable geographies.
- Winsorize extreme values, orient all indicators so higher means better, and record the transform.
- Compute pillar scores with versioned weights and compare against equal-weight and PCA baselines.
- Publish confidence from source quality, recency, missingness, geographic precision and model stability.
- Run fairness audits on aggregate opt-in groups without using those attributes in scoring.
Opt-in evidence
Credentialsmax weight 22%
Prefer issuer verification, expiration dates and domain relevance over brand prestige.
Work profilesmax weight 18%
LinkedIn or portfolio data must be normalized by role, seniority, geography and opportunity access.
Peer praisemax weight 12%
Endorsements are capped, domain-specific, duplicate-resistant and weighted by rater reliability.
Health activitymax weight 16%
Activity signals need device provenance and disability-aware interpretation.
Civic impactmax weight 14%
Donations, volunteering and open work are scored by verified impact, not dollar amount alone.