Methodology

GeoScore methodology

How the index is built: design principles, the absolute scale, the aggregation scheme, and the rules that decide whether an indicator gets into the index at all.

Document status

Version 9 architecture. Axis and subindex weights are being calibrated and are not published here — what is published is the structure, the aggregation operators and the admission rules. If you are evaluating us ahead of an integration, this is the page to read, not the marketing one.

Ten design principles

These are constraints the index imposes on itself. Each one closes off a specific way of producing a number that looks good and means nothing.

  1. 1

    One index, one value, one moment in time

    GeoScore describes the state of a location today. Everything about the future — construction, plans, migration — lives in a separate momentum index. Otherwise it is unclear whether "72" means now or in three years.

  2. 2

    Every element answers exactly one question

    And belongs to exactly one parent. An industrial zone is split by the kind of effect it has: emissions go to environment, accident probability and the protection zone go to technogenic risk, jobs go to the economic base.

  3. 3

    An absolute scale with external anchors

    No score in the main index depends on the national distribution.

  4. 4

    The equal-coverage rule

    An indicator enters the main index only with national coverage of at least 90% and no systematic regional bias. Everything else goes into a city extension layer or into context flags.

  5. 5

    Not everything is monotonic

    Built density, building height and the jobs-per-resident balance have an optimum — more is not simply better. They need an inverted-U transform, not a linear ramp.

  6. 6

    Access is not only distance

    Every service is measured on four facets: proximity, capacity, choice and quality. "One pharmacy" and "twenty pharmacies" must not score the same.

  7. 7

    Risk is not averaged in with amenities

    Risk is aggregated non-compensatorily and applied to the index as a multiplier, not as a term in a sum.

  8. 8

    Missing data is not a value

    Three explicit modes: a neutral value with a coverage flag; the worst value, used only where the absence of the facility is itself the measurement; and full exclusion, with the weight redistributed inside the component.

  9. 9

    No price inside the index

    GeoScore is a feature in ASINT's valuation model. Any indicator derived from listings or transactions makes the index circular and is forbidden.

  10. 10

    The structure is checked after it is built

    If a subindex correlates above 0.90 with a neighbour it is a duplicate and gets merged. If a principal-component analysis over the subindexes yields fewer than three significant factors, the structure has not earned its size.

What "absolute" actually requires

Six requirements. The index is absolute only if all six hold — not five.

3.1

The scale is tied to an external anchor

For every indicator the 0 and 100 points come from a source outside the project: a building code, a WHO guideline, a physical limit, an international standard. Percentiles are forbidden in the main index.

3.2

The scale saturates

Every indicator has a point beyond which things do not get better: 500 m to a school and 50 m to a school are both "the norm is met". A cell's score does not change when a different cell changes.

3.3

The meaning of a value is fixed

0 — the facility is absent or the norm is grossly violated · 25 — the lower bound of acceptable · 50 — the national norm is met · 75 — good practice · 100 — international best practice or the physical maximum.

3.4

One anchor for the whole country

The norm is not adapted per region. If the average distance to a clinic in Karakalpakstan is 6 km, the score must show that rather than normalising to the Karakalpak average.

3.5

The source is uniform nationwide

One indicator, one source, the same collection logic everywhere. Mixing open map data in cities with a registry in villages is not allowed: a gap in completeness becomes a gap in scores.

3.6

A coverage admission gate

Coverage of at least 90% of inhabited cells; coverage in the worst region at least 0.8 of the best; missingness uncorrelated with territory type. Fail it and the indicator goes to a city layer, a context flag, or the backlog. Never a neutral 50 across most of the country.

Hierarchy levels

Six levels, from the index down to a single measurement. Codes are stable: renaming a label does not change its code, so an integration does not break because of an editorial edit.

GeoScoreone index, 0–100, absolute
Axis3 of them — different aggregation regimes
Subindex9 of them — a standalone question, published in the API
Componenta thematic block inside a subindex
Subcomponenta facet of measurement: proximity, capacity, choice, quality
Indicatora leaf: one measurement, one source, one anchor

The indicator passport

Mandatory fields for every leaf of the tree. An indicator without a complete passport does not make it into a build.

FieldWhat it describes
code / name_ru / name_en / name_uzCode and names in all three languages
definitionDefinition: what exactly is measured, in words, without ambiguity
unit, directionUnit and direction: metres, downward — less means better
spatial_opSpatial operator: cell, ring, distance decay, isochrone, two-step catchment
transformTransform: linear ramp, logarithmic, or a band with an optimum
anchors, anchor_sourceAnchors and their source: 500 and 1500 m per the building code
coverage, missing_policyCoverage and the missing-data policy
source, refresh, owner, versionSource, refresh frequency, owner, version

Indicator admission rules

All seven are mandatory. An indicator that fails even one does not enter the main index.

  • A theoretical justification — why this affects the quality of a place at all
  • Coverage, per the equal-coverage rule
  • An external anchor for the 0 and 100 points
  • Discriminating power — the share of tied values below a threshold
  • Non-redundancy — absolute correlation below 0.85 with neighbours inside the component
  • Reproducibility — the calculation repeats from the same inputs
  • No circularity with price

The aggregation scheme

Weights are not set here. What is set is the operators, and where compensation is allowed and where it is not.

LevelOperator
indicator → subcomponentweighted mean
subcomponent → componentweighted mean
component → subindexmean with an imbalance penalty
subindex → axes I, IImean with an imbalance penalty
subindexes 8, 9 → axis IIInon-compensatory: a severe hazard is not damped by a mild one

Final assembly

GeoScore = Q(I, II) × RiskFactor(III)

Opportunity and Quality of Place are combined with limited compensation, and Risk is applied to the result as a multiplier — ranging from a floor value up to one.

Limited compensation instead of full

A plain weighted mean lets transport at 95 fully compensate for an environment at 20. An imbalance penalty pulls down locations with a collapse in one dimension — without creating the geometric-mean problem where a single zero wipes out everything.

Risk is a multiplier, not a term

A place in a dam-breach zone does not get better because there is a school nearby. The multiplier's range is fixed explicitly so the penalty stays interpretable.

Influence is published, not just weight

Alongside every weight we publish an influence index: with correlated inputs, a nominal weight is almost never equal to actual influence. In the previous version transport had the largest weight and only the third-largest influence.

The missing-data policy

Missing data is not zero and not the average. Three explicit modes; the choice is recorded in the indicator's passport.

Neutral with a flag

A neutral value plus a coverage flag. Used when the absence of data says nothing about the place itself.

Worst value

Used only where the absence of the facility is itself the measurement — for example, no school at all within the access radius.

Exclude the indicator

The indicator is not computed at all and its weight is redistributed inside the component. This is the working mode for subindexes that currently ship in reduced form.

Published next to the index, but not inside it

The reasons differ: time horizon, circularity with price, being categorical, or being a metric about the index rather than about the place.

momentum — development dynamics

A separate 0–100 score with the same architecture: construction activity, planned infrastructure, demographic dynamics, market activity, investment pressure. The market block is our own data, and that is precisely why it lives here: an index containing price cannot be used as a feature in a price model.

The context layer

Attributes without a score: administrative names, settlement type, land category and designated use, master-plan zone, protection zones, coverage flags. The rule: a attribute present in fewer than 90% of cells, or that is a category rather than a magnitude, is a flag, not an indicator.

The relative view

The percentile of the absolute score within its own territory class. The absolute score answers "how good is this place"; the relative one answers "how good is it for a place of this type". The two numbers are published side by side and never mixed.

Derived outputs

Investment priority as the shortfall in score multiplied by population — where an investment reaches more people. And a population-weighted Gini coefficient by region — how unevenly quality is distributed within a region.

The registry and its gaps

329 indicators. Readiness is uneven, and we publish it as it is.

Nearly half the registry is available without a single external agreement. The remaining gaps cluster into three blocks: utility networks and amenities; welfare, demography and safety; hazardous facilities and pollution. Three or four agreements close half of all gaps. The hungriest subindexes are built environment and social environment: both are new, and until the data arrives they ship in reduced form.

What this document does not settle

An honest list of open questions is more useful than the absence of one.

  • The numeric weights of axes and subindexes — being calibrated, not published here
  • The floor value of the risk multiplier — fixed by a separate decision
  • Exact anchor thresholds for those indicators where no national norm exists
  • The composition of the built-environment and social-environment subindexes until external data arrives

Need the detail on a specific indicator?

We will send the indicator passport, its source, anchors and current coverage