Normalization
Rescaling indicators measured in different units onto a common scale so they can be combined. The choice of method and bounds shapes the result — which is why the MEI publishes its bounds and uses a unit-invariant method.
Rail: Macro · Updated: 2026-07-09
What It Is
Normalization is the process of transforming indicators measured in different units onto a common, comparable scale, so that they can be aggregated. An index that combines dollars, transaction counts, and percentages can't add them directly — the units are incommensurable, and without rescaling, the variable with the largest raw numbers would dominate the result regardless of the intended design. Normalization strips the units and maps each indicator onto a shared, dimensionless scale.
The main methods are min-max scaling (rescaling a value by its position between a chosen minimum and maximum, to a fixed range like 0–100), z-score standardization (centering on a mean of zero with unit standard deviation), and distance-to-a-reference (measuring against a benchmark). For min-max, the choice of minimum and maximum — the "goalposts" — is a consequential structural decision: those bounds define the whole scale, so setting them too wide compresses real differences and too narrow exaggerates small ones.
For quantities that are highly skewed or grow multiplicatively — incomes, market values, counts that compound — a logarithmic transformation is standard practice before or during normalization, so that equal proportional changes register as equal steps and a few large values don't crush the rest of the scale into a corner. The UN Human Development Index is the canonical example: it uses min-max goalposts for each dimension (life expectancy bounded between 20 and 85 years, for instance) and applies a logarithm to the income dimension specifically to reflect the diminishing marginal value of additional income.
Why It Matters for the Machine Economy
Normalization is where the MEI made its most concrete methodological correction, and the fix is worth stating precisely. The platform's earlier approach normalized unbounded metrics with a formula of the form log(1+value)/log(1+max) — and that formula has a serious defect: it is not unit-invariant. The same economic quantity produced a different normalized score depending on the units it happened to be expressed in — roughly 80 measured in dollars versus roughly 42 measured in millions of dollars, for the identical underlying figure. A normalization whose output depends on an arbitrary choice of units isn't measuring anything objective. The redesign replaced it with log-space min-max normalization, which is unit-invariant (converting the units multiplies value, floor, and ceiling by the same factor, which cancels out), and which provides the explicit published floor the geometric mean's scale requires.
The bounds themselves are treated as the most load-bearing parameters in the system, and are handled the way the composite-indicator literature (and the HDI's goalpost tradition) recommends: every metric's floor and ceiling are published and versioned, because a normalized score of "82" is meaningless without knowing it means "82% of the way from a stated floor to a stated ceiling on a log scale." And the bounds aren't arbitrary — each floor is defined as a dormancy threshold (the level below which the thing being measured is effectively absent) and each ceiling as a maturity goalpost (the level at which that metric's growth phase would be complete). Publishing the full bounds table is the platform's core replicability commitment: it's what lets an outside party reproduce and check the index rather than take it on trust.
Real-World Example
The HDI rescales income between a chosen minimum and maximum after taking a logarithm, so that the jump from very poor to modestly better-off counts for more than an identical dollar jump between two wealthy countries. The MEI applies the same family of technique to a metric like on-chain payment volume: a log-space min-max scale between a published floor (below which activity is just pilot noise) and a published ceiling (a stated maturity level), so that proportional growth registers consistently and the score means a definite, checkable position between two disclosed bounds.
Related Terms
- MEI (Machine Economy Index) — the index whose metrics are normalized this way
- Geometric Mean — the aggregation that requires the floored, normalized scale
- Saturation and Rebasing — what happens when a normalized metric outgrows its bounds
- Composite Index — the general construction normalization is a step in