How a statistic becomes a score

• All 125 countries set the scale. A frame built around one country would only describe that country.
• A high score is not a recipe, because what works depends on local conditions. The profile shows where to look, and local context decides what to build.
• Capability changes below the national level, in groups small enough to act. A country score is a coarse proxy for their conditions.
• Treat this as a measuring instrument that tests whether an attempt to raise a capability worked. Confidence, gaps and revisions sit beside each score.
• The diagnostics test whether dimensions track income. The limits page records known failures.

1. Take the latest comparable value for each country and indicator, with its source and year.
2. Apply the declared transform: per million people, log, or distance from a target.
3. Winsorize extreme outliers with Tukey fences at three interquartile ranges.
4. Normalize to 0 through 100 against the frame set by all 125 countries. Reverse indicators where lower is better.
5. Average the available indicators inside a dimension with equal weights.
6. Compute confidence separately as coverage × recency × source quality.

Missing indicators lower coverage and drop out of the mean, and nothing is imputed. Equal weighting keeps v0 easy to challenge.

The registry has 75 indicators: 34 with data, 18 gaps and 9 retired rows. Gaps have no comparable dataset; retired rows have a rejected one. Both lower confidence and define the collection agenda. 12 values come from reproducible source adapters and 2 from published tables, with retrieval dates stored.

• Cdirect capability measuredirect capability measure. Measures the thing itself. Days to register a company measures how hard it is to start a business.
• Icapability inputcapability input. Measures something that supports the capability. Research spending is an input. It does not prove a country reads the future well.
• Odownstream outcomedownstream outcome. Measures a result that usually follows from the capability. Patents can show experimentation, but also defensive filing. Korea files at volume, so its patent number says less than it looks.
• Pperception proxyperception proxy. Records what people or experts say, not what they did. The Worldwide Governance Indicators aggregate expert opinion. Seven were retired because they tracked income per head more than capability.

A tier says who published a number. The sources page lists the publisher, database and request.

• Coverage is the share of a dimension's indicators that have a value.
• Recency decays after two grace years, over a twelve-year window, to a floor of 0.1.
• Source quality is the mean tier weight of the values that are present.
• The product stays below 1 in practice, so the bands reflect real values.

• Historical values use today's frame, so the change reflects the country.
• The same indicators are used at both ends. A new indicator cannot create movement.
• The basket may be smaller than the dimension, so the trend level can differ from the score. Its size is printed beside the trend.
• Ten-year, twenty-year and longer spans are published when the matched evidence supports them. A missing span shows how far the data reaches.
• Each indicator has its own line back to 1960 where data exists. Nothing is carried forward or filled in.
• Each point carries the published value, normalized value and source tier.
• Each run compares its data with the previous file and logs restated, added or dropped values.
• Values more than five years old do not count for a year. Historical values outside the frame clamp to 0 or 100, and the clamp is recorded.
• Adoption indicators often rise for every country. Compare each change with the median before calling it progress.

• Each record carries a published number, reference period, source and retrieval date.
• Each record states what the case does not show.
• Records never affect scores or confidence.
• A gap becomes scorable when a comparable series covers at least two countries.

• Round 1: each panelist scores dimensions with thin source coverage from the evidence brief and its knowledge.
• Round 2: each panelist sees the anonymized round-1 scores and rationales, then revises or defends its scores.
• We keep the median and interquartile range. A range above 25 points is unresolved disagreement.
• Panel estimates stay in their own file and never enter the indicator score or confidence. The blended view uses one only when no indicator is observed.
• The panel also rates each indicator's class, validity, wealth-proxy risk and redundancy.
• The Delphi page shows the current run and its provenance. The active run is a working session, so it carries no panel distribution.

• 0 and 100 are the weakest and strongest values among the 125 countries. They are not a sample of the world, and a low score is not a percentage of capability.
• Scores use only the latest observation. Trends use a matched basket against today's frame. Nothing is back-filled or imputed.
• With 125 countries, diagnostics are hints, not established results.
• Doing Business series are frozen at 2019 and are marked down by the recency term.
• Retiring the perception composites left Coordination, Trust and Shared Purpose with one or two indicators each. The limits page carries the detail.
• Political uniformity is never treated as a capability.

How to argue with any of this and what evidence would overturn each decision.