CBDC privacy posture v10.2e · exploratory
Which official CBDC records document privacy-increasing choices, and which emphasise traceability or state visibility? This page preserves the original ranking interface but replaces the legacy index with the current ensemble measures.
Two different signals
How much coded material concerns privacy relative to selected design families.
A signed aggregate of supported design choices after ODR, relation, strength and ensemble weights.
Their ensemble correlation (95% CI [−0.102, 0.525], n = 39): attention is not commitment.
Exploratory ranking
Posture is shown on its native −1 to +1 scale. A zero is neutral or balanced, not missing. Rows are sorted by the current ensemble point estimate; the ranking is descriptive and will be revisited after consensus adjudication of the completed blind human audit.
| # | Entity | Posture | Privacy share | Dominant purpose | Candidate mass | Docs | |
|---|---|---|---|---|---|---|---|
| Loading… | |||||||
Institutional composites and non-national monetary areas are retained where they have decision/proposal evidence. This is not a validated league table of implemented privacy.
How it is measured
Exact evidence first. Every accepted statement has a source-verifiable span. Unsupported spans are rejected; source-only retry is allowed once.
One candidate, one vote. Each deduplicated candidate contributes total mass 1. Shared candidates split that mass across providers and distinct codes, so duplicate model output cannot increase weight.
Signed posture. Privacy direction is combined with whether the design affects visibility from the state or from intermediaries, its observation/decision/recommendation status and evidence strength. Decisions receive full ODR weight and proposals half weight.
Important limit. Calibration precision was 0.980 and span fidelity 1.000, but union recall was 0.766 and failed the frozen recall gate. The blind audit produced reviewer-specific validity estimates of 78.6% and 76.1%, yet inclusion agreement was 62.6%. These production values are therefore exploratory machine-assisted measures, not automated human gold.