Methodology
One static dataset
The warehouse prepares observations, quality flags and permitted comparisons. The website reads those files; it does not reconstruct national totals or silently replace missing data with an older number.
Periods and geography
Each observation has its own period. District boundaries can change between a historical survey and the available map. A current outline does not make an old observation a current-boundary estimate. Missing boundaries and inferred figures are labelled.
Comparisons
Ranks are ordered from the highest value, not from best to worst. Ties share a rank. A difference from the median uses the named cohort, and the denominator counts eligible observations in that cohort. Inferred, suppressed, suspect and small-sample observations are excluded from these statistics.
Map bands use stored national-cohort quartiles. Filtering to a state does not recalculate them. Ties can leave colour bands empty. A blocked comparison remains unavailable even when its individual observations may be shown.
Sources and missingness
Source IDs are recorded for each published observation and period. Inferred values do not claim a neighbouring observation’s publisher. Suppressed values are withheld; missing values are not zero. Source warnings accompany the relevant observations.
Release limits
This release focuses on districts, with selected state and national context. Some national headlines and cross-grain benchmarks are unavailable. Village-level analysis, automatic trend claims, correlations and authored stories are deferred.