The Observatory measures conduct against the five truce stipulations. This volume measures the substrate underneath it: health, housing, air, infrastructure, work and civic participation, across the same 193 member states and from named public endpoints. Nothing here is a compliance finding. It is the context a compliance finding has to be read in.
Every sentence below is arithmetic over the dataset this page carries and nothing else. Where an association is reported it is a rank correlation with its sample size printed beside it, and it is never presented as a cause.
Each card carries the unit, the number of member states with a reading, the vintage span of those readings, the observed distribution and the reference that defines good. Where an external target exists it is named and dated. Where none exists the 90th percentile of the observed population is used and is labelled an observed benchmark, never a target. The five best and five worst readings are listed under each.
A single number over many series is the easiest thing on this page to misuse, so the construction is published in full and the admission of a series is decided by a rule rather than by judgement. A series enters only if it covers at least of the 193 member states and carries a median vintage no older than . A state receives a score only if it has readings for at least 80% of the admitted series. Scores are percentile ranks within each series, so a single extreme reading cannot dominate.
| Series removed | Weight% of index | Median rank movementplaces | Largest movementplaces |
|---|
The Observatory's conduct reading is the forced-displacement trajectory: the change in the stock originating in a state between 2023 and 2024. Below, each structural series is rank-correlated against that trajectory, then against it again with the INFORM Risk index held constant, so that an association is not simply restating that hard places are hard. A correlation is a statement about co-movement in a cross-section. It is not evidence that changing the first quantity would change the second.
A cross-section of levels compares rich states with poor ones, so it cannot say whether conditions move conduct. This section does two things a cross-section cannot. It differences every series within the state, which removes anything about that state that did not change over the decade. And it puts the structural change first and the conduct outcome second, so that a coefficient at least has the order of events on its side. The eleven-year displacement panel behind it covers member states and reconciles exactly with the Observatory on every year it shares.
Each row is one structural series. Each column is a different question asked of the same data. Reading across a row shows how much of an association survives when the design is tightened, which is the only way to tell a real relationship from an artefact of the specification.
| Structural series | A. Level, then outcome2024 level vs 2019 to 2024 change | B. Change on changesame window, both differenced | C. Structure first2014 to 2019 vs 2019 to 2024 | C adjustedC, structural risk held constant | Reverseoutcome first, structure second |
|---|
A distance to a percentile is a distance to other states, so it cannot say where a state's largest shortfall is. Only the five series that carry a published external target are used here: under-five mortality against SDG 3.2.1, fine particulate exposure against the WHO 2021 guideline, informal settlement share against SDG 11.1, electricity access against SDG 7.1 and secondary enrolment against SDG 4.1. For each state the shortfall on each is ranked against every other state's shortfall on the same series, and the worst of those is reported as that state's leading deficit.
| State | Leading deficit | Reading | Target | Shortfall percentileagainst all states | Second | Third |
|---|
Every member state, every admitted series, the substrate score and the Observatory's conduct reading side by side. Sort any column. Click a row for the full profile including the series the composite excluded.
A detector that did not connect is reported here in the same rack as one that did, with the reason. Four of the quantities that were asked for have no keyless machine-readable source at member-state scope, and the page says so rather than substituting something adjacent and calling it the same thing.
The dataset behind this page canonicalises to a single SHA-256 digest. Strip the two fields the seal cannot cover, serialise with sorted keys and no whitespace, and hash. The digest below is what you should get. The Observatory volume carries its own digest and this page records it, so the pair can be shown to have been built from the same ingest.