Methodology

MOSAIC reads the public Displacement Tracking Matrix API, version 3, operated by the International Organization for Migration. The API exposes aggregated, non-sensitive figures for internally displaced persons at country, Admin 1 and Admin 2 level, drawn from DTM baseline assessments and multi-sectoral location assessments.

Five endpoints are used: a country list, an operation list, and one displacement endpoint per administrative level. Requests carry a subscription key issued by the DTM developer portal. The key is held server-side and never reaches the browser.

DTM figures may be estimates derived from key informants and secondary-source cross-checking rather than exact enumeration. Assessment rounds are deployed at different frequencies depending on the needs of each context, so cadence varies by operation and by country.

A DTM v3 row is not a total. Rows arrive disaggregated simultaneously by operation, displacement reason, origin Admin 1 and assessment type, so a single administrative unit in a single round is represented by several rows. MOSAIC sums those rows to a present-IDP figure keyed on operation, administrative unit, round number and reporting date. Any analysis that treated one row as a unit total would understate the figure.

Nothing is aggregated across operations. Two operations covering the same country may differ in geographic scope, cadence and method, so their totals are not additive and their series are not comparable. Where a country is covered by several operations, MOSAIC selects one as the analytic frame, by most recent reporting date and then by widest unit coverage, and the operation is selectable on every country page.

Sex disaggregation is reported only where at least one contributing row supplied it. A missing split is left missing rather than inferred.

DM = (IDP_t − IDP_t−1) / IDP_t−1

Displacement Momentum is the proportional change in present IDP individuals between an administrative unit's latest comparable DTM assessment round and the preceding round within the same DTM operation. Bands are MOSAIC-defined monitoring thresholds, not IOM classifications. A momentum value is not computed where a baseline is missing or zero.

These thresholds are product-design choices. They have not been calibrated against country context, assessment cadence, sample or coverage shifts, or historical variability, and they should be calibrated before operational use. No momentum value is produced where the baseline is missing or zero. A rise from zero is reported as a new observation with its absolute value, never as an infinite or imputed percentage.

Comparison runs only between consecutive rounds of the same operation for the same administrative unit. Where that condition is not met, the reason is stated rather than silently absorbed:

• No comparable previous round
• Newly reported area
• Area absent from the latest round
• Non-consecutive round numbers
• Reporting date missing
• Previous value was zero
• Assessed-unit count changed between rounds
• Built from a single source record

A unit present in one round and absent from the other is never treated as zero. Absence from an assessment is a statement about the assessment, not about displacement.

Each rule is a transparent statement about a unit's own trailing series within one operation. No model is fitted and no parameters are learned, so every signal can be reproduced by hand from the underlying observations.

Sudden surge

Latest round-to-round change is at or above +30 per cent and the absolute increase is at least 1,000 individuals.

Sustained multi-round increase

Three or more consecutive rounds of positive change, with cumulative growth of at least +25 per cent.

Deviation from recent trend

The latest change exceeds 3.0 times the median absolute change of the prior rounds in the comparison window, requiring at least four prior observations and a non-zero median.

Abrupt decrease

Latest round-to-round change is at or below -25 per cent and the absolute decrease is at least 1,000 individuals.

Newly observed location

The unit reports displacement in the latest round and has no prior observation in this operation.

Location reappeared after absence

The unit reported displacement, was absent for at least one round, and reports again in the latest round.

A signal states that a figure moved unusually relative to its own history. It is not evidence of conflict escalation, and it does not distinguish a population change from a measurement change. Signals are cues for collection.

MOSAIC reports simultaneous increases among units sharing a parent Admin 1 as a clustered increase. It does not claim spatial contiguity, because the public API supplies place codes but no adjacency graph or boundary geometry.

Concentration is computed over the administrative units reported in the selected round of the selected DTM operation. It uses the Herfindahl-Hirschman Index of unit shares, normalised so that zero is an even distribution across assessed units and one is total concentration in a single unit. Units not assessed in the round are absent from the calculation, so a narrowing of assessment coverage will raise measured concentration without any population movement.

MOSAIC confidence describes how well a change is grounded in comparable DTM observations. It begins at 100 and applies fixed deductions for missing baselines, non-consecutive rounds, coverage changes, short unit histories and assessment age. It is a statement about data comparability and recency only. It carries no claim about the cause of a change.

The matrix records how each item of evidence bears on each hypothesis. MOSAIC counts inconsistent entries and weights them by source reliability, information credibility and stated relevance. It does not compute a probability, rank hypotheses by likelihood, or select a conclusion. Ordering by weighted inconsistency reflects only which hypotheses the recorded evidence has not yet eliminated.

The seeded hypothesis set deliberately includes two measurement explanations, coverage or enumeration change and return or onward movement effects, alongside the conflict and disaster explanations. A hypothesis set offering only conflict explanations would bias the analysis before any evidence was recorded.

Workspaces are stored in the analyst's browser. Nothing is transmitted to a server and nothing is shared between users. Exports are deterministic: the same workspace always produces the same Markdown and the same JSON.

The global map colours whole countries from Admin 0 figures, drawn over Natural Earth 110m public-domain outlines bundled with the application. No external tile service is contacted.

MOSAIC does not draw subnational polygons. The public API supplies place codes but no boundary geometry, and rendering approximate subnational boundaries would assert a spatial precision the data does not carry. Subnational distribution is presented as ranked quantities instead, which is what the data supports.

• It does not attribute a cause to a change in recorded displacement.
• It does not predict conflict, escalation or displacement.
• It does not impute missing observations or interpolate across gaps in a series.
• It does not process, infer or display personal data or individual locations.
• It does not reconstruct sensitive microdata from aggregated figures.
• It does not score or select a hypothesis on the analyst's behalf.
• It does not present any derived indicator as an IOM classification.