OPEN-SOURCE GEOSPATIAL DATA FUSION · v0.2.0

See what the map knows—and where its sources disagree.

This live demo combines two fictional geospatial sources into one canonical world without erasing their evidence. Follow one building from raw records to a resolved identity, twelve traceable claims, and an intentionally preserved conflict.

Start the guided demo How synthesis works ↓

THE 60-SECOND VERSION

Sources go in. Evidence stays attached.

  1. 01
    Read bounded sources

    Every provider keeps its identity, license, geometry, and health status.

  2. 02
    Reconcile conservatively

    Strong identity evidence can unite records. Uncertain cases remain separate.

  3. 03
    Inspect neutral output

    Canonical entities retain every claim, conflict, relationship, and provenance link.

INTERACTIVE WALKTHROUGH

Follow the Harbor Workshop through the pipeline.

Use the three steps below, select any map feature, or open the evidence tabs. Everything shown comes from the generated synthesis result.

Fictional CC0 dataNo loginNo network providers
Canonical entities—building · parcel · road · POI
Evidence claims—each linked to a source record
Source records—across two available providers
Preserved conflicts—visible, not silently discarded

SPATIAL EVIDENCE

Source geometry and canonical output

OGC:CRS84

The same building footprint appears in both sources. The survey geometry was transformed from EPSG:3857 before synthesis.

Synthesized fictional harbor features Overlaid source and canonical building, parcel, road and point of interest geometries.
Click a green feature to inspect itSelected: Harbor Workshop
MATCH

RECONCILIATION EXPLAINED

Two source records, one defensible identity.

Both building records carry the same stable GERS identifier. The engine uses that explicit shared identity—not a vague confidence score—to group them.

building-17+survey-feature-a→Harbor Workshop

SOURCE HEALTH

What actually answered this run

2 available

A provider failure would be reported here—not misread as an empty place.

DISAGREEMENT

The conflict the engine kept

unresolved

Both sources describe the same building, but report different heights. The canonical entity selects an effective value while retaining both direct claims and their dates.

READING THE OUTPUT

Evidence classes are labels, not one fake confidence score.

Direct source

A provider explicitly supplied this value.

Derived

A repeatable method produced this relationship or value.

Conflicting

Supported claims disagree and remain inspectable.

Synthetic

Generated content must always be disclosed as such.

WORKING 0.2 CAPABILITIES

Use the same core with real bounded data.

The hosted demo stays offline and reproducible. The repository also includes live-source workflows, indexed matching, repeatable configuration, persistence, and four neutral export paths.

INGEST

Providers developers already meet

Local/HTTP GeoJSON, OGC API Features, ArcGIS Feature Services, OSM APIs, and pinned Overture Buildings through its maintained CLI.

npm run example:overture-osm
RECONCILE

Five entity-specific policies

Buildings, POIs, addresses, parcels, and roads produce MATCH, AMBIGUOUS, or NO_MATCH with the measured candidate signals attached.

npm run evaluate:reconciliation
INSPECT + UPDATE

Explain and compare snapshots

The CLI exposes contributing records, property selection reasons, provenance, conflicts, and affected entities between runs.

gws inspect · gws diff
EXPORT

Keep the result renderer-neutral

Canonical JSON retains the complete model; GeoJSON and FlatGeobuf carry the spatial view; PROV-JSON carries a compact lineage graph.

canonical · geojson · fgb · prov-json
Largest demonstrated matcher run200,000total generated input records
Reconciliation time771 ms8 GB Apple Silicon validation host
Evaluated candidates100,000instead of 10 billion all-pairs

These are deterministic grid-throughput measurements, not real-world accuracy or global-scale claims. Read the complete validation boundary ↗

USE THE ACTUAL SOFTWARE

Clone it. Run the same synthesis. Inspect the files.

The first run is deliberately offline. It writes canonical JSON and GeoJSON you can use without this viewer or World Explorer.

Download canonical JSON Download GeoJSON Download FlatGeobuf Download PROV-JSON
Terminal
git clone https://github.com/RRG314/geospatial-world-synthesis.git
cd geospatial-world-synthesis
npm ci
npm run example
What this demo does—and does not prove.

It demonstrates the real public data model, deterministic synthesis, evidence tracing, conflict retention, CRS normalization, and neutral exports on a small fictional dataset. It is not a full GIS client, universal conflation benchmark, cadastral authority, or safety-critical system.

Read validation and limitations ↗