Methodology & data reference

The seven scored domains below blend to the composite (default ULTRAPLAN §1 weights shown). Each row names the underlying dataset, the data's real granularity before we attribute it to a statistical area, and the join method. Direction records the honest reading (“higher / lower is better”).

Context layers & pins - never scored

These are compiled for transparency and exploration. They never enter the composite, the weights, or the data-confidence index.

Every dataset in the default Greater Melbourne build, rendered straight from the committed manifest (data/generated/sources.json). Other capitals resolve their map/profile source ids against their active region manifests (data/generated/sources.<region>.json). Each non-derived source records a sha256 of its raw file (see provenance). “Derived” rows are computed from other sources and carry no raw file of their own.

Each scored indicator is percentile-ranked withinGreater Melbourne (relative, not absolute), inverting indicators where higher is worse (rent, crime, hazard overlay). A domain score is the sub-weighted blend of its indicators; the composite is the weight-blend of the seven scored domains. Default weights: Rent burden 30, Transport 18, Crime/Safety 14, Health 14, Hazards 8, Education 8, Income/Economy 8 - adjustable by sliders / lens presets. The score uses each weight's share (ratio) of the total, so a slider holds the value you set.

Missing data is never imputed: a missing indicator gets a null percentile, is excluded from the weighted total, and its weight is re-distributed across the present scored domains. The composite and lens-weighted scores are optional lenses, never a definitive ranking.

Sources arrive at different geographies. Statistical-area-direct ABS series are used as-is. Suburb/LGA or territory-level crime series are aggregated to the statistical area by population-weighted spatial intersection (area-weighted fallback where mesh-block population is unavailable); every aggregated value records which method it used. Point data (hospitals, GP, schools, pins) needs no crosswalk - it is assigned to the area it falls in, with proximity measured from the area centroid. Polygon overlays (hazards) are area-weighted against each statistical area.

Non-residential areas (estimated resident population < 200 - airports, parkland, industrial, water) are excluded from percentile baselines and rankings, and drawn in neutral no-data grey (#d9d6cf) rather than a misleading score.

Each place profile is a tabbed drawer (adapted from the Analisa.pt municipality-drawer pattern): an Overview tab (composite breakdown, key facts, resident-population trend), one tab per scored domain, and context tabs (Home buyer, Walk & cycle, Equity & community, Data coverage). Re-weighted reads of the seven domains live in the map's Lens picker (Balanced, Renting, Buying, Family, Retiree, Data quality), not on the profile.

Inside a domain tab, each indicator card shows its value, unit, honest direction, active-region percentile, source, and a benchmark band - this area's raw value against the active-region median and P25-P75 range across residential areas, computed at build from the full region dataset. Time-series are shown only where we hold ≥3 real points (population; property & violent crime, labelled council-level (LGA) so they are not misread as area-precise); every trend line states its geography, the number of points and period range, and links here for any boundary break. Indicators with a single period simply omit the trend rather than fabricating one.

Equity (SEIFA IRSAD/IRSD deciles), community (tenure / renter %, apartment %, Year-12 completion %, and - among residents who hold a post-school qualification - the bachelor+ / postgraduate share), population trend, and home-buyer index appear for transparency only. The home-buyer index blends indicators we already hold (cost-pressure 28%, safety 18%, schools 16%, transport 14%, low hazard 14%, walk access 10%) into a active-region percentile - using no sale-price data, so it is not a price or capital-growth estimate.

15-minute access counts how many of eight everyday-amenity categories sit within ~1.2 km of the area centroid (straight-line, not street-network). Cyclability is the density of OSM cycle infrastructure per km² (an infrastructure measure, not a safety/comfort rating). Both are OSM-derived (ODbL), community-maintained and uneven in coverage, and appear as a profile panel plus an optional map layer.

Map pins (hospitals, GP, pharmacy, police, schools, childcare, post, pathology/NDIS, supermarkets, parks, gyms, cafes) are off by default and toggled per category, colour-coded by a categorical palette kept separate from the YlGnBu choropleth ramp. NDIS and pathology are sparsely tagged in OSM - treat their coverage as indicative, not complete.

Each area carries a 0–100 data-confidence index combining domain coverage, completeness (non-missing sub-indicators), freshness, and aggregation-method confidence (directly measured > crosswalk-estimated > proximity). It describes how well-measured an area is - a property of our pipeline, not a judgement of the place - and is shown as an optional map layer and a per-area report card. Across Greater Melbourne it is near-uniform (about 86-95) and shows no correlation with income or SEIFA (r ≈ 0). The profile's Data-coverage panel states, per domain, what the data actually represents and which indicators are measured, missing, or stale.

Each source records a cadence and, where the upstream API exposes it, a last-updated date plus a sha256 of its raw file. A scheduled job re-fetches, rebuilds, and re-hashes monthly; when a raw file's hash changes the map redeploys, and when upstream is unchanged the build is a no-op. A blank hash in the manifest above means that source was added to the manifest and will be stamped on the next full build.

When we find and fix an error in the underlying data - a wrong geographic join, a stale figure, or a licence we had to drop - we keep a dated record of it. This is the public version of that record. It is a transparency log of what changed and when, not a ranking or a judgement of any area.

Fixes are reviewed and applied in the data pipeline, then reflected in the map and profiles on the next build. This log is published separately from our internal review notes and starts empty.

• Crime is recorded at suburb/LGA level and allocated to the statistical area by crosswalk - not resident point-level. Resident-population rates can overstate inner-city areas with large daytime worker/visitor populations.
• Hazard scores reflect designated regulatory overlay land, not insurance-grade fire/flood likelihood. An area may be partly overlaid yet most dwellings unaffected. In the default Melbourne build the SBO flood layer is currently unavailable from the planning API (LSIO only); other capitals use their active region hazard source labels, including Queensland SPP bushfire-prone mapping for Brisbane.
• GP/clinic count is OSM nodes only - by design, so widening the map-pin query never shifts the scored Health composite. This excludes clinics mapped as building polygons (ways), so it can undercount.
• Schools and GP use community-maintained OpenStreetMap; preschool uses Census counts. NDIS and pathology pins are sparsely tagged in OSM.
• Labour-force indicators are Census 2016 - older than income/rent (2021). GTFS is a static timetable export, not real-time service quality.
• 15-minute access distances are straight-line, not street-network; rivers, freeways and rail crossings are not modelled.

This is the canonical attribution page for the default Greater Melbournemanifest - per-dataset licences are in the manifest above. Map and profile surfaces for other capitals resolve source ids through their active region manifests so their crime and hazard licensors are not inherited from Melbourne. By licensor:

• Commonwealth of Australia - ABS (Census, Data by Region, SEIFA, ERP, Building Approvals; CC BY 4.0); AEC (2022 election results CSVs; CC BY 4.0); Bureau of Meteorology (solar-exposure climatology; BoM copyright - no CC notice verified on the product page).
• © State of Victoria - Department of Transport and Planning (Vicmap, VicPlan overlays, PTV GTFS, VIF2023, traffic volumes, activity centres), DEECA (fire history, coastal inundation, urban heat / tree canopy), Department of Education (school locations and zones) - CC BY 4.0; Crime Statistics Agency (recorded offences) - CC BY 3.0 AU.
• © EPA Victoria - AirWatch air quality and Beach Report (CC BY 4.0).
• © Overture Maps (OpenStreetMap + Microsoft, ODbL) - building footprints and heights for the sun/shade view.
• © OpenStreetMap contributors (ODbL) - schools, GP/clinics, transport fallback, everyday amenities, cycle infrastructure, post offices, pathology/NDIS points, noise/nuisance proxies.
• OpenFreeMap - basemap style and tiles, with © OpenStreetMap contributors.

This product charges for tooling and presentation, never for reselling the underlying open data, and retains attribution. The OSM-derived files are derivative databases under ODbL and are available from the public repository (DATA-LICENCE.md).

Spotted a data problem or want a dataset added? Use the Feedback button in the top bar - reports are reviewed against the next refresh and never folded directly into scores.