Minimum viable dataset

# Minimum Viable Habitability Dataset v1

## Purpose

The objective is to test whether Habitability can be measured credibly with a compact, source-diverse dataset before building a large territorial data lake.

This dataset is intentionally construct-first. We select the metrics that matter, then use the strongest practical source for each metric. Eurostat is one source among several, not the framework.

## Design principles

1. Prefer observed or administrative data over opaque precomputed indicators.
2. Keep derived Habitability metrics reproducible from their raw inputs.
3. Preserve provenance, geography, time period, source methodology and quality flags.
4. Do not silently mix city proper, municipality, FUA, region or grid data.
5. Do not force symmetry across H-D-A-R if the evidence is weak.
6. Do not create a composite score until coverage and source quality are validated.
7. Use Grenoble as the first end-to-end test territory, then validate across European FUAs.

## Minimum Viable Dataset

The first target set is 18 metrics.

| # | Construct | Metric | Role | Preferred source family | Evidence class | Priority |
|---|---|---|---|---|---|---|
| 1 | Economic opportunity | Unemployment rate | H | OECD / national statistics | L1 official statistic | P1 |
| 2 | Income capacity | Median disposable income per consumption unit | H/A input | national statistics / OECD | L1 official statistic | P1 |
| 3 | Housing affordability | Median dwelling price or rent relative to disposable income | A | national transaction/rent data + income | L2 derived | P1 |
| 4 | Housing space | Living area per person or crowding proxy | H | national housing statistics / Eurostat fallback | L1/L2 | P2 |
| 5 | Premature mortality | Deaths under 65 per population, preferably age-standardised when available | H | national health statistics / OECD | L2 derived | P1 |
| 6 | Infant mortality | Infant mortality rate | H | national health statistics / OECD / Eurostat fallback | L1 | P2 |
| 7 | Personal safety | Homicide/violent-death rate | H | police/justice/public-health statistics | L2 derived | P2 |
| 8 | Road safety | Road deaths per population | H | national road-safety statistics / Eurostat | L2 derived | P2 |
| 9 | Air quality | Population-weighted PM2.5 exposure | H/R | EEA + population grid | L2 derived from measured/modelled data | P1 |
| 10 | Heat exposure | Annual days above heat threshold / heatwave exposure | R/H | Copernicus ERA5/Land | L2 derived | P1 |
| 11 | Flood exposure | Share of population/buildings in flood-risk zones | R | Copernicus CEMS / national flood maps + GHSL | L2 derived | P1 |
| 12 | Water stress | Drought/water-stress indicator relevant to local supply | R | Copernicus / EEA / national hydrology | L2/L3 | P2 |
| 13 | Green/natural access | Population-weighted access to green/natural space | H/D | Copernicus Land + OSM | L2 derived | P2 |
| 14 | Healthcare accessibility | Population-weighted travel time to primary/emergency care | H/R | national directories + OSM/network data | L2 derived | P1 |
| 15 | Essential services accessibility | Access to grocery, school and everyday services | H | OSM + official directories | L2 derived | P2 |
| 16 | Public transport accessibility | Access to frequent public transport / jobs by transit | H/D | GTFS/official transit + network data | L2 derived | P2 |
| 17 | Revealed desirability | Net population/migration trajectory | D | OECD / national statistics | L1 official statistic | P2 |
| 18 | Built-up intensity / land pressure | Built-up area or imperviousness per resident and change over time | Context/R | GHSL / Copernicus Land | L2 derived | P2 |

## What is deliberately excluded from v1

The first version does not include subjective happiness, generic city rankings, arbitrary amenity counts, tourism popularity, social-media signals, prestige indicators or opaque commercial scores. These may later be useful as explanatory or desirability signals, but they should not define the first measurement engine.

## Source waves

### Wave 1 — Socio-economic backbone

- OECD Cities/FUAs
- National statistical institutes
- France pilot: INSEE Filosofi and relevant employment/demographic datasets

Goal: establish unemployment, income, population trajectory and the denominator backbone on a defensible geography.

### Wave 2 — Spatial/environmental backbone

- Copernicus Climate Data Store / ERA5
- Copernicus Land Monitoring
- GHSL
- EEA air-quality data

Goal: create metrics that Habitability derives itself rather than inheriting from city-statistics tables.

### Wave 3 — Affordability and accessibility

- National property transaction/rent sources
- France pilot: DVF + INSEE income
- OSM / official service directories
- GTFS / transport authority feeds

Goal: add the strongest missing H-D-A-R components, especially Affordability and access to services.

## Minimum launch condition

The data architecture is ready for normalization testing when Grenoble and at least a meaningful cross-city peer set have:

- at least 12 of the 18 metrics,
- at least one strong metric in H, A and R,
- explicit provenance and geography for every observation,
- quality status for every source/metric combination,
- sufficient history to test trajectories where the metric is temporal.

D may remain a thinner revealed-preference layer in v1.

## Decision gate

Do not resume composite normalization merely because data exist. Resume only when the cross-source quality audit confirms that the reference distributions are comparing like with like.