Spatial data science capability
Geospatial Careers is a market research project. The technical story matters because it shows how messy labour-market signals can become spatial evidence: collected responsibly, structured carefully, analysed geographically and prepared for public release.
This is the capability layer behind the research, not an operating guide. It explains the shape of the work while keeping operating detail, source handling and sensitive implementation choices private.
Evidence ledger
A dated account of findings that have passed the current publication checks. Each row keeps its sample, method and limitations attached to the claim.
Research snapshot generated 11 July 2026. Raw and governed counts are different evidence grains and should not be read interchangeably.
SQL
Measured
23,840 active canonical vacancy postings pass the governed cohort, from 45,187 raw rows. 9,167 have confident skill links (38.5%) and 7,465 have salary evidence (31.3%).
Read with care
DL
Measured
17,987 governed active vacancy postings have a versioned transformer embedding (75.5% coverage).
Read with care
ML
Measured
Across 500 valid governed job bootstraps, the modal community count appeared in 31.4% of resamples; median ARI was 0.55. The community labels remain withheld.
Read with care
What it demonstrates
The research is a working example of end-to-end spatial data science: research framing, data modelling, geospatial enrichment, quality review, interpretation, web publication and dataset release planning.
Public labour-market signals are organised into a research structure that separates source coverage, role language, geography, salary evidence and confidence.
The research moves from messy records to spatially useful aggregates: normalised roles, geographies, skill signals, salary context and quality flags.
The analysis asks where demand is visible, where skills are moving, which salary evidence is reliable, and which regions remain under-reported.
Public outputs are designed as citable research artefacts, not raw operational dumps. The planned release path favours open formats and clear provenance.
SDSC fit
The strongest version for Spatial Data Science Conference is a practical case study: how to turn labour market noise into a spatial evidence layer the community can inspect, challenge and improve.
Open research direction
The public release should be useful beyond this website. The right output is not every internal process, it is a documented, privacy-conscious research dataset that others can cite, inspect and build questions from.
Best fit for open geospatial data publication when the public artefact is a documented spatial dataset with stable URLs and standard access patterns.
Best fit for making spatial research assets discoverable as collections and items, with clear metadata, licence, extent and release lineage.
Best fit for a dataset card, Parquet preview, notebooks and machine-readable examples for people who discover research through data science workflows.
Community flywheel
The contribution loop should feel like joining a public research effort: read a finding, correct a gap, get release notes, share the improved evidence.
Lead with a specific market signal that practitioners recognise: a country gap, a salary caveat, a cloud-native skill shift or a regional demand pattern.
Ask the community to challenge weak coverage, add missing sources and explain what the data cannot see from the outside.
Turn corrections into versioned public notes, then package stronger aggregate data for open publication and citation.
Share missing geographies, salary evidence, tools, skill signals or reports. The more practitioners correct the evidence, the stronger the public release becomes.