digi.raise 2026 — starter kit

The QR-linked companion to “Mapping the Geography of Giving”. Four self-contained Claude Code examples. Each one produces both an analysis (answer.md) and an interactive MapLibre map (answer.html) from your donor data.

Nothing here sends donor data to a third party. Reference data is fetched live from public ABS endpoints (no auth required) — nothing ABS-related is bundled.

The four examples

#	Folder	Question	What the map shows
01	`examples/01_donor_fingerprint/`	What’s the demographic profile of my donor file?	Choropleth of donor postcodes coloured by donor count.
02	`examples/02_spot_the_bias/`	Which postcodes look like our donors but we’ve never asked?	Three-layer map: donors vs campaigned vs unasked look-alikes (the highest-leverage view in the kit).
03	`examples/03_score_next_campaign/`	Rank a list of postcodes I’m about to mail.	Targets coloured by similarity score 0-100, with the recommended cut-line outlined in gold.
04	`examples/04_donor_cluster_locator/`	Where do my donors cluster?	Two side-by-side maps: top by count vs top by rate per 1,000 adults. The delta is the story.

Each folder contains:

prompt.md — full spec (analysis + map) — paste into any LLM that can fetch URLs
CLAUDE.md — auto-loaded by Claude Code on session start
README.md — what the folder is for
expected_output.md — what a good answer.md looks like (no answer.html reference — the spec lives in the prompt)
the sample CSV(s) needed to run it

Running an example in Claude Code

cd starter/examples/01_donor_fingerprint
claude
# inside Claude Code:
"Run CLAUDE.md."

Claude reads the prompt, fetches boundary geometry + ABS Census features from the public endpoints, writes the analysis to answer.md, and the interactive map to answer.html. Open the HTML with open answer.html to view.

Map stack each prompt requires

All four prompts mandate the same stack so the outputs look consistent:

MapLibre GL JS (loaded from CDN)
CARTO light_all raster basemap, 4-domain sharded, no API key: https://{a-d}.basemaps.cartocdn.com/light_all/{z}/{x}/{y}@2x.png
Geometry fetched from the ABS ArcGIS REST POA endpoint at runtime
Embedded GeoJSON in the HTML so the file works after a single page-load

Public ABS endpoints the prompts use

No registration, no API keys.

Source	URL pattern	What you get
ABS ArcGIS REST — POA boundaries	`https://geo.abs.gov.au/arcgis/rest/services/ASGS2021/POA/FeatureServer/0/query?where=poa_code_2021%20IN%20(<list>)&outFields=poa_code_2021,poa_name_2021&outSR=4326&f=geojson`	GeoJSON polygons for any Australian postcode
ABS Data API — Census 2021	`https://data.api.abs.gov.au/rest/data/ABS,C21_T01_POA,1.0.0/...?format=jsondata`	SDMX-format demographic features per POA (income, age, religion, housing, etc.)
AEC 2022 booth-level TPP	`https://results.aec.gov.au/27966/Website/Downloads/`	Federal voting CSVs (optional — only if voting features come up)

If the SDMX endpoint proves too painful at runtime, every prompt lets Claude fall back to general knowledge of Australian postcode demographics — provided that fallback is disclosed at the top of answer.md.

Plot a Lorenz curve of your own giving

The Sydney slide showed 9% of postcodes generate 50% of the dollars. Run the same chart for your donor file (local, no network):

python3 scripts/lorenz_curve.py your_donors.csv

Produces lorenz.png in the current directory. Works on any CSV with postcode and gift_amount columns.

Privacy

Prompts run inside your LLM session — your donor data goes to Claude / OpenAI only.
The Lorenz script reads a CSV from disk and writes a PNG to disk. No network calls.
If you’re nervous about sending postcodes to an LLM, swap them for SA1 IDs first using the ABS boundary endpoint.

Regenerating sample data

python3 scripts/_gen_sample_data.py

Writes:

examples/01_donor_fingerprint/sample_donors.csv
examples/02_spot_the_bias/sample_donors.csv + sample_campaign_list.csv
examples/03_score_next_campaign/sample_donors.csv + sample_campaign_targets.csv
examples/04_donor_cluster_locator/sample_donors.csv