Common Ground

A non-partisan, side-by-side comparison of Australian federal party policy positions (Labor, Coalition, Greens, One Nation) — a matrix view, an overlap/agreement view, and a filter for how a policy affects you personally.

Live: https://spheric.github.io/common-ground/ — rebuilt automatically from data/dataset.json on every push.

The accuracy promise

Every position shown is either cited or explicitly marked as unknown. Nothing is guessed:

• Every position that isn't "No clear position" links to at least one source that was independently fetched and checked to confirm it actually supports the stated claim.
• Where no clear, verifiable position exists, the app says so — "No clear position" — instead of inferring one from a party's general ideology or past behaviour.
• Summaries are paraphrases of the cited sources, not the sources themselves. Always check the link before repeating a claim.

Principles, in priority order: accuracy → neutrality → accessibility → ease of use → beauty.

Quickstart

node scripts/build.mjs
open dist/index.html

That's it — no install step, no dependencies. dist/index.html is a single self-contained file; it also opens fine from a file:// URL. If data/dataset.json doesn't exist yet, the build falls back to data/dataset.sample.json (a small fixture) so the app always has something to render.

dist/artifact.html is a body-only variant of the same build, for publishing to claude.ai as an Artifact.

Updating the data

Data isn't hand-edited or scraped — it's produced by an agent-native ingestion pipeline: a documented Claude Code skill that fetches a source, extracts positions, verifies every source URL by fetching it, merges the result into data/dataset.json, validates, and rebuilds. See ingest/README.md for the full flow, ingest/sources.md for where to look per party, and the skill itself at .claude/skills/ingest-policy-source/SKILL.md.

Stated positions are one half of the picture; the other half is parliamentary voting records from They Vote For You (OpenAustralia Foundation, Open Data Commons ODbL). A hand-verified mapping in data/tvfy/mapping.json links 39 of the 63 issues to TVFY policy propositions, and scripts/tvfy/ aggregates how current MPs and senators actually voted, per party — shown beside each party's stated position in the app. See the "Parliamentary voting records" section of ingest/README.md.

To check the current dataset without changing anything:

node scripts/validate.mjs

Repo map

data/dataset.json                # the real dataset (produced by the ingestion pipeline; may not exist yet)
data/dataset.sample.json         # small fixture, shape-identical — used for dev when dataset.json is absent
data/schema/dataset.schema.json  # JSON Schema (draft 2020-12) for the dataset
docs/app-spec.md                 # the build spec — source of truth for data model, app, and pipeline
ingest/README.md                 # how ingestion works
ingest/sources.md                # per-party source registry with reliability notes
ingest/prompts/extract.md        # the extraction/normalisation prompt template
.claude/skills/ingest-policy-source/SKILL.md  # the Claude Code skill that runs the pipeline
scripts/validate.mjs             # dataset validator (zero dependencies)
scripts/build.mjs                # single-file bundler (zero dependencies)
web/index.html, web/style.css, web/app.js  # the app source
dist/index.html                  # built standalone file — open directly in a browser
dist/artifact.html               # built body-only variant — for claude.ai artifact publishing

Data model, briefly

A dataset has meta (title, subtitle, as-of date, disclaimer, methodology), the four parties, the four fixed stances (For / Against / Mixed / No clear position), a fixed vocabulary of impact_tags, and topics[] each containing issues[]. Every issue has a neutral yes/no question and exactly one position per party: a stance, a ≤40-word summary, impacts (tag ids), an impact_note ("what this means for you"), confidence, verified status, and sources[] — which may be empty only when the stance is no_position. Full shape: data/schema/dataset.schema.json; canonical example: data/dataset.sample.json.

Disclaimer

Independent and non-partisan. Positions are paraphrased from cited public sources and may change — always check the linked source. Not affiliated with any political party.