Australian Taxation Office Legal Database — Machine-Readable

60,000+ ATO rulings, determinations, and private advice. Structured CSV. RAG-ready on day one.

The ATO Legal Database spans 15 years of Australian tax law — private advice, interpretative decisions, public rulings, and compliance guidelines. It's all public. It's all messy HTML. Getting it into a format an LLM can actually use takes weeks of data engineering.

This dataset skips that. Every document is one CSV row. Every logical section is its own column. Source URLs included.

---

This repository contains the full structured dataset and data dictionary. Full column reference is in data_dictionary.md. If this dataset saves you time, consider supporting the project:

☕ Buy Me a Coffee

---

Table of contents

• Direct download links
• What's in the dataset
• Why this is RAG-ready out of the box
• Schema highlights
• Quickstart
  • RAG chunking with LangChain
  • Filter query example
  • Recommended metadata fields
• Recommended chunking strategy
• Who this is for
• Support the project
• Data source

---

Direct download links

• edited_private_advice_all.zip
• ato_interpretative_decisions.zip
• public_all.zip
• practical_compliance_guidelines.zip

---

What's in the dataset

Dataset	Document Type	Est. Documents	Coverage
Edited Private Advice	Private rulings issued to individual taxpayers	~50,000	2011–present
ATO Interpretative Decisions	Binding ATO ID decisions	~10,000	All years
Public Rulings & Determinations	TR, TD, GSTR, CR, GSTD + 22 other type codes	~2,000	All years
Practical Compliance Guidelines	PCG, Draft PCG	~200	All years

Total size: ~400–500 MB uncompressed across all four datasets. Token estimate: ~80–120 million tokens (at ~1,500–2,000 tokens/document average).

Each zip contains a single UTF-8 CSV file. Files are updated automatically on weekdays when new ATO documents are detected.

---

Why this is RAG-ready out of the box

One row = one complete document. No reassembly, no PDF parsing, no HTML cleanup.

Section-level columns. Each logical section (Facts, Ruling, Reasons, Legislative References) is its own column — map directly to RAG chunks without regex.

Pipe-delimited sub-sections. Multi-part fields use | as a consistent separator (e.g. Q1: [text] | A1: [text]). Fine-grained chunking is trivial.

Source URLs included. Every row links back to the live ATO page. Citations are built in — no hallucination risk on provenance.

Status flags. Is_Archived, Is_Withdrawn, Is_Draft let you filter to current authoritative documents in one line. Critical for legal RAG accuracy.

Structured metadata. Date fields, type codes, legislative references, and subject references are all separate columns — use them as vector database metadata filters.

---

Schema highlights

Full column reference with types, nullability, and format notes: data_dictionary.md

Edited Private Advice — 11 columns Authorisation_Number · Date_of_Advice · Subject · Ruling · Ruling_Period · Date_Scheme_Commenced · Relevant_Facts_and_Circumstances · Relevant_Legislative_Provisions · Reasons_for_Decision · Is_Archived · Source_URL

ATO Interpretative Decisions — 16 columns ATO_ID_Number · Status · Title · Issue · Decision · Facts · Reasons_for_Decision · Date_of_Decision · Year_of_Income · Legislative_References · Related_Public_Rulings_and_Determinations · Related_ATO_Interpretative_Decisions · Subject_References · Business_Line · Is_Archived · Source_URL

Public Rulings & Determinations — 22 columns Document_Reference · Document_Type · Document_Type_Code · Title · Status · Date_of_Issue · Date_of_Effect · Date_of_Withdrawal · What_This_Ruling_Is_About · Class_of_Entities_or_Scheme · Ruling · Examples · Appendix_Explanation · Other_Appendices · Previous_Rulings · Related_Rulings · Legislative_References · Case_References · Subject_References · ATO_References · Is_Withdrawn · Source_URL

Practical Compliance Guidelines — 21 columns PCG_Number · Document_Type · Title · Status · Date_of_Issue · Date_of_Effect · Date_of_Withdrawal · Replaces · Related_Rulings_and_Determinations · Legislative_References · Summary · Purpose_and_Scope · Background · Compliance_Approach · Examples · Appendices · Other_Sections · Compendium_Reference · Is_Archived · Is_Draft · Source_URL

---

Quickstart

import pandas as pd

# Load a dataset
df = pd.read_csv("edited_private_advice_all.csv")

# Filter to current documents only
active = df[df["Is_Archived"] == False]

# Filter by topic
cgt = active[active["Subject"].str.contains("CGT|capital gains", case=False, na=False)]

print(f"{len(active):,} active documents, {len(cgt):,} CGT-related")

RAG chunking with LangChain

from langchain.schema import Document

def row_to_chunks(row):
    chunks = []
    metadata = {
        "source": row["Source_URL"],
        "authorisation_number": row["Authorisation_Number"],
        "date": row["Date_of_Advice"],
        "subject": row["Subject"],
        "is_archived": row["Is_Archived"],
    }
    # Ruling is the primary retrieval chunk
    if row["Ruling"]:
        chunks.append(Document(page_content=row["Ruling"], metadata={**metadata, "section": "ruling"}))
    # Reasons are the secondary chunk
    if row["Reasons_for_Decision"]:
        chunks.append(Document(page_content=row["Reasons_for_Decision"], metadata={**metadata, "section": "reasons"}))
    return chunks

active_docs = []
for _, row in active.iterrows():
    active_docs.extend(row_to_chunks(row))

Filter query example (Pinecone / Weaviate / Chroma)

Find current Tax Rulings about CGT main residence exemption issued after 2015:

filter: Document_Type_Code = "TR"
    AND Is_Withdrawn = False
    AND Date_of_Issue >= "2015"
semantic search on: Ruling + What_This_Ruling_Is_About

Recommended metadata fields to index in your vector store

Document_Type_Code · Status · Date_of_Issue · Year_of_Income · Business_Line · Is_Archived · Is_Withdrawn

---

Recommended chunking strategy

Column(s)	Chunk type	Suggested use
Subject / Title	Document title	Metadata / retrieval filter
Ruling	Operative text	Primary retrieval chunk
Reasons_for_Decision	Reasoning chain	Secondary retrieval chunk
Relevant_Facts_and_Circumstances / Facts	Context	Supporting chunk
Legislative_References	Structured references	Metadata filter
Source_URL	Citation	Always attach to answer
Is_Archived / Is_Withdrawn	Freshness flag	Pre-filter before search

Sub-sections within a field are delimited by | — split on this to create paragraph-level chunks without regex.

---

Who this is for

User	Use case
LegalTech / TaxTech startups	Building Australian tax AI without months of data engineering
Big 4 / mid-tier accounting firms	Feeding an internal AI the full history of ATO rulings
Law schools & tax researchers	The first structured, citable corpus of ATO private and public rulings
LLM fine-tuning teams	~100M tokens of expert Australian tax reasoning, already cleaned
Government / policy consultants	Bulk analysis of ATO trends across 15 years by topic, legislation, or date

---

Support the project

This dataset is free. If it saves you time, a coffee is appreciated:

☕ Buy Me a Coffee

---

Data source

All documents are sourced from the ATO Legal Database (ato.gov.au). Data is publicly available under ATO copyright. This dataset is a structured, machine-readable representation of that public information.