Recorded criminal incidents by council area and month, NSW
liveRecorded criminal incidents for each NSW local government area, by offence category and subcategory, for every month since January 1995. BOCSAR publishes it as a workbook with one column per month; each cell is one row here, keyed by council area, offence category, subcategory and month.
Also called: BOCSAR recorded crime statistics, NSW recorded criminal incidents by LGA, NSW crime data by local government area, BOCSAR RCI offence by month.
- Why is this one row per month when the publisher's file has one column per month?
- The publisher's workbook has a column for every month since January 1995 and gains three columns each quarter. Each cell here is one row, with its council area, offence category, subcategory and month, so a new release adds rows instead of columns and every version has the same fields. No count is changed.
- Are these counts of offences or of people?
- They are criminal incidents recorded by NSW Police in each month. One incident can involve more than one person and more than one offence category. The publisher's notes on the portal explain how incidents are counted.
- How often is it updated?
- BOCSAR releases this dataset every quarter. Recent months can change when police records are updated, so a later version can revise an earlier month.
Made for agents
Connect your AI agent
Any agent that connects to publicdata.au/mcp can find this dataset, read its 5 fields, count and filter across all 3,070,116 rows and diff its versions. Every answer names the version and carries BOCSAR's attribution. No key.
Then ask it: What does Recorded criminal incidents by council area and month, NSW hold, and what changed in the newest version?
claude mcp add --transport http publicdata https://publicdata.au/mcp One command. No key, no account, no sign-up.
Settings, then Connectors, then Add custom connector, and paste https://publicdata.au/mcp.
On the web and in the desktop app.
Open https://chatgpt.com/plugins/plugin_asdk_app_6abbc4cec1a08191b61eaccfe2091100 and connect the publicdata.au plugin.
Listed in ChatGPT's plugin directory. publicdata.au asks for no key and no account.
https://publicdata.au/mcp Streamable HTTP. Cursor, VS Code or any client that connects to a remote MCP server.
Using https://publicdata.au/llms.txt, how many people died on Queensland roads in 2025, by month? For an assistant that reads the web and has no connector.
The dashboard above, live. Click to filter, add panels, share a link or embed a frame.
Query APIFilter and count from a URLaggregate?group=offence_category&metric=sum.incidents&month=eq.2026-06-01
Incidents by offence where month is 2026-06-01: 14,726 Theft, 8,000 Against justice procedures, 5,983 Assault, 4,130 Intimidation, stalking and harassment.
Download it as CSV, Parquet, SQLite and 4 more formats
Pick a format and a version. CSV is picked first because it opens in the tools most offices have. The URL is yours to keep. A dated version keeps its content.
UTF-8 without a byte-order mark, RFC 4180 quoting, one header row using the normalised field names. Opens in Excel, Numbers and Sheets.
Columnar and typed. The smallest download and the fastest to query. Reads directly in DuckDB, pandas, Polars, Arrow and Spark. The provenance header is in the file metadata.
One database with a records table, a fields table and a publicdata metadata table. Open it with any SQLite client.
One DuckDB database with a records table of typed columns, a fields table and a publicdata metadata table. It attaches read-only over HTTPS, so DuckDB, Python and R can query it without downloading it.
The first line is the provenance header, then one JSON object per line. For streaming and for tools that read line by line.
Arrow IPC file (Feather v2), zstd compressed, with the provenance in the schema metadata. Memory-maps in pyarrow, Polars, Arrow JS and R without parsing.
The same CSV, gzipped, at about a tenth of the size. curl, pandas, DuckDB and R read it directly.
One JSON file per value of lga, 131 files. The example is by/lga/albury.json. Every file carries the same header and a partition note.
Excel is not offered because the table is over 1,048,575 rows, which is as many as a worksheet holds below its header row. JSON is not offered because the table is over 2,000,000 rows, too many for one document a reader parses at once.
In your own tools
Use it in Excel, R, Python and more
Excel and Power BI read the CSV from its address and refresh from it. The R and Python packages take any dataset on this site by its slug, so a new dataset needs no new release. The DuckDB file attaches read-only over HTTPS, and a query reads only the blocks it touches.
The dated URL in the code never changes. https://publicdata.au/d/nsw-recorded-crime-by-lga/latest/ redirects to the newest version.
https://publicdata.au/d/nsw-recorded-crime-by-lga/latest/data.csv In Excel choose Data, then From Web, and paste this address. Excel keeps it, so Refresh All reads the newest version. A sheet holds about a million rows; past that, load the query to the Data Model.
https://publicdata.au/d/nsw-recorded-crime-by-lga/latest/data.csv In Power BI Desktop choose Get data, then Web, paste this address and choose Anonymous when asked how to sign in. A scheduled refresh reads the same address, so the report follows each new version.
library(publicdataau)
df <- pd_read("nsw-recorded-crime-by-lga")
pd_attribution(df) install.packages("publicdataau"). pd_read() fetches the version's Parquet file; pd_rows() and pd_aggregate() ask the query API instead, and pd_connect() attaches the DuckDB file.
import publicdata_au as pd_au
df = pd_au.read("nsw-recorded-crime-by-lga")
df.attrs["publicdata"]["attribution"] pip install "publicdata-au[pandas]". read() fetches the version's Parquet file; rows() and aggregate() ask the query API, and connect() attaches the DuckDB file.
INSTALL httpfs; LOAD httpfs;
ATTACH 'https://publicdata.au/d/nsw-recorded-crime-by-lga/v/2026-09-16/data.duckdb' AS nsw_recorded_crime_by_lga (READ_ONLY);
SELECT lga, count(*) FROM nsw_recorded_crime_by_lga.records GROUP BY 1 ORDER BY 2 DESC; The DuckDB file attaches read-only over HTTPS and only the blocks a query touches are read. Parquet works the same way: FROM read_parquet(url).
const res = await fetch("https://publicdata.au/api/v1/datasets/nsw-recorded-crime-by-lga/aggregate?group=offence_category&metric=sum.incidents&month=eq.2026-06-01");
const { rows, publicdata } = await res.json();
console.log(rows, publicdata.attribution); The query API answers a page on any site as well as Node, with no key. It returns the rows and the attribution the licence asks for.
What is in it
Column names are made snake_case and the publisher's original header is kept beside each one. Blank cells are null. Nothing is added, removed, ranked or summarised.
| Field | Type | Publisher's header | Note |
|---|---|---|---|
| lga | string | LGA | Local government area as BOCSAR names it. |
| offence_category | string | Offence category | |
| subcategory | string | Subcategory | Blank where the category has no subcategories. An asterisk on a name is BOCSAR's own. |
| month | date | (column header) | First day of the month the incidents were recorded in. |
| incidents | integer | (cell) | Criminal incidents recorded by NSW Police in that month. |
Download this table as a data dictionary (Excel), or read it as a Frictionless Table Schema at schema.json. The key is lga, offence_category, subcategory, month.
A sample of 10 rows
The council area and subcategory with the most incidents in each of ten offence categories in the newest month, largest first. A blank cell is shown as null.
| lga | offence_category | subcategory | month | incidents |
|---|---|---|---|---|
| Liverpool | Transport regulatory offences | null | 2026-06-01 | 543 |
| Central Coast | Against justice procedures | Breach bail conditions | 2026-06-01 | 314 |
| Sydney | Assault | Non-domestic violence related assault | 2026-06-01 | 257 |
| Central Coast | Intimidation, stalking and harassment | null | 2026-06-01 | 254 |
| Sydney | Theft | Other theft | 2026-06-01 | 252 |
| Blacktown | Malicious damage to property | null | 2026-06-01 | 208 |
| Sydney | Drug offences | Possession and/or use of cannabis | 2026-06-01 | 137 |
| In Custody | Sexual offences | Sexual assault | 2026-06-01 | 118 |
| Sydney | Liquor offences | null | 2026-06-01 | 106 |
| Sydney | Disorderly conduct | Trespass | 2026-06-01 | 90 |
The first row as JSON
As it appears in data.ndjson.
GET https://publicdata.au/d/nsw-recorded-crime-by-lga/latest/data.ndjson
{
"lga": "Albury",
"offence_category": "Homicide",
"subcategory": "Murder *",
"month": "1995-01-01",
"incidents": 0
}
Query it without downloading
The query API returns only the rows you ask for, as JSON, NDJSON or CSV, from the same typed rows as the files. It needs no key. Build a query here and run it, then copy the URL or the code into your own work.
How filters work
Each filter is field=operator.value in the query string, and every filter must match. A number or date field compares as a number or date, and a text field compares as text. Values are case-sensitive, except that like and ilike ignore case in the ASCII letters A to Z. Prefix not. to negate a filter, as not.eq.value.
| Operator | What it matches |
|---|---|
| eq.value | Equal to the value. |
| neq.value | Not equal to the value. A blank cell does not match. |
| gt.value | Greater than the value. |
| gte.value | Greater than or equal to the value. |
| lt.value | Less than the value. |
| lte.value | Less than or equal to the value. |
| like.*text* | Matches a pattern where * stands for any run of characters, ignoring case in the ASCII letters A to Z. |
| ilike.*text* | The same as like. Both ignore case in ASCII letters only, so É and é differ. |
| in.(a,b,c) | Equal to any value in the list. A value cannot contain a comma. |
| is.null | Blank in the source, or suppressed by the publisher. |
| Parameter | What it does |
|---|---|
| select | Fields to return, comma-separated. Every field when absent. |
| order | field.asc or field.desc, comma-separated. When absent, rows come in the order of the file they are read from. For the newest versions that is the version's own data.parquet, which keeps the publisher's order unless the dataset sorted it. Any other version is read from the Parquet file the answer names in file, and when that file is sorted the answer names its sort in order, which ties follow too. |
| limit | Rows per page, 1 to 10,000. 100 when absent. |
| offset | Rows to skip. The next URL in each answer sets it for you. |
| group | On aggregate, fields to group by, comma-separated. |
| metric | On aggregate, count, sum.field, avg.field, min.field or max.field, comma-separated. count when absent. |
| format | json, ndjson or csv. JSON carries the provenance header, and the others carry it in response headers. |
Without a version the API answers from the newest version, and that answer changes when the publisher releases again. Put versions/<date>/ before rows or aggregate for an answer from that version alone. Every version with a table can be queried, and /api/v1/datasets/<slug>/versions lists them. A query on a version the database does not hold is read from its Parquet file, and one that would read too much of it is refused with DuckDB SQL that answers it from the file. The 1 newest version is loaded; versions lists them.
The API allows 60 requests in 10 seconds from one address. Above that it answers 429 for 10 seconds with a Retry-After header, a RateLimit-Policy header and a JSON body that gives the limit. Every API answer carries the same RateLimit-Policy. A client should wait the Retry-After seconds, or read the files, which have no limit. openapi.json describes this dataset's query paths for client generators and agents.
https://publicdata.au/mcp is a remote MCP server over Streamable HTTP with the same tools, defined in the same place, so a tool added to the pages is added here too. It needs no key and no account. Its row tools and the query API answer every version through one query path, so a tool's answer and the query URL it cites give the same rows. A query that would read too much of a file is refused with DuckDB SQL that answers it from the file. It also has search and fetch, the two tools ChatGPT's deep research and company knowledge call by name, which run search_datasets and read a dataset's record in the shapes those features expect. The row tools are held to the query API's limit of 60 queries in 10 seconds from one address, and each call is two queries because it also counts the matching rows. Add it to Claude Code with claude mcp add --transport http publicdata https://publicdata.au/mcp, add the URL in Claude as a custom connector, connect the publicdata.au plugin in ChatGPT, or give it to any client that connects to remote MCP servers. Each dataset with a table is also a resource at https://publicdata.au/d/<slug>/fields.json, which lists its fields with their types, the publisher's descriptions, their ranges and the values they hold, so an agent can read a dataset's shape before it writes a query.
Smaller files
The same rows split by a field, one JSON file each. Each file carries the same provenance header.
By lga (131 files)
by/lga/index.json lists every file with its row count.
What changed
One version for every release the publisher has made since this site started following the dataset. The date is the day the file changed on the portal.
- 2026-09-16 3,070,116 rows5 fieldsxlsx9b13b9e87887
versions.json · changes.json · history.tar.zst (1.3 MB, every version's Parquet and manifest)
Questions
How do I download Recorded criminal incidents by council area and month as a CSV file?
Open https://publicdata.au/d/nsw-recorded-crime-by-lga/latest/data.csv. It redirects to the newest dated version, which is https://publicdata.au/d/nsw-recorded-crime-by-lga/v/2026-09-16/data.csv today. The same path serves Parquet, SQLite, DuckDB, NDJSON, Arrow. A dated URL never changes, so use it when the file must stay the same.
Can I open Recorded criminal incidents by council area and month in Excel?
Not as a workbook. Excel is not offered because the table is over 1,048,575 rows, which is as many as a worksheet holds below its header row. Load the CSV or the Parquet file with Power Query, or take one partition file at a time. https://publicdata.au/d/nsw-recorded-crime-by-lga/v/2026-09-16/data.csv.gz is the CSV at about a tenth of the size.
What years does Recorded criminal incidents by council area and month cover?
The current version covers 1995 to 2025. Each release from BOCSAR becomes a new dated version here, and earlier versions stay online.
How often is Recorded criminal incidents by council area and month updated?
BOCSAR releases it quarterly. This site checks the portal every week and adds a dated version when the file changes.
Can I use Recorded criminal incidents by council area and month commercially?
Yes. CC BY 4.0 allows commercial use, redistribution and derived works as long as the attribution is kept. The attribution string is in this page's side column and inside every file.
Is this the official source for Recorded criminal incidents by council area and month?
No. The publisher is NSW Bureau of Crime Statistics and Research, and its page is https://data.nsw.gov.au/data/dataset/nsw-crime-and-policing-statistics. National Digital converted the publisher's file to other formats. It renamed and typed the columns and kept only those the schema lists. It turned column headers into rows, one row for each value. National Digital made these changes, and the publisher is not responsible for them. The original sits beside every version as source.xlsm with its SHA-256, so the two can be compared.
How do I get only the rows for one lga?
Every version has one JSON file per value of lga, 131 files in the current version, listed with row counts at https://publicdata.au/d/nsw-recorded-crime-by-lga/v/2026-09-16/by/lga/index.json. For example https://publicdata.au/d/nsw-recorded-crime-by-lga/v/2026-09-16/by/lga/albury.json holds the 23,436 rows where lga is Albury.
What this site did to the data. Cells were typed, headers were renamed and the encoding was made UTF-8. Rows were left alone. The publisher's file sits beside every version as source.xlsm so the change can be checked.
The licence requires the publisher's attribution. We ask that you also name publicdata.au and link to the version you used.
NSW Bureau of Crime Statistics and Research, Recorded criminal incidents by month by LGA, https://data.nsw.gov.au/data/dataset/nsw-crime-and-policing-statistics, accessed 28 September 2026, licensed under CC BY 4.0. Serialised and versioned by National Digital at publicdata.au, https://publicdata.au/d/nsw-recorded-crime-by-lga/v/2026-09-16/
The attribution alone is inside every file. Harvard is the author-date form a report or a brief asks for. Paste the HTML on a web page so the links carry.
NSW Bureau of Crime Statistics and Research states the licence, CC BY 4.0, on its own page, which was read and recorded here on 28 September 2026.
Every fetch reads the licence again and stops when it has changed. The fetch that made version 2026-09-16 read it on 28 September 2026.
- Hugging Faceload_dataset("National-Digital/nsw-recorded-crime-by-lga")
- Kagglewith a starter notebook
- ZenodoDOI 10.5281/zenodo.23076818
Each copy carries the same attribution and links back to this page.
- datapackage.jsonFrictionless data package pointing at the latest version
- schema.jsonTable Schema for every field
- schema.xlsxthe same field list as an Excel data dictionary
- versions.jsonevery version with date, rows, fields and hash
- changes.jsondiff for every consecutive pair
- manifest.jsonwhere the bytes came from and when
- Query APIfiltered rows and counts without a download
- openapi.jsonOpenAPI 3.1 for this dataset's query paths
- explore/charts and dashboards in the browser, shareable and embeddable
- index.mdthis page as Markdown
This page carries a schema.org Dataset record and is listed in /.well-known/ard.json and /catalog.json.
NSW Bureau of Crime Statistics and Research publishes this table on bocsarblob.blob.core.windows.net. Questions about the content go to the publisher. Questions about the files here go to corrections.
Fetched 28 September 2026. Source SHA-256 9b13b9e878875b01…
This page is built from register/nsw-recorded-crime-by-lga.yaml on GitHub. Anyone can propose a fix to it, such as a clearer description or a field the publisher renamed, as a pull request.
Report a file that differs from the publisher's · How to contribute