DIY Accounting Spreadsheets

The diya-gl format

A year of accounts in two text files, and the mapping, the checks and the evidence behind them.

What diya-gl is

diya-gl is the file format behind DIY Accounting's books pages, its command line tools and the spreadsheets it generates. A whole year of a small business's accounts is two text files in a zip. book.toml says who the business is, what period the books cover and what the chart of accounts holds. lines.jsonl holds one transaction per line. A sole trader's year runs to about 15 KB.

It is for people who want to own their accounting data outright. You can read it in a text editor, keep it in git and see what changed, and recalculate it in a browser or on a command line without Excel and without a server. This page is the specification: the fields, the figures they compute, the checks they have to pass, the zip they travel in, and the evidence that the figures agree with the spreadsheets we have shipped for twenty years.

The declared subset

Every field name in the two schemas is taken from the XBRL Global Ledger Taxonomy Framework 2015, the XBRL Standards Board's recommendation of 25 March 2015. The tables below name the element each field comes from.

That is a lineage and a published mapping, and this page claims nothing past what the tables show. XBRL GL runs no conformance programme and issues no certificate for a data format. A diya-gl file is TOML and JSON Lines; it is not an XBRL instance document, and no XBRL processor reads it.

The format names 48 elements, from four of the framework's modules: gl-cor (core), gl-bus (business), gl-muc (multicurrency) and gl-taf (tax audit file). No other element of the framework is named in either schema. The framework's nested tuples are flattened: an entry header and its detail are one line, and gl-cor:account becomes the pair accountMainID and accountMainDescription. Where a UK small business needs something the framework has no element for, the schema declares a diya-gl extension: a diya-gl: prefix on a field, or a whole table marked as one. There are 135 extension fields, and the tables below mark every one.

The schemas themselves are published beside this page: diya-gl-book-v2.schema.json and diya-gl-lines-v2.schema.json, both JSON Schema draft 2020-12. A longer commentary on the mapping is in diya-gl-docs.md.

Document info

The two tables at the head of book.toml: what period the file covers, and whose business it is.

Accounts and registers

The chart of accounts, the registers a year-end needs (assets, hire purchase, members, charges) and the tax rates the period was calculated at. The rate tables hold HMRC's published figures for the year; the framework has no element for a rate table, so those rows name none.

One account

Each four-digit account code in a section holds one of these.

Transaction lines

One JSON object per line of lines.jsonl. Four fields are required on every line, and a journal line must also say which side it posts.

Computed figures and their SA103S boxes

The three sole trader products fill the short self-employment pages, SA103S. The table gives every box on the 2026 form. figure means the product computes the box. derived means the page works it out from the profit and loss account. blank means the format carries nothing for that box, so it prints empty. Box numbers follow the 2026 form; nothing here is the HMRC document itself.

The last column is HMRC's own field name for the box in the Making Tax Digital Self Employment Business API, from HMRC's published SA103 mapping (sa103f_mapping_v3.csv, read 2026-09-04). 15 of the 34 boxes have one.

Self Employed books fill the full pages, SA103F, as well, and both they and Limited Company books fill a VAT return. Limited Company books file a CT600 and micro-entity accounts rather than SA103S. Every one of those views is on the product's own books page, and the figures behind them are on its reconciliation scorecard.

The check catalogue

A book is read in three stages. The two JSON Schemas validate book.toml and every line of lines.jsonl. The validator then refuses a file whose lines reach an account the book does not declare, or name a hire purchase agreement, an asset or a member the registers do not hold. Last, the 27 book checks below run over the data itself.

A check passes or fails, and a failure names the entries that caused it; 13 of the rules are checks. A warning passes or warns and never blocks a save; 14 of them are warnings. Which rules a book runs depends on the product it declares. Every result is written to bookchecks.json in the zip.

Beside these, the engine runs its reconciliation checks: one for every figure the workbook and the engine both compute, compared to the penny. Those depend on the books in hand rather than on the format, and every one of them is published on the scorecards below.

The zip layout

The zip has no directories. Its entries are:

Two writes of the same book produce the same bytes, on any machine and in any of the three surfaces (the browser page, the command line tool and the MCP server). The rules that make that hold:

• Lines sort by posting date, then source journal, account, entry number, line number, document reference and amount.
• Fields within a line follow the order the lines schema declares them in, and a field with no value is left out.
• book.toml follows the book schema's table and key order, and every array of tables sorts by its own id field.
• Money carries exactly two decimal places and a rate exactly four, trailing zeroes included.
• Dates are ISO 8601, YYYY-MM-DD, with no time.
• Every zip entry is stamped 1 January 1980, so the timestamp never varies between two writes.

The same book and lines also travel as one JSON file, with a { "format": "diya-gl-books", "version": 1, "product": ..., "book": ..., "lines": [...] } envelope, and as that JSON zipped. The books pages read all three, plus a workbook and a package zip.

Versioning and provenance

The format is diya-gl-books version 1. The two schemas are at v2, and the version is part of each schema's $id URL. A v1 existed but nothing outside this repository ever read it, so v2 replaced it in place.

Five stamps sit in the book's document info and at the head of report.json. They answer the question a file raises years later. Recalculate a 2026 book in 2030 and it either reproduces its report.json byte for byte, or names the stamp that differs.

Reconciliation evidence

Every figure this format computes is checked against the same figure in the spreadsheet. For each product, CI generates the package from a set of books, recalculates the workbook in LibreOffice, computes the same figures from the diya-gl data alone, and compares the two to the penny. Each scorecard carries the whole run: its status, the checks, the input transactions, screenshots of the recalculated sheets, the accounting statements and the tax review.

All reconciliation reports

Getting the tools

The books pages read and write the format in your browser. Nothing you load leaves the machine.

• Basic Sole Trader books
• Taxi Driver books
• Self Employed books
• Limited Company books

The same engine runs on the command line and as an MCP server, published as the npm package @diy-accounting-uk/diya-gl. It is not on npm yet; this page will carry the install line when it is.

The spreadsheets themselves stay free to download on a donation basis. Download a package, or chip in if the tools earn it.