How machine-readable are UK company accounts, really? This benchmark parses a full day of Companies House accounts filings (Inline XBRL) and measures what fraction of real filings expose each core accounting concept as a structured, machine-readable fact, no OCR or layout heuristics required.


Filings parsed	8,856
Distinct taxonomy concepts seen	899
Median tagged facts per filing	49
Source	Companies House daily accounts bulk (iXBRL)
Licence	Open Government Licence v3.0

The headline

Core balance-sheet concepts are exposed as structured facts in the large majority of filings: Equity in 97.2%, Net assets in 86.6%, Creditors in 75.6%, Total assets less current liabilities in 73.3%. This is the quantity any claim that "structured data beats PDF for AI-driven reporting" ultimately rests on: not that the data could be structured, but that in the real filed population it already is, at a measurable rate, per concept.

Files

data/benchmark.json — concept-coverage table + core-concept exposure
data/filings.jsonl — per-filing concept counts and core concepts present
scripts/extract.py — reproducible parser over the daily bulk archive

Reproduce

curl -L -o data/accounts.zip \
  "https://download.companieshouse.gov.uk/Accounts_Bulk_Data-2026-07-08.zip"
python3 scripts/extract.py

Contains public sector information from Companies House licensed under the Open Government Licence v3.0. Independent, self-initiated open research by Tesseract Academy.

Downloads last month: 24

Total file size:

2.18 MB