Methodology, and what this cannot tell you

How 23,878 offences were found, what was measured, and the questions this dataset is not able to answer.

The corpus

Every statutory instrument published by legislation.gov.uk across nine types — UK, Welsh, Scottish and Northern Irish instruments, Orders in Council, ministerial orders, Church instruments and the historical statutory rules and orders — from 1823 to the present. That is 156,107 documents.

The text read is the text as made

This is the decision that matters most, and it is counter-intuitive. legislation.gov.uk publishes two texts for an instrument: the text as originally made, and a revised text kept up to date with amendments.

In the revised text, a repealed provision is not marked repealed — it is gone. Regulation 9 of S.I. 2020/350, which created three offences, is simply absent from the revised version. Reading the revised text would therefore silently lose every offence that has since been repealed, which is a large share of the ones a practitioner most needs.

So the as-made text is read, and current status is recovered separately: from other instruments' revocation and saving provisions, and from the amendment-history apparatus in the revised text of the instruments that do create offences.

Revoked offences are kept

An offence does not stop mattering when the instrument creating it is revoked. The standard drafting keeps it prosecutable for conduct before the revocation:

“Notwithstanding the revocation of the first Regulations, they continue in force in relation to any offence committed under the first Regulations before these Regulations came into force.”

— reg. 2(2), Health Protection (Coronavirus, Restrictions) (England) Regulations 2020

Every offence therefore carries a status: in force, revoked, revoked with a saving, or not determined. Where a saving was found, the cut-off date and the saving provision are recorded.

How an offence is recognised

Detection separates the words that create an offence from the many that merely mention one — fixed penalty machinery, prosecution provisions, body-corporate liability extensions, savings, definitions and cross-references. Fault classification follows the standard distinction between strict liability and objective fault, and treats “without reasonable excuse” as an element of the offence rather than a fault word, following R v Charles [2009] EWCA Crim 1570.

Against a hand-checked set of labelled provisions, extraction scores 1.00 precision and 1.00 recall. That set is small, and a perfect score on it means the rules do not fail on the cases someone thought to check — not that they never fail.

NULL means “not determined”

Never zero, never inferred. Where a page says a value was not determined, that is a statement about the state of knowledge, not a statement that the value is nil. This matters most for penalties and for prosecution figures: no figure is not the same as no prosecutions.

The examples are generated

The “what would breach this” section on each offence page is written by machine from the provision's own words — its conduct clause, the person it binds, and the instrument's subject. Where the provision's own words could be quoted, they are, unaltered. The rest is illustration, and is labelled as illustration. It is not a statement that particular conduct is or is not caught.

What this cannot tell you

Whether an offence is ever prosecuted

Nothing in the statute book says how often an offence is charged, and offences created by statutory instrument very largely do not have their own code in the Ministry of Justice's Outcomes by Offence data tool. Where a code could be matched, it is shown, with the warning that the code usually covers more than one provision.

Whether a court has read a provision differently

The classifier reads words, not cases. Where the House of Lords read down a reverse burden — as it did to s.28 of the Misuse of Drugs Act 1971 in R v Lambert — this dataset does not know.

Offences in Acts of Parliament

Out of scope by design. Most of the well-known strict liability offences — speeding, driving without insurance, possession of a prohibited firearm, TV licence evasion — are in primary legislation.

Instruments published only as a PDF

A large share of the corpus has no machine-readable text, peaking at 42% of the 2010s, because whole modern families are published that way: air navigation flying restrictions and temporary traffic orders. Those were read, and they contain no offence-creating language at all — they are purely prohibitory, and take their offence from a parent Act or Order. They are kept out of this site's counts because seventeen thousand ephemeral local orders alongside the Environmental Permitting Regulations would make every proportion here meaningless.

How Parliament handled an instrument

Procedure, motions and votes come from Parliament's own Statutory Instruments service, and recorded divisions from the two Houses' votes services. That source is a live business system rather than a historical archive: it begins in May 2017, and holds 362 of the 6,423 offence-creating instruments here.

So most instrument pages say nothing about procedure, and where they say nothing they say why: an instrument the service has no record of and an instrument that predates the service entirely are different facts, and only the first is a fact about Parliament.

Neither House can amend a statutory instrument. That is not an inference — across every procedural step recorded for these instruments, in 75 distinct step descriptions, not one is an amendment. An instrument is put to each House as a whole thing, to be approved or not. What the Houses amend, when they amend anything, is the motion about the instrument: a peer moves “as an amendment to the motion to approve the Regulations” to leave out the approval and insert a refusal. That is what a fatal motion, and a motion of regret, actually is.

Sentencing guidelines

The guideline links on each offence page are a curated mapping by subject, not a finding about the individual provision, and they point at a search of the Sentencing Council's own site rather than reproducing any of its content. The Council's site asks automated agents not to crawl it, and that request is honoured here.

Being a good citizen of the source

legislation.gov.uk is a free public service run by The National Archives. The harvest that produced this report is rate limited by default, honours the service's non-standard HTTP 438 “rate limit exceeded” response, and permanently halves its own request rate when throttled. The text it fetched is cached, so changing a classification rule costs a local re-extraction rather than another crawl of somebody else's bandwidth.