Coverage methodology note

Headline corpus figure: 69,885 fully digitised pieces of in-force UK legislation — every item that legislation.gov.uk currently exposes with substantive machine-readable CLML body content (NumberOfProvisions > 0, full_text >= 200 chars). This is the analyser input. (Up from 69,462 after ingesting 423 net-new recoverable items surfaced by the completed API-exhaustion sweep — see “Exhaustion sweep” below.)

Within the wider National Archives in-force universe of 119,841 catalogued items, the remaining 49,956-item difference is structural — overwhelmingly local/private Acts and pre-modern instruments held only as PDF scans (Tier 2 / Tier 3 below). The picture depends on which scope of the statute book is in view:

Scope-stratified coverage

The full-universe 58.3 % coverage figure conflates two structurally different populations: (i) general-application legislation — Public Acts, Statutory Instruments, devolved primary and secondary law — which is essentially fully digitised, and (ii) local and private Acts — bespoke statutes addressed to named persons, places or companies — which have historically been published only as PDF scans. Reporting a single ratio masks this. The three scopes below disentangle them:

Scope Description In-force universe In corpus w/ text Coverage
A — Full universe Every in-force item across all 28 leg_types 119,841 69,885 58.3 %
B — General application Excludes ukla / ukppa / gbppa / eppa / gbla / uklp (Local + Private + Personal Acts) 89,548 69,695 77.8 %
C — Post-1990 general application Scope B AND year ≥ 1990 (modern regulatory law) 67,403 61,745 91.6 %

The 91.6 % figure for modern general-application law is the most directly relevant denominator for any analysis of contemporary regulatory burden. The 58.3 % full-universe figure is included for completeness and so the structural composition of the gap can be inspected directly. (The Scope B and C corpus numerators — 69,695 and 61,745 — are lower than the 69,885 headline because they exclude the 190 texted local/private-Act items, which fall outside a general-application denominator by nature.)

Three-tier structural breakdown (Scope A)

These are exhaustive per-item counts from the completed API-exhaustion sweep (every in-force item with a coverage gap probed across all retrieval channels; see “Exhaustion sweep” below), not extrapolations.

Tier Description Items
Tier 1 — Fully digitised (corpus headline) Substantive machine-readable CLML; in corpus and analyser-ready 69,885
Tier 2 — Digitised but no analyser-usable text On legislation.gov.uk but PDF-only / NumberOfProvisions = 0 / Akoma-Ntoso-only — data.xml yields no CLML body 31,407
Tier 3 — Not digitised /data.xml (and all channels) return HTTP 404; catalogued but no record on legislation.gov.uk 18,505
Pre-modern name-slug Acts (counted in Tier 1) In corpus under chapter-coded URIs; catalogued under slug IDs — see “Pre-modern identifier reconciliation” 44
In-force universe (catalogued) National Archives InForce CSV manifests (_inforce_definitive_count.py) 119,841

Tiers reconcile to the universe: 69,885 + 31,407 + 18,505 + 44 = 119,841. Tier 2 = 31,407 = 29,872 + 1,535 — the same set of items counted two ways: 29,872 metadata-only/ PDF shells (the shell verdict in the exhaustion-sweep table below) plus 1,535 items that are digitised but not CLML-ingestable (Akoma-Ntoso/PDF-only, deferred to phase 2). The 29,872 reported in the sweep table is therefore the shell component of this same Tier 2, not a competing total.

legislation.db total rows: 218,089. Rows flagged na_inforce = 1: 85,358 (of which 69,885 carry length(full_text) >= 200 and so enter the analyser; the remaining 15,473 are in-DB shells retained for amendment-chain tracking). The wider Tier-2 figure (31,407) counts all digitised-but-empty in-force items, most of which are not held in the DB.

In-force universe by decade

The in-force statute book has a long pre-modern tail. Counting the full National Archives in-force universe (the manifest denominator, not just the items held in legislation.db), only ~56 % of in-force items date from 1990 onward, and the pre-1900 universe alone is 25,842 items — overwhelmingly local/private and pre-modern Acts surviving only as PDF scans. The analyser-ready Tier-1 corpus, by contrast, is overwhelmingly modern: ~89 % of it dates from 1990 onward and the 2000s and 2010s alone account for ~56 % of it, precisely because digitisation coverage rises sharply at the 1990 Statute Law Database base date (below). The table sets the in-force universe against the Tier-1 corpus (na_inforce = 1 AND full_text >= 200); the coverage column is the universe-based ratio and reconciles to the “All types” column of the per-type matrix that follows.

Decade In-force universe In Tier-1 corpus Coverage
pre-1900 25,842 724 3%
1900s 2,087 48 2%
1910s 1,322 57 4%
1920s 1,431 99 7%
1930s 1,493 97 6%
1940s 1,037 198 19%
1950s 1,591 396 25%
1960s 2,972 743 25%
1970s 5,918 1,534 26%
1980s 8,528 4,065 48%
1990s 16,797 14,207 85%
2000s 21,213 19,876 94%
2010s 20,455 19,237 94%
2020s 9,149 8,604 94%
Total 119,835 69,885 58%

Coverage rises sharply at ~1990 (3–26 % for pre-1980 decades vs 85–94 % from the 1990s on). This is the Statute Law Database baseline effect: older in-force items (those carried into the revised dataset at the 1 February 1991 base date — see “Definitional rule” below) are disproportionately held as PDF-only or NumberOfProvisions = 0 shells without machine-readable CLML, so they are catalogued as in force but lack analyser-usable text. The universe totals 119,835 here rather than 119,841 because six in-force items carry a blank or unparseable year and fall into no decade bucket. Of the 119,841 in-force items, 85,358 are held in legislation.db — of which 69,885 carry substantive text (the Tier-1 corpus) and 15,473 are retained as in-DB shells for amendment-chain tracking.

Corpus composition by decade and leg_type

Tier-1 corpus (analyser-ready, na_inforce = 1 AND full_text >= 200) cross-tabulated by decade and the seven largest leg_types; other aggregates the remaining 18 types (chiefly ukla, nisi, asp, apni, nia, ukcm, aep, aosp). Totals reconcile to the 69,885 headline.

Decade uksi nisr ssi eur ukpga wsi eudn other Total
pre-1900 0 0 0 0 559 0 0 165 724
1900s 0 0 0 0 45 0 0 3 48
1910s 0 0 0 0 57 0 0 0 57
1920s 0 0 0 0 89 0 0 10 99
1930s 0 0 0 0 87 0 0 10 97
1940s 47 0 0 0 126 0 0 25 198
1950s 184 0 0 1 140 0 0 71 396
1960s 373 0 0 4 257 0 2 107 743
1970s 1,050 6 0 22 318 0 9 129 1,534
1980s 3,434 13 0 50 382 0 54 132 4,065
1990s 12,053 1,165 75 79 405 19 135 276 14,207
2000s 11,754 2,567 2,528 626 303 1,267 494 337 19,876
2010s 9,530 2,110 2,475 2,224 290 1,232 1,030 346 19,237
2020s 4,969 900 1,258 234 202 765 115 161 8,604
Total 43,394 6,761 6,336 3,240 3,260 3,283 1,839 1,772 69,885

Reading the matrix: UK statutory instruments (uksi, 43,394 / 62 % of the corpus) dominate every modern decade. The devolved secondary-law streams switch on with devolution — Scottish SIs (ssi), Welsh SIs (wsi) and the bulk of NI Statutory Rules (nisr) appear from 1999/2000 onward. Retained EU law (eur + eudn) peaks in the 2010s. Primary legislation (ukpga) is the only stream with meaningful pre-1900 presence and stays roughly flat (~200–400 per decade) throughout, reflecting the far lower volume but longer survival of Acts. Reproduced from legislation.db via the na_inforce flag; query in _inforce_definitive_count.py lineage.

Coverage by decade and leg_type

The table below converts the corpus counts above into coverage ratios: each cell is the share of that leg_type’s in-force universe in that decade — counted from the National Archives manifests under the same definitional rule (is_in_force in _inforce_definitive_count.py) — that is present in the Tier-1 corpus with substantive text. Numerator = Tier-1 corpus (na_inforce = 1 AND full_text >= 200); denominator = in-force manifest items of that type and decade. A dash (–) means no in-force items of that type exist in that decade; cells shown as 100 % include rounding (numerator ≈ denominator). The bottom-right cell reconciles to the headline ratio (69,885 / 119,841 = 58 %).

Decade uksi nisr ssi eur ukpga wsi eudn other All types
pre-1900 – – – – 46% – – 1% 3%
1900s – – – – 66% – – 0% 2%
1910s – – – – 61% – – 0% 4%
1920s – – – – 82% – – 1% 7%
1930s – – – – 72% – – 1% 6%
1940s 20% – – – 85% – – 4% 19%
1950s 24% – – 100% 81% – 0% 11% 25%
1960s 22% – – 67% 90% – 12% 11% 25%
1970s 29% 1% – 22% 94% – 9% 12% 26%
1980s 60% 1% – 33% 98% – 18% 30% 48%
1990s 100% 39% 100% 19% 100% 100% 29% 90% 85%
2000s 99% 77% 100% 77% 100% 98% 63% 100% 94%
2010s 95% 78% 97% 100% 100% 94% 99% 100% 94%
2020s 96% 79% 97% 100% 100% 95% 100% 100% 94%
All decades 85% 54% 98% 82% 79% 96% 65% 5% 58%

Reading the coverage matrix: modern general-application law is near-complete — from the 2000s on, uksi, ssi, wsi and ukpga all sit at 94–100 %. The two persistent gaps among general-application types are nisr (NI Statutory Rules, 54 % overall; ~77–79 % even in recent decades — the largest general-application shortfall, confirmed structural 404s) and retained EU law (eudn 65 %, eur 82 %, dragged down by PDF-only Commission Decisions/Regulations). The other column reads 5 % overall because its denominator is dominated by local and private Acts (ukla, gbppa, ukppa, eppa) that survive only as PDF scans — by constitutional nature these impose no general-application burden and are excluded from the Scope B/C denominators (see below). Pre-1980 coverage is low across the board: the in-force statute book of those decades is disproportionately held as pre-CLML shells under the InForce1991 baseline.

Definitional rule (in force)

An item from the National Archives InForce CSV manifests was counted as in force if it imposes legal obligations or prohibitions somewhere in the UK at the snapshot date. Operationally:

Unqualified repealedBy, revokedBy, repealedByLegislature and all NotInForce* variants were excluded (no current legal force anywhere in the UK).

Open question on InForce1991 (flagged for The National Archives). The InForce1991 status (6,892 items in the in-force universe) marks legislation carried into the revised dataset as in force at the 1 February 1991 Statutes in Force base date. It is not independently confirmed here whether every such item remains in force today, or whether the status records only its standing as at the 1991 baseline, not subsequently reconfirmed. These items are disproportionately pre-modern, so the pre-1980 coverage denominator (and hence the low pre-1980 coverage ratios in the by-decade table above) is mildly sensitive to this question; the post-1990 Scope-C figures are unaffected. Pending clarification from The National Archives, InForce1991 is counted as in force, consistent with its treatment on legislation.gov.uk.

Reproducibility script: _inforce_definitive_count.py.

Pre-modern identifier reconciliation (name-slug vs chapter-coded)

Coverage matching compares National Archives in-force identifiers against corpus identifiers by normalised-string equality (host, /id/ and trailing slash stripped, lowercased). This is exact and reliable for the modern type/year/number scheme and for numeric regnal citations (ukla/Geo3/41/100), which the catalogue and legislation.gov.uk share. It does not reconcile a small set of pre-1800 Acts that the Archives cite by descriptive name-slug (aep/Hen3/52/stat-marlbr) while legislation.gov.uk serves them under a chapter-coded regnal URI (aep/Hen3c23/52/23). The two schemes are disjoint, so a string match fails and the Act is wrongly counted as missing / not retrievable.

These are reconciled by an explicit override table (medieval_slug_overrides.csv, built by build_slug_overrides.py). The match is decided at the (leg_type, monarch, regnal-year) group level, refined by statute/session number for stN-style slugs; it confirms that the Act’s regnal-year group is present in the corpus, not that each work-list row is independently title-matched to a specific provision. On that basis 17 medieval Acts (44 work-list rows) are reconciled as already in the corpus under a chapter-coded URI — including Magna Carta (1297), the Statutes of Marlborough (1267) and Westminster the First (1275) and Second (1285), Quia Emptores (1290), Confirmation of the Charters, the Statute concerning Tallage, and the Treason Act 1351. For 8 of the 17 the regnal-year cell contains a single statute (so group presence is equivalent to per-statute presence); for the multi-statute cells the slugs are either statute-number-disambiguated or correspond one-to-one to the cell’s corpus entries, so in this dataset every reconciliation is also correct at the specific-statute level — but the matching guarantee is at the regnal-year-group granularity. A further 2 statute-number variants (5 Ric. 2 Stat. 1, 1381; 1 Geo. 2 Stat. 2) were reviewed and confirmed as genuine gaps — the former repealed by the Criminal Law Act 1977, the latter not served by legislation.gov.uk as a distinct in-force item.

The residual numeric impact on coverage is ≈0.1 % and does not affect the headline figures. The limitation is recorded here, and patched by the override table, rather than addressed by a general slug↔chapter-code canonicaliser, which the scale (≈140 candidate items across the whole work-list, of which 44 reconciled) does not warrant. Note the strict distinction maintained throughout: items the matcher reports absent are “not ingested”, which is not the same claim as “not digitised on legislation.gov.uk” — the gap between the two is the subject of the exhaustion sweep below.

Exhaustion sweep — all retrieval channels tested

Every in-force item with a coverage gap (50,765 items) was probed across all retrieval channels on legislation.gov.uk — /data.xml, /data.xml?version=enacted, /enacted/data.xml, /data.xht, /data.akn, and per-year point-in-time endpoints — to an exhaustive per-item verdict (355,343 probes total). This replaces the earlier sample-based extrapolation. Per-item verdicts:

Verdict Items Meaning
recoverable (substantive CLML) 2,344 a channel returned ingestable body text
already in corpus 44 pre-modern slug false-404, present under chapter-coded URI
shell (digitised, no CLML body) 29,872 200 but NumberOfProvisions = 0 / PDF-only
not digitised (404) 18,505 absent from every channel
incomplete 0 the 200 timed-out items were re-probed to definitive verdicts

Validation. A stratified independent re-check (validate_frontier.py, hitting the canonical HTML page — a different signal than the data.xml the sweep keyed on) confirmed the negative verdicts: false-404 rate 0 / 487 (95% CI 0.0–0.8%) among sampled not_digitised items, and 50 / 50 of a shell control re-confirmed as PDF-only. The only systematic false-404s were the ~140 pre-modern name-slug items, reconciled separately above.

Recoverable ingest. Of the 2,344 recoverable, 386 were already in the corpus (a stale work-list — chiefly the 385 Best-Collection items ingested earlier) and 423 were ingested as net-new CLML Tier-1 via the existing pipeline (_ingest_recoverable.py; source priority revised-current → Best-Collection revised → made/enacted; no-force provisions stripped; na_inforce = 1). This is what moved the headline 69,462 → 69,885. Ingested by type: uksi 145, nisr 130, eur 92, ssi 22, wsi 17, eudn 7, ukpga 6, asc 1, asp 1, nisro 1, mwa 1.

Deferred to phase 2 (counted): 1,535 items. The remaining recoverables are digitised but not CLML-ingestable — their data.xml is a 0-provision / PDF-only shell and they registered as “recoverable” only via an Akoma-Ntoso representation the analyser does not parse. These are not pipeline-ingestable today and are deferred to the phase-2 target list (below), counted toward Tier 2. Composition is predominantly uksi 755 + nisr 425 (not retained-EU: eur 50 + eudn 18 = 68), then ukla 178, ukpga 31, nisro 25, wsi 22, uksro 17, ssi 10, nisi 4.

Reproducibility: sweep checkpoint _exhaustion_ckpt_v3.db (md5 681772dd…); aggregation aggregate_exhaustion_sweep.py → exhaustion_recoverability_by_type.csv / ..._by_type_year.csv; validation validate_frontier.py → validation_results.csv; ingest _ingest_recoverable.py. Earlier sample-based work is in api_exhaustion_findings.md / api_exhaustion_verification.md with artefacts under _probe_sweep/.

Confirmed unrecoverable types

Six leg_types returned 0 recoverable items in the completed sweep (n ≥ 20): every in-force gap item is a metadata-only/PDF shell or a hard 404 across all channels. 11,104 in-force gap items, surviving only as scanned PDFs in the original print collection:

type gap items shell (PDF-only) not digitised (404)
gbppa (GB private acts, pre-1801) 5,317 0 5,317
ukppa (UK private & personal acts) 3,839 842 2,997
eppa (English private acts, pre-1707) 1,611 20 1,591
apni (Acts of the NI Parliament) 172 172 0
aip (Acts of the Irish Parliament, pre-1800) 144 0 144
aosp (Acts of the Old Scottish Parliament, pre-1707) 21 0 21

The earlier sample-based list named eight types; the completed sweep found recoverable items in uksro and gbla, so they no longer qualify, and uklp (11 items) falls below the n ≥ 20 threshold. Counts are in-force gap items per type — some of these types also hold items already in the corpus from earlier ingestion (e.g. aosp has 65 in corpus, not part of the gap); those are unaffected.

Items catalogued in-force but not digitised in full

The 49,956-item structural gap (Tier 2 + Tier 3) is dominated by local/private Acts and pre-modern instruments. The composition, from the completed sweep:

Why the local and private Acts gap does not materially affect the regulatory burden measure

Local and private Acts apply, by their constitutional nature, to named persons, places, or institutions identified within the Act itself — a particular railway company, named landowners, a specific bridge, a single corporation — and impose no obligations or prohibitions of general application on the public or on regulated entities of a class. Excluding them from the analyser denominator therefore does not reduce the regulatory burden being measured, because they never formed part of the regulatory burden that any UK private actor outside the named subjects could face. This is why Scope B (77.8 %) is the most defensible denominator for general-application regulatory-burden analysis, and Scope C (91.6 %) for modern regulatory-burden analysis.

Regulatory layers beyond the statute book — what is out of measure

The measure covers the centrally published statute book — the in-force legislation legislation.gov.uk exposes as machine-readable text (the Tier-1 corpus above). Legally binding rules also live in other layers this measure does not reach, and they must be distinguished from non-binding material, not lumped together:

Frontier proxies — the one place the statute book reaches across the boundary. Where the statute book itself imposes a duty to comply with one of these out-of-measure instruments (“it is an offence to contravene a permit condition”; “contravention of a byelaw under this section is an offence”), that compliance duty is counted, once, as a frontier proxy — standing at the visible frontier for the invisible layer of rules behind it. This is the deepest-visible-layer rule in rubric §3. The count is one proxy per out-of-measure target, never an enumeration of that target’s internal contents. Frontier proxies are tagged (frontier_hook = true, frontier_target_type) so the population is enumerable and reportable as “N frontier duties standing proxy for out-of-measure layers.” These out-of-measure layers are a natural subject for a later phase (the objective anchor already names regulatory rulebooks as a future extension); bucketing them is confirmed as a job for later, not now.

Growth analysis — regulation migrates between layers

Because binding rules live in several layers, regulation can migrate between them with no net change in regulatory burden. The clearest live case is the Financial Services and Markets Act 2023 transfer of retained EU law (REUL) into the FCA/PRA rulebooks: as those requirements move from the statute book (in measure) to the regulators’ Handbooks (out of measure), the measure will record a decline in statute-book financial-services burden that is layer-migration, not deregulation. The growth / flow analysis must name this as an adjustment when the data exists, so a migration is not read as a removal. (Counts are not additive across scope expansions: if a future phase brings a rulebook layer into measure, its contents are counted there and the corresponding frontier proxies re-classify to counted_at_source — rubric §3 — so the same rule is never counted twice.)

Phase 2 target list

missing_inforce_legislation.csv contains the in-force items not present in legislation.db, with columns item_url, title, year, leg_type, status, data_xml_url. This is the Phase 2 target list for PDF-OCR digitisation or further per-item retrieval as the National Archives completes its digitisation programme.

The completed sweep adds a second, counted phase-2 category: 1,535 items that are digitised but not CLML-ingestable — their data.xml is a 0-provision/PDF-only shell, and they register as recoverable only via an Akoma-Ntoso representation the current pipeline does not parse. Composition: predominantly uksi 755 + nisr 425 (not retained-EU: eur 50 + eudn 18), then ukla 178, ukpga 31, nisro 25, wsi 22, uksro 17, ssi 10, nisi 4. A future Akoma-Ntoso parser is the natural route to capture the subset that carries real AKN provision text; the PDF-only remainder needs OCR.

Reproducibility

Artefact Purpose
_inforce_definitive_count.py Computes the 119,841 headline figure from the CSV manifests.
_inforce_db_coverage.py Cross-references the in-force universe against legislation.db.
_probe_sweep_all_types.py 50-item /data.xml probe across every leg_type.
_reprobe_db_shells.py 100-item probe of DB shells (na_inforce=1, full_text < 200).
_alt_endpoint_retry.py Six-endpoint probe of high-impact shell pools.
_alt_xht_reclassify.py Refined XHTML content classifier (rejects chrome-only responses).
_bulk_collection_index.py Indexes every file in the Best Collection bulk download.
_bulk_collection_xref_v2.py Cross-references BC against project-canonical in-force universe.
_ingest_bc_792.py Ingests the 792 BC items identified as recoverable; updates shells / inserts new rows.
_scope_stratified_coverage.py Computes the three-scope coverage table.
_verify_three_checks.py Independent verification of the three headline claims.
_build_coverage_table_v2.py Builds the type-by-type coverage_table.csv.
aggregate_exhaustion_sweep.py Aggregates the completed sweep checkpoint to per-item verdicts (incl. slug + re-probe overrides).
build_slug_overrides.py Builds the pre-modern name-slug reconciliation table (medieval_slug_overrides.csv).
validate_frontier.py Independent canonical-page validation of negative verdicts (false-404 / shell).
_ingest_recoverable.py Ingests CLML-substantive recoverables via the existing pipeline; defers AKN/PDF-only to phase 2.
InForce_results_47/result_table_*.csv Source manifests from The National Archives (18 files).