Getting started with danlex

danlex provides access to Danish legislation published by Retsinformation, the official state legal information system. This vignette works through a single instrument in detail and then scales up to the corpus.

This vignette was pre-computed: the code was run against the live service when the package was built, so the results below are real but the examples do not call the API when the package is checked or installed.

Finding a document

Danish acts and regulations are cited by year and number — “bekendtgørelse nr. 763 af 1998”. That maps directly onto an identifier:

doc <- dlx_get_doc("lta", 1998, 763)
doc$title
#> [1] "Bekendtgørelse om indbetaling af Den Særlige Pensionsopsparing for medlemmer af arbejdsløshedskasser for lønmodtagere ved udbetaling af dagpenge m.v."
doc$document_type
#> [1] "Bekendtgørelse"
doc$ministry
#> [1] "Beskæftigelses- og Ligestillingsministeriet"

The first argument is the collection. lta is Lovtidende A, which carries both acts and regulations in a single number space — so the instrument above is a bekendtgørelse even though it sits in the same series as acts. The type is never inferable from the identifier alone, which is why document_type is always returned.

The other collections are ltb and ltc (Lovtidende B and C, the latter carrying treaties), mt (Ministerialtidende, discontinued in 2012), retsinfo (administrative decisions), fob (Ombudsman decisions) and ft (parliamentary documents).

A document that does not exist is a normal answer, not an error:

dlx_get_doc("lta", 1998, 999999)$status
#> [1] "not_found"

Text availability

Retsinformation returns full text only for material published from late September 2007 onward. Earlier documents carry metadata alone, and status reports this:

doc$status
#> [1] "metadata_only"

The boundary is sharp. Act 1080 of 2007, announced on 14 September, has no text; act 1081, announced on 25 September, does:

dlx_get_doc("lta", 2007, 1080)$status
#> [1] "metadata_only"
dlx_get_doc("lta", 2007, 1081)$status
#> [1] "ok"

This is a property of the source. Metadata for earlier material is complete and reaches back to 1852 for Lovtidende A — and, in the retsinfo collection, to 1665.

Reading a document

For a modern instrument, dlx_get_text() returns the running text with its structural markers in place:

txt <- dlx_get_text("lta", 2025, 50)
txt$n_char
#> [1] 9601
cat(substr(txt$text, 1, 400))
#> I medfør af § 7 b, stk. 2, § 22, stk. 1, nr. 1, 3, 4, 5, 6 og 7, og stk. 3-4, § 23 og § 30 i lov om erhvervsakademiuddannelser og professionsbacheloruddannelser, jf. lovbekendtgørelse nr. 396 af 12. april 2024, fastsættes efter bemyndigelse i henhold til § 4, stk. 1, i bekendtgørelse nr. 1118 af 28. oktober 2024 om delegation af uddannelses- og forskningsministerens beføjelser til Uddannelses- og

Working with provisions

Danish legal drafting is highly structured, and Retsinformation preserves that structure in its XML. dlx_get_paragraphs() returns one row per subsection, so provisions can be handled as data rather than as text to be parsed:

p <- dlx_get_paragraphs("lta", 2025, 50)
nrow(p)
#> [1] 25
p[1:4, c("paragraf", "stk", "text")]
#> # A tibble: 4 × 3
#>   paragraf stk     text                                                         
#>   <chr>    <chr>   <chr>                                                        
#> 1 § 1.     <NA>    Formålet med erhvervsakademiuddannelsen som maritim teknolog…
#> 2 § 1.     Stk. 2. Den uddannede skal have viden, færdigheder og kompetencer (l…
#> 3 § 1.     Stk. 3. Den uddannede, som har gennemført de maritime valgfag, skal …
#> 4 § 2.     <NA>    Uddannelsen giver den uddannede ret til at anvende titlen ma…

Note that the first subsection of each section carries no marker: Stk. 1. is implicit in Danish drafting practice and simply absent from the source.

Amending instruments

An amending instrument contains provisions that are not its own — they are new text destined for another instrument. The is_amendment column makes the distinction:

own <- dlx_get_paragraphs("lta", 2025, 50)
amd <- dlx_get_paragraphs("lta", 2025, 1)

table(own$is_amendment)
#> 
#> FALSE 
#>    25
table(amd$is_amendment)
#> 
#> TRUE 
#>   12

The second instrument amends a regulation from 2018; every provision it contains belongs to that regulation, not to itself. Treating the two alike would badly distort any corpus-level analysis of what Danish law says.

References between instruments

Regulations record the provisions they were issued under, and dlx_get_references() returns these as an edge table:

refs <- dlx_get_references("lta", 1998, 763)
refs[, c("ref_accn", "ref_date", "ref_title")]
#> # A tibble: 2 × 3
#>   ref_accn     ref_date   ref_title                                             
#>   <chr>        <date>     <chr>                                                 
#> 1 A19990059229 1999-07-14 Bekendtgørelse af lov om arbejdsløshedsforsikring m.v.
#> 2 A19990066629 1999-08-19 Bekendtgørelse af lov om en aktiv arbejdsmarkedspolit…

The ref_accn values are accession numbers, which dlx_get_doc() accepts directly — so a citation network can be traversed without building a lookup table first:

parent <- dlx_get_doc(accn = refs$ref_accn[1])
parent$title
#> [1] "Bekendtgørelse af lov om arbejdsløshedsforsikring m.v."
parent$document_type
#> [1] "Lovbekendtgørelse"

References are maintained, not historical

One property deserves emphasis before anyone builds a network from this data. The referring instrument is from 1998, yet its references are dated 1999:

refs$ref_date
#> [1] "1999-07-14" "1999-08-19"

This is not an error. Retsinformation updates these links to point at the current consolidated version of the parent act, so they reflect the state of the law now rather than at the moment of enactment. An analysis that treats the edges as timestamped to the referring document’s date will be systematically wrong.

Scaling up

Enumerating the corpus

dlx_list_documents() retrieves the full index of everything published through the service:

idx <- dlx_list_documents(progress = FALSE)
nrow(idx)
#> [1] 202924
table(idx$collection)
#> 
#>      fob       ft      lta      ltb      ltc       mt retsinfo 
#>     2957    41492    63218       45     4783     6438    83991

The first call takes about a minute, because the index is published as a sitemap split across 21 pages whose ordering does not permit fetching a single year. The result is held for the rest of the session, so later calls are immediate.

Nothing is written to disk. To keep the index between sessions, save it:

saveRDS(idx, "danlex-index.rds")

The index carries identifiers and modification dates only. Note that lastmod records when Retsinformation last revised its record, which is unrelated to when the instrument was enacted or repealed.

Coverage varies by collection

The index is the quickest way to see what is actually available:

lta <- subset(idx, collection == "lta" & !is.na(year))
range(lta$year)
#> [1] 1852 2026

mt <- subset(idx, collection == "mt" & !is.na(year))
max(mt$year)
#> [1] 2012

Ministerialtidende ends in 2012 because it was discontinued. Ombudsman decisions (fob) begin in 1980, and treaties (ltc) drop off sharply after 2008. These are genuine historical discontinuities rather than gaps in coverage, and an empty result for a given year may simply mean nothing was published.

Retrieving in bulk

dlx_get_docs() takes a vector of identifiers. Here is every treaty published in 2025:

treaties <- dlx_list_documents(year = 2025, collection = "ltc",
                               progress = FALSE)
docs <- dlx_get_docs(treaties$eli, progress = FALSE)

nrow(docs)
#> [1] 6
table(docs$status)
#> 
#> ok 
#>  6

Failures are returned as rows with status = "error" rather than raised as conditions, so a long run survives a dropped connection:

sum(docs$status == "error")
#> [1] 0

Once retrieved, the metadata is ordinary data:

sort(table(docs$ministry), decreasing = TRUE)
#> 
#>         Udenrigsministeriet Skatte- og Vækstministeriet 
#>                           5                           1

Tracking changes

The index is refreshed monthly, so recent material may be missing from it. dlx_get_changes() reads a separate feed covering at least the last 60 days:

ch <- dlx_get_changes()
nrow(ch)
#> [1] 405
range(ch$updated)
#> [1] "2026-07-02" "2026-08-29"
table(ch$published_in)
#> 
#>    Lovtidende A    Lovtidende C Retsinformation 
#>             101               2             302

The feed itself records only the publication channel, not the collection. danlex derives collection from the identifier, which makes it possible to separate legislation from administrative decisions:

recent_acts <- dlx_get_changes(collection = "lta", reason = "NewDocument")
nrow(recent_acts)
#> [1] 83
head(recent_acts$title, 3)
#> [1] "Bekendtgørelse om Ledningsejerregistret"                                                                           
#> [2] "Bekendtgørelse om landsplandirektiv for to arealer til højspændingsstationer i hhv. Køge Kommune og Stevns Kommune"
#> [3] "Bekendtgørelse om kriterier for vurdering af kommunale projekter vedrørende vandløbsrestaurering"

Because the feed is a static file with no query interface, these filters apply after retrieval. Asking for a date earlier than the feed reaches will return a truncated result and warn.