| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Bisher nahm ÄndGgner nur verkündete Artikelgesetze und Gesetzentwürfe an, ohne
den Unterschied zu kennen. Dokumente, die einen Entwurf ändern statt ein
Gesetz, lieferten still null Befehle.
Neu:
* Dokumentart-Erkennung aus dem Rohtext (DokumentErkenner, DokumentArt,
DokumentKopf) — vor der Bereinigung, weil gerade die Drucksachenköpfe als
Kolumnentitel wegfallen. Erkannt wird nie am Dateinamen: Die Beispieldaten
enthalten eine Datei namens Beschlussempfehlung, die in Wahrheit ein
Entschließungsantrag ist. Dokumente ohne Befehle erzeugen jetzt eine
benannte Warnung statt stiller Leere.
* Änderungsanträge (AenderungsantragParser, EntwurfsPatcher): Der Antrag wird
auf den Entwurfstext angewandt, der geänderte Entwurf danach wie gewohnt auf
das Stammgesetz — eine komponierte Synopse. Drucksachen- und Gesetzesstellen
bleiben getrennte Typen; die Zuordnung Antrag→Entwurf läuft über die
Drucksachennummer, nicht über die Argumentreihenfolge. Der 1910 Zeilen große
BefehlAnwender bleibt unberührt.
* Beschlussempfehlungen werden erkannt und mit Begründung übergangen, die den
brauchbaren Entwurf beim Namen nennt. Die Spaltentrennung der
Zusammenstellung ist gebaut (FontgroessenFilter.Spalte, koordinatenbasiert
am Steg) und belegt: Die linke Spalte ergibt Befehl für Befehl den
Regierungsentwurf. Die rechte Spalte bleibt bewusst offen — sie vermerkt
„unverändert“ auch zeilenweise innerhalb zitierter Blöcke, ihre
Anführungszeichen gehen daher nicht auf; nötig ist die zeilenweise Zuordnung
beider Spalten über die gemeinsame Grundlinie.
Die Synopse kennzeichnet eine Entwurfsfassung als solche.
296 Tests grün (vorher 269); alle gepinnten Akzeptanzzahlen unverändert.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Change-Id: I0001d25fcbd9969fc06eea675edc4326ce2e02f9
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Investigating "Anlagen as amendment target" corrected the diagnosis from
yesterday: Anlagen are supported. In gii-XML they are norms of their own
("Anlage 8", "Anhang"), Stelle.anlagenEnbez() resolves them, and the GEG
sample applies annex-targeted commands — only two of its fifty manual
entries touch annexes, both for content reasons.
What actually failed in Berlin was narrower, and is fixed:
* A law with a single annex names it after the provision it belongs to
("Die Anlage zu § 2 Absatz 4 Satz 1"). StellenParser demanded a
number after "Anlage" and rejected the phrase, so the article's frame
command went unrecognised and its items lost their context. The
annex now carries the designation "Anlage", like "Anhang". The naming
suffix is skipped deliberately: read along, the "§ 2" inside it would
become the target and the wrong norm would be amended.
* C0 control characters are now stripped. Berlin's enumeration indent
carries a U+0007 that sits invisibly in front of the command and made
its start-of-line anchor miss.
* Command verbs torn apart by a stray space in the official typesetting
("ein gefügt" for "eingefügt") are rejoined, restricted to sequences
that are not valid German word order.
Three of Article 1's four commands are now recognised. The fourth places
the new unit by a word anchor ("Vor den Wörtern … wird folgender Absatz 5
eingefügt"), which StrukturEinfuegung cannot express; that and the
inability of the plain-text stem format to carry an Anlage at all are
recorded in Landesrecht-Beispiele.adoc.
231 tests green; all pinned figures unchanged.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Change-Id: Ic5804bf343dd5d8ce435c07b8f882ca0372890d2
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The plan assumed the two-column Berlin gazette would need x-coordinate
column detection. It does not: the content stream already emits the left
column fully before the right one, so commands come out in reading order.
The test pins that finding.
What does go wrong there is different, and two parts of it are fixed:
* The article's amendment formula carries its own target ("§ 2 Satz 1
des Gesetzes zur Errichtung … wird wie folgt geändert"), but the
following items never inherited it. The pattern's leading \b could
never match before "§" — a non-word character needs a word character
in front of it — so the paragraph branch was dead code and only the
Bavarian "Art." branch ever fired.
* The running page header is not a line of its own; it sits inside the
body text of the following column, so it has to be cut out rather
than dropped as a column title.
Two blockers remain and are recorded rather than papered over: Article 1
amends an *Anlage* with its own numbering, which the model does not carry
(the same gap that shelved Baden-Württemberg), and full-width frames
(title block, imprint) sit elsewhere in the stream than on the page,
which would need a real XY-cut. Stems are unobtainable either way —
gesetze.berlin.de is an authenticated juris app like the Schleswig-
Holstein portal.
229 tests green; all pinned figures unchanged.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Change-Id: I4619dbe96a023d61104d7107a9a34533123d1bc5
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The NEFG acceptance case left two commands as "check manually". Both are
now supported, so Lower Saxony applies 7 of 7 with no remainder.
* Lower Saxony cites inserted paragraphs with a space before the sub
number ("§ 2 a"); the rest of German legal usage writes "§ 2a", which
is what the Stelle and level parsers expect. TextBereiniger now pulls
the form together, leaving enumeration markers of the amendment act
alone ("… nach § 8 c) In Absatz 2 …").
* "In Kapitel 4 wird nach § 12 der folgende neue § 13 angefügt" — the
division only names the section the new paragraph lands in; the
anchor governs the position, as in the plain form. The optional
"neue" is now accepted in both.
Applying that insertion needed one more thing: the paragraph it creates
is only vacated by the *following* command ("Der bisherige § 13 wird
§ 14"), so in document order the two collided and the insertion was
rejected. "Bisherig" denotes the state before the amendment, so the
renumbering logically precedes the reoccupation and is now pulled ahead.
The log still lists commands in the order of the amendment act.
226 tests green; all other pinned figures unchanged.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Change-Id: I0e2ea5b0389f8002157eeb8471800da51febe739
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Article 3 of the 22. Rundfunkänderungsgesetz (GV. NRW. 2026 S. 202) amends
the Telemedienzuständigkeitsgesetz. All four commands apply cleanly, and
the result matches the official consolidated version of 1 April 2026.
recht.nrw.de turned out to be the one remaining state portal that serves
consolidated laws as server-rendered HTML, with the version history
encoded in the URL date prefix — so the pre-amendment stem (27 Apr 2022)
was directly obtainable.
Two fixes were needed:
* LandesRechtTextParser read the parenthesised title suffix wholly as
the abbreviation. State laws routinely carry both short title and
abbreviation there ("(Telemedienzuständigkeitsgesetz – TMZ-Gesetz)"),
and the amendment act cites the short title, so article selection
never matched. The parser now splits at the dash.
* ZitatExtraktor swallowed the rest of the document when a closing
quotation mark is missing in the official text (Article 2 no. 11
here), which hid Articles 3 to 5 entirely. An amendment act never
quotes across an article heading, so an open quotation is now closed
before a bare "Artikel N" line, with a warning. Quoted paragraphs are
unaffected: Bavarian norm heads read "Art. N".
223 tests green; all previously pinned figures unchanged.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Change-Id: I78470075d9be2433966b821b6fe1e21808996c46
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The GVOBl. Schl.-H. amendment act (Kommunalrecht, 2026/27) brings three
conventions the tool had not seen:
* a running two-line page header that sits between articles and glued
itself onto the following article heading,
* superscript footnote markers on article headings ("Artikel 1 ¹)"),
which defeated the article split, and
* ragged-right typesetting, where the geometric margin detection finds
no alignment cluster and reports full lines as hard line ends, so a
trailing hyphen was never joined.
All three are fixed in TextBereiniger; the hyphen rule now trusts the
extractor's trailing-space signal over the geometric class.
A full acceptance test is not possible: the Schleswig-Holstein law portal
serves document bodies only through an authenticated API, and the last
freely archived consolidated Gemeindeordnung (14 Nov 2022) predates the
amended § 34a. The new test therefore covers everything up to and
including command recognition; the blocker is recorded in
Landesrecht-Beispiele.adoc.
219 tests green; Bund/Bayern/Sachsen/Niedersachsen figures unchanged.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Change-Id: I9b3ab8a756372504b51c4d410b491fded255da96
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Second §-structured Landesrecht case for wave six: the Lower Saxony ELER Support
Act (NEFG, unchanged since 2022, derived from the two-column Nds. GVBl 2022 No. 33)
plus the amending act of 28 Jan 2026 (Nds. GVBl 2026 No. 10). Five of seven
commands apply — both "erhält folgende Fassung" re-enactments (§ 1(1), § 2), the
citation replacement in § 1(5), the words insertion in § 6(1), and the § 13 → § 14
renumbering.
- LandesRechtTextParser: a norm head no longer accepts a title ending in a full
stop, so a cross-reference sentence at line start ("§ 7 GAPInVeKoSG findet
entsprechend Anwendung.") is not misread as a norm head "§ 7" (which had made the
monotonicity check swallow the norms in between).
- TextBereiniger: strip the Lower Saxony GVBl running footer and publisher line.
The footer sat between two enumerated items and otherwise attached to the
preceding re-enactment quote, defeating its end-of-sentence anchor.
- Add sampledata/Niedersachsen/NEFG-alt.txt and EndToEndTest.nefgAcceptance.
Two insertion forms remain documented residues (manuell prüfen): inserting a
§-block with a space-separated sub-number ("§ 2 a") and a chapter-scoped §-block
insertion ("In Kapitel 4 wird nach § 12 … angefügt").
221 tests green; federal, Bavarian and Saxon pinned numbers unchanged.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Change-Id: Ie23eee1f596ec8eecc6a73f6ce585a22e73ec6c9
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Complete the first §-structured Landesrecht acceptance case for wave six: the
Saxon Civil Servants' Pension Act (SächsBeamtVG) in its 1 Feb 2025 version plus
the amending act of 13 April 2026 (SächsGVBl. S. 134). All five commands are
recognised and applied without residue — the amount replacements in § 47 and the
re-enactment of § 80(1) sentence 2.
- LandesRechtTextParser: recognise arabic, keyword-first structural headings
("Abschnitt 1", "Unterabschnitt 2", "Teil 3") via GLIEDERUNG_ARABISCH, in
addition to Bavaria's roman "I. Abschnitt" form. Bavaria behaviour unchanged.
- TextBereiniger: strip the footer of revosax full-text output (REVOSAX_FUSS,
"https://…revosax… Fassung vom … Seite N von M").
- Add the consolidated pre-amendment stem as canonical plaintext
(sampledata/Sachsen/SaechsBeamtVG-alt.txt, derived from the archived revosax
full text) and EndToEndTest.saechsBeamtVgAcceptance (5 applied, 0 manual).
220 tests green; federal and Bavarian pinned numbers unchanged.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Change-Id: I8e878bc9506f24028cf643a873de000f3d066d64
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Prepare wave six (Landesrecht of the remaining Länder) by lifting the
Bavaria-specific stem-law loader to a shared one. Unlike Bavaria (Art./§
inversion), the other Länder structure their stem laws in §, so the loader must
handle both sigils.
- Move gesetz/bayern/{BayRechtLoader,BayRechtTextParser} to
gesetz/land/{LandesRechtLoader,LandesRechtTextParser}; the norm head now
matches "§ N" as well as "Art. N" and derives the sigil per norm from the
match. Cross-reference keywords in the norm head are excluded only as whole
words (so a title "Satzungen" no longer trips on "Satz"), and the juris
abbreviation may be multi-token ("(GO NRW)").
- Derive the superscript mode data-drivenly from the loaded stem law
(Superskript.traegtSatznummern) instead of from "is it gii-XML": Bavaria and
Lower Saxony keep their amtliche Satznummern, the Bund and Länder without
official sentence numbering drop them.
- Recognize the neufassung idiom "erhält/erhalten folgende Fassung" (Schleswig-
Holstein, Niedersachsen) via a NEUFASSUNG_VERB building block, additive to
"wird/werden wie folgt gefasst".
Federal and Bavarian behaviour is unchanged (219 tests green, pinned acceptance
numbers UWG 19/0, GEG 66/53, GEG-GModG 90/9, IfSG 42/24, BayJG 149/154 hold).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Change-Id: I55a70a7932bf657a2346ca70f3fa05e173bf80ad
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Extend ÄndGgner from federal-only to Bavarian state law. Bavarian base
laws are structured in Artikel while their amending acts are structured
in Paragraphen (inverted from the Bund); official sentence numbers and
footnote markers are carried as Unicode superscripts (¹²³, ⁶)).
- Generalise Stelle.Paragraph to carry a sigil ("§" or "Art.") and route
it through recogniser, applier and resolver (bit-identical Bund output).
- Superscript pipeline: geometric detection in FontgroessenFilter
(SuperskriptModus BEHALTEN), Superskript util, exact sentence splitting
in SatzTeiler, label-based sentence resolution.
- BayRechtLoader/BayRechtTextParser for gesetze-bayern.de PDF/plaintext.
- §-structured amending acts, GVBl/Landtag column titles, non-breaking
spaces, the GVBl continuation quote; Bavarian command forms (footnote
aufhebung, Satznummerierung streichung, Wortlaut forms, Halbsatz,
gapping chains).
Acceptance (EndToEndTest.bayJgGvblAcceptance): the pre-2026 BayJG
fassung (BayJG-alt.txt, reconstructed from Wayback single-article
snapshots) with GVBl 6/2026 §§ 1-2 applied — 154 commands, 0 unknown,
149 applied automatically, 5 pinned residuals (follow-up edits inside
two multi-step renumbering sequences in Art. 29a and Art. 56).
Application-side fixes surfaced by the acceptance run (all Bund-safe):
sentence-start superscript before §; sentence boundaries = {0} ∪ {each
number ≥ 2}; gapping scope inheritance for bare word operations;
footnote definition lines hidden from word operations; absatz aufhebung
marks "(weggefallen)" keeping its number, and absatz renumbering
overwrites an empty placeholder (weggefallen/gegenstandslos) target.
211 tests green (mvnw verify); federal reference numbers unchanged.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Change-Id: Iebae4c17ca90755c5fd36251362042f3d5796fd0
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The percentile-based right-margin estimate from the previous commit
misclassified two real-world layouts: at the column switch of the
two-column old-BGBl format a window contains both column margins, so the
90th percentile returned the right column's margin and marked every
left-column soft wrap as deliberate (vetoing hyphen mending, IfSG 2020:
"Impf- surveillance"); and series of equally wide centered "Artikel N"
headings in draft bills formed the top of their window's distribution,
so they counted as full-width and were reflowed into the following line,
breaking teileInArtikel's line anchor (ProdHaftG RegE). Classify against
alignment clusters instead: a line end is soft when at least five window
lines end within 4 pt of it (the justified block's own margin), hard when
such a cluster runs at least 10 pt above it, unclassified otherwise.
Structure anchors ("Artikel 2", "§ 19", bare Gliederung labels) are
always exempt from reflow since equal-width heading series still form
sham clusters. Also flush the pending line end-X at page ends — without
it, the first line of the next page inherited the previous page's footer
geometry and was misclassified (UWG: "Artikel 2" swallowed the closing
provisions).
Markerless joins additionally never cross into a following enumeration
marker line ("...vorgesehen und" + "d) die Überwachung" no longer glues
to "undd)"), a text-level veto that also works without geometry.
On the base-law side, ContentFlattener now separates sibling <LA>
elements within a <DD> — the short-label/definition pairs of the UWG
Anhang and § 2 IfSG were previously glued without any separator
("Irreführung über Unternehmereigenschaftdie unwahre Angabe...") in both
synopsis columns. The continuation line is indented two spaces deeper
than its enumeration line so StellenAufloeser.zeilenBlock keeps it inside
the unit's block ("Anhang Nummer 31 Buchstabe b"). BefehlAnwender writes
the same canonical shape when inserting or recasting quoted units
(rueckeZitatEin), so XML-derived and PDF-derived items agree:
" 2a. Stichwort" + " Definitionstext".
Verified by mvnw verify (179 tests, 7 new) and a before/after sweep of
all 21 sample law/amendment combinations: applied/manual counts are
unchanged throughout (UWG 19/0, GEG-BGBl 66/53, GModG-RegE 77/20,
IfSG-0645 42/24, ...), no marker characters leak into the HTML output.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Change-Id: Ib7faae1bc5c59bda83f648a579af16a4c77e4285
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
indentation
Replaces the heuristic character-count full-width check with real geometric
layout analysis. FontgroessenFilter collects the right-edge X coordinate of
the text runs on each page and classifies line endings against the local
90th percentile of page margins (using a 20-line sliding window) into soft,
hard, or unclassified breaks.
TextBereiniger uses this classification to reflow only soft line wraps (WEICH) and
keep deliberate ones (HART or UNBEKANNT).
Also aligns quote normalization and GII-XML flattening to indent continuation
lines in lists (e.g., hanging definitions in UWG Anhang):
- BefehlAnwender.normalisiereZitatText applies to single-unit Neufassung and indents
continuation lines deeper (4 spaces) than list items (2 spaces).
- ContentFlattener generates the same structure for sibling <LA> elements inside <DD>.
This ensures the paragraph parser correctly groups these lines as child lines.
Change-Id: I1fd7039d4933a273cc2dc55f958d34b439e2302c
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
TextBereiniger.verbindeUmbrueche assumed any letter-ending, no-trailing-space
line followed by a lowercase continuation was a hyphen-less mid-word split
(PDF extraction artifact) and joined it with zero separator. That assumption
breaks for deliberate word-boundary breaks, e.g. the short-label/hanging-
indent definition format in the UWG Anhang ("...Nachhaltigkeitssiegels" +
"das Anbringen..."), producing glued words in the rendered synopsis.
Gate the markerless join on the candidate line reaching a locally-typical
"full column width" (90th percentile in a ±20-line window), since automatic
wraps always land near the column edge while deliberate breaks don't. The
window is local rather than document-wide because some source PDFs mix
column widths within one document (narrower Regelungstext vs. wider
Begründung), which a global statistic would otherwise penalize.
Also route the single-unit Neufassung fallback through the existing
normalisiereZitatText normalization, matching its sibling code paths, so
that internal line breaks now more often preserved by the fix above don't
leak into the HTML output as spurious line breaks instead.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Change-Id: I0a79f22f9286cd4a90eea32bad9b54cb4a082cf4
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
Gliederungs-Überschriften insertion/replacement
Fourth enablement wave: Anhänge/Anlagen resolve as ordinary norm targets with
nested Nummer/Buchstabe block resolution, InhaltsuebersichtAnwender applies
Angabe commands automatically, GliederungsUeberschriften handles inserting and
replacing structural headings, the law's own heading can be recast, and
FontgroessenFilter now determines body text per page instead of document-wide.
Sample data (UWG/AGG/ProdHaftG) now applies with 0 manual cases.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Change-Id: I1a76750afbecfa1aae734bcdabc91470a28420ca
|
| |
|
|
|
|
|
|
|
|
|
|
| |
blocks
Adds bisStelle to StrukturErsetzung for coordinated target ranges, recognizes
deletion/renumbering of whole structural units (paragraphs and Gliederung
entries), and handles insertion/replacement of multi-paragraph blocks split on
§-headings. Updates FASSUNGEN.txt accordingly.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Change-Id: I637cc5effbe32c094e62c070b63c470149a77c85
|
|
|
Record the changes of the "Fassung vom 13.7.2026" in bureaucratic legal
German, mirroring the amendment-act style the tool itself processes.
Convention: add a dated section (newest first) for every day the code
changes.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Change-Id: Ib39e4f78405b33b439be006325e2c310b72b1511
|
