legaldown-validator 📐

The reference implementation of LegalDown

Parse, validate, and serialize LegalDown documents, and assemble templates — from the command line or from Python.

Targets specification v0.2 · Every diagnostic carries a stable rule id · One dependency: PyYAML

---

What it does

LegalDown is an open plain-text standard for legal documents. This package is the toolkit for working with those documents in code:

📥 Parses a .lgd file into a typed document model — frontmatter, sections, blocks, parties, definitions, anchors.

✅ Validates it against the specification's rule set (§16): broken cross-references, undefined terms, duplicate identifiers, malformed party metadata, bad dates and money values, and more.

📤 Serializes the model back to LegalDown source, so you can edit documents programmatically and write them out again.

🧩 Assembles templates (§15.7): a template and an answers set in, the finished document out, byte for byte as the specification defines it.

🏷️ Names every finding. Each diagnostic carries the specification's stable rule id (§16.1) alongside its severity and message, so you can suppress one check, escalate another, or gate a build on exactly the rules you care about. Rule ids survive specification renumbering — they are the part of a diagnostic that is safe to depend on.

🖨️ Rendering a document to HTML or plain text, with its numbering, cross-references, and defined terms, is the job of legaldown-render, the reference renderer built on this package (pip install legaldown-render).

Install

pip install legaldown-validator

Python 3.11 or newer. The distribution is named legaldown-validator; the import name is legaldown:

import legaldown

The package ships a py.typed marker, so type checkers use its annotations directly.

Quick start

Given a LegalDown document:

---
title: Services Agreement
document_type: contract
sides:
  - name: providers
    label: Provider
    parties:
      - name: acme
        label: Acme
        type: legal_entity
        legal_name: Acme Corporation
  - name: clients
    label: Client
    parties:
      - name: beta
        label: Beta
        type: legal_entity
        legal_name: Beta Industries Inc.
language: en
---

# Fees {#fees}

{{party: beta}} shall pay {{money: 5000}} under {{ref: payment-terms}}.

Validate it:

legaldown validate contract.lgd

contract.lgd:24: error: [ref-broken] Broken section reference: 'payment-terms'.
contract.lgd:24: warning: [money-missing-currency] Money directive without currency parameter.

1 error(s), 1 warning(s), 0 info(s)

One diagnostic per line, each prefixed with its file and line (§16.9) and its rule id, in source order. A clean document reports No issues found. and exits 0.

Diagnostics go to stdout; the trailing summary and No issues found. go to stderr, so legaldown validate contracts/ > report.txt captures the findings alone. --quiet drops the summary entirely.

Command line

legaldown validate contract.lgd                      # one document
legaldown validate contracts/                        # a whole directory, recursively
legaldown validate --format json contract.lgd        # machine-readable output
legaldown validate --ignore def-unreferenced doc.lgd # mute one rule
legaldown validate --warnings-as-errors doc.lgd      # tighten a CI gate
legaldown validate --final signed-contract.lgd       # nothing left to fill in

Option	Effect
--format {text,json}	Output format (default text)
--ignore RULE_ID	Suppress a rule by its stable id — repeatable
--final	Check a document is ready for signature: a remaining placeholder, drafting note, or template construct is an error (§15.9)
--warnings-as-errors	Report warnings at error severity
--strict	Exit non-zero on any diagnostic, not just errors
--quiet	Drop the trailing summary line

Directories are searched recursively for *.lgd, *.legaldown, and *.legal.md, and legaldown --version reports the validator version.

Exit codes: 0 clean · 1 diagnostics found (errors, or any diagnostic under --strict) · 2 a file could not be read.

Changed in 0.4.0: a document that amends another (amends.file, §7.5) or declares LegalDown attachment files (§12.4) is now checked against the definitions those files declare, read from beside it (never from outside its directory). Before, legaldown validate never read them, so a term only the amended original defines was an Info (amend-term-unresolvable) and is now found, and an amend-def-override or def-duplicate-id can be reported that was not. A CI check on such documents can start to fail for that reason: it is the specification's result, and the files must be beside the document. A file that is not there, or cannot be read, is as if it had not been asked for.

Assembling a template

legaldown assemble template.lgd --answers answers.yaml          # the document to stdout
legaldown assemble template.lgd --answers answers.yaml -o out/  # with its fragments and attachment files

legaldown assemble template.lgd -i -o out/                         # asked in the terminal for what is missing
legaldown assemble template.lgd -i --answers a.yaml --save-answers a.yaml -o out/   # ... and remembered
legaldown questions template.lgd --answers answers.yaml           # what is asked, given the answers so far
legaldown questions template.lgd --answers answers.yaml --format json

assemble -i asks, in the terminal, for the answers --answers does not give, as the questions are reached — a decision opens or closes the questions under it. Each prompt shows the question, what to type (yes or no, an amount and a currency, like 5000 EUR) and the default in brackets, and each answer is checked as it is typed. A decision is asked again until it is answered; an empty line leaves a value question to its default, or its blank. The prompts go to standard error, so standard output (or -o) is still the assembled template. Without a terminal on standard input, -i lists what must still be answered and exits 1, or assembles if nothing must be. At the end of the input the interview stops and assembly goes on with what was answered (the questions left are left to their defaults or blanks, or refuse the assembly if they are decisions); Ctrl-C stops it (exit 2). A template that cannot be assembled asks nothing: its problems are reported first. --save-answers FILE writes the answers used, with those typed, as YAML — also after an interruption or a failure, so that typing is not lost — and the file is an answers set for the next run. It is written as a whole or not at all, beside the file and moved into place, so it may be the answers file that was read; but it is the answers, not the file's text, that are written, so comments in it are lost (with or without -i).

questions lists the questions reached, which are unanswered, which stop assembly, and what is wrong with the answers; --format json is Form.as_dict() (below), for another program or an agent. It exits 0, whatever the answers, unless the template has problems or cannot be read; --check makes a form that cannot be assembled exit 1.

The answers set is a YAML mapping of question ids to answers (§15.7.1); the shapes plain YAML gets wrong are put right for the template's questions (5000 EUR, yes, a number as an amount). Include fragments and LegalDown attachment files are read relative to the template, never from outside its directory; the ones assembly keeps are written under -o at their relative paths. Answer problems are reported as answer-invalid, answer-missing and answer-unknown (§16.12) on stderr, and nothing is written when there is an Error. A template with Errors in the validator's template rules (questions, conditions, {{choose:}}, placeholders, insertions) is refused, with the rule ids, as §15.7.2 gives assembly a valid template. Exit codes are as for validate.

Changed in 0.4.0: legaldown assemble refuses a template that has Errors in the validator's template rules (question-invalid, condition-invalid, choose-invalid, the placeholder-* rules, insertion-boundary, def-term-variable, drafting-note-def, template-fragment-invalid, ...), with exit 1 and the rule ids on stderr, where 0.3 assembled it. §15.7.2 gives assembly a template that validates; a gate that assembles templates in CI can start to fail on one that never did. --save-answers does not write when there was nothing to save (no answers file read, nothing typed), so that an answers file you meant to resume from is not replaced by an empty one.

JSON output

--format json emits the structured shape described in §16.9 — ideal for CI annotations, editor integrations, and dashboards:

{
  "legaldown_spec": "0.2",
  "validator_version": "0.4.0",
  "diagnostics": [
    {
      "file": "contract.lgd",
      "rule": "ref-broken",
      "level": "error",
      "line": 24,
      "message": "Broken section reference: 'payment-terms'."
    },
    {
      "file": "contract.lgd",
      "rule": "money-missing-currency",
      "level": "warning",
      "line": 24,
      "message": "Money directive without currency parameter."
    }
  ]
}

In CI

- run: pip install legaldown-validator
- run: legaldown validate contracts/ --warnings-as-errors

The exit code fails the job; the rule ids let you grant exceptions without turning whole checks off.

Python API

Three functions cover the common path:

from legaldown import load, serialize, validate

document = load("contract.lgd")
result = validate(document)

for diagnostic in result.diagnostics:
    print(diagnostic.level, diagnostic.rule, diagnostic.message)

if result.is_valid:                      # no Error-level diagnostics
    print(serialize(document))           # the source text; save(document) writes it to the file

load(path) takes a str or os.PathLike, reads the file as UTF-8 and returns the Document. It names the document for you: document.filename is the file's name and document.path its absolute path (symbolic links are not followed), which is where the files it refers to are to be looked for. For text that is not in a file (an editor buffer, an HTTP body, a test), use parse(text), which load is built on; it takes an optional filename= to name the document in its diagnostics.

load raises FileNotFoundError (or another OSError) when the file cannot be read and UnicodeDecodeError when it is not UTF-8. Both it and parse raise FrontmatterError (a yaml.YAMLError and a ValueError) when the frontmatter cannot be read — the frontmatter-invalid-yaml rule, which validate, given a parsed document, cannot report.

validate(document) checks it and returns a ValidationResult; validate(document, final=True) is the signature-ready check (§15.9). A document that amends another (amends.file, §7.5) or declares LegalDown attachment files (§12.4) is checked against the definitions those files declare. They are read from beside the document — document.path is where load found it — and never from outside its directory (§2.3); only the definitions each file declares itself are read, not those it refers to in turn. A file that is not there, is not UTF-8, or has frontmatter that cannot be read is as if it had not been asked for: no diagnostic of its own. A document from parse(text) has no path, so there is nothing to read (resolve=lambda path: None does the same for a loaded one, to validate it without reading anything); give validate a resolve= function, from a relative path to the file's text (or None), to read them from elsewhere — a database, an upload. Such a function is a LoadFile; file_loader(directory) makes one for files on disk, and parse_template takes one as resolve= too.

Deprecated: parse_document is now parse (string; same arguments) or load (file), validate_document is now validate, serialize_document is now serialize (text) or save (file), and the importer callbacks import_definitions= and import_attachment_definitions= (and their types, DefinitionsImporter and AttachmentDefinitionsImporter) are replaced by resolve=. They still work, and calling one raises a DeprecationWarning; they are deprecated since 0.4.0 and will be removed in 0.5.0 (the template functions too: Assembly). Every deprecation warning legaldown raises is a legaldown.LegaldownDeprecationWarning (a DeprecationWarning subclass), so one filter catches them all: warnings.simplefilter("error", legaldown.LegaldownDeprecationWarning) in code, or filterwarnings = ["error::legaldown.LegaldownDeprecationWarning"] in pytest's configuration (Python's own -W option and PYTHONWARNINGS ignore a category from a package, as they read it before the package can be imported).

Changed in 0.4.0: ValidationResult is a plain value of two parts, diagnostics and index. errors, warnings and infos are read-only lists taken from diagnostics (changing them changes nothing; change diagnostics). What validating resolved, besides what it found, moved to result.index (below): result.sections is now result.index.sections, result.is_template is result.index.is_template, and so on, and the five inline_* lists are result.index.values.dates, .money, .durations, .fields and .placeholders. The old names are gone. So are the result's recording methods (error, warning, info, at) and used_terms, which only the validator used.

The public API is what the legaldown and legaldown.validator packages export (their __all__), and legaldown.grammar and legaldown.syntax. legaldown is the workflow and the model: loading, validating, saving and assembling documents, what that returns, and the Document dataclasses. legaldown.syntax and legaldown.grammar are the tooling API (see Tooling API): reading source text the way the validator does, and the language's constants and rules. Changed in 0.4.0: the lexer, the fragments, the definition readers, render_block, the dict factories and the like moved out of legaldown.__all__, so from legaldown import * no longer brings them. Imported by name they still import from legaldown, as the same objects and without a warning, so code written for earlier releases keeps working; new code imports them from their home (the table at the end of Tooling API). Changes to the API are listed in the notes of each GitHub release; before 1.0 a minor release may change it, a patch release does not. Other modules are internal and may change in any release.

Working with the result

Validating a document builds the indices the checks need — section numbers, resolved definitions, party display text, every inline value found in the body. result.index, a DocumentIndex, hands all of it back, so a renderer or a UI can reuse the work instead of re-deriving it. The result itself is what was found, kept nowhere but in diagnostics:

result.…	Contents
diagnostics	Diagnostic(rule, level, message, line, file) — the authoritative record
is_valid	True when no Error-level diagnostic was reported
errors / warnings / infos	Message strings by severity, taken from diagnostics
rules(level=None)	Set of rule ids present, optionally filtered by severity
index	A DocumentIndex. Values and markers in it are as written: one that is invalid is reported in diagnostics too, so check is_valid before relying on them:

result.index.…	Contents
sections, section_lookup	Numbered section index, as SectionIndexEntry(title, identifier, path, level, number, alternative), in document order and by identifier (a paragraph's or item's anchor too, to its section); resolves {{ref:}} targets. Numbers count from the shallowest heading level, and a level a heading skips counts as 1 (#, ###, ## → 1, 1.1.1, 1.2), so no two sections share a number except alternatives and what they contain (§15.8); an entry's alternative is true for a section that is an alternative to its preceding sibling — the section just before it at its level under the same parent, with the same identifier, never present together with it (§15.4) — and so shares its number (§15.8)
definition_lookup, party_lookup, side_lookup, attachment_lookup	Resolved display text
values	The field-spec values the checks met, as written, as InlineValues: dates, money, durations, fields, placeholders (those of the frontmatter too; one with malformed arguments is not among them)
blanks	The document's blanks, a template's or not, by placeholder id in order of first occurrence: Blank(id, type, fixed, in_frontmatter, consistent) — type the type the validator settled on: the effective type of the first occurrence with a valid one (§10.7, §15.2; text when none is written), or None when no occurrence has a usable type (each is written with an invalid one, or the id is a decision question's); fixed the currency of a money blank or the unit of a duration blank that every occurrence fixes ("" when one fixes none, or the blank is not consistent); consistent false when its occurrences have different types or fix two currencies or units (placeholder-type-inconsistent); in_frontmatter whether one occurrence is in the frontmatter (§3.10)
is_template	Whether the document is a template (§15.1): it declares questions, carries a condition, or holds a {{choose:}}
placed_markers	The markers in body text that apply (§5.7, §15.3), in document order: PlacedMarker(section, block, fragment, offset, source, identifier, condition, field, item, include_only, line) — in fragment fragment of block_fragments(block) (legaldown.syntax), at offset, which is the block's field (text, or suffix after a lifted {{ref:}}/{{term:}}); item is the list item it marks, counted in pre-order over all the list's items, nested and empty ones included, as list_fragments counts them; identifier is "" where it does not apply (an include-only paragraph, §12.2). Identifiers and conditions are as written: check is_valid before relying on them

is_template(document) (top-level) gives the template decision without validating (§15.1). A renderer builds from these decisions rather than re-deriving them, reading the source with the validator's own helpers (Tooling API). legaldown-render is built this way.

Reading and editing the document model

load and parse return a Document of plain dataclasses — Metadata (with Amends for amends and an object supersedes, and Attachment), Section, Block and ListItem, Side, Party (with Representative and CustomField) — that you can inspect, edit, and write back out:

from legaldown import load, save

document = load("contract.lgd")
document.metadata.governing_law = "Czech Republic"

save(document)                           # to document.path, where load found it

save(document, path=None) mirrors load: it writes the document to path (a str or os.PathLike), or to document.path when it has none to be given, and returns the absolute path. A document from parse(text) or built in code has no path, so save raises ValueError until you pass one. The write is atomic — the file is either as it was or the new text, never half written — a symbolic link is written through, and a file keeps its permissions. Missing directories on the way are created. The text is UTF-8 with LF line endings and no byte-order mark, so a file that was read with CRLF line endings or a byte-order mark is written back without them. Afterwards document.path is the path written, as after load: that is where validate looks for the files the document refers to (and document.filename, which diagnostics name, is its name). serialize(document) is the same text as a str, the inverse of parse, for when the file is not what you want to write (an HTTP response, a database).

Content before the first heading — typically the sentence identifying the parties — is the document's preamble (§4.4): it is unnumbered, so it lives in document.preamble rather than in sections. document.iter_blocks() walks every body block, preamble first.

A Section's identifier is the explicit {#id} written after its heading, or "" when it has none. The identifiers the validator generates (§5.3, §5.5) are not written back into the model: read them from result.index.sections.

document.to_dict() and Document.from_dict(data) round-trip the model through JSON-friendly structures (missing fields take their defaults), except the fields that describe the parsed source rather than the document, Metadata.not_line_editable, Metadata.frontmatter_absent, Document.source_map and Document.path. A blank document is Document(); empty_document(), still importable from legaldown, is not one but a starter contract with two sides and two sections.

Each diagnostic names its line (from 1) and its file (the document's filename), as §16.9 requires: the line of the directive, marker, heading or block it is about, or of the frontmatter key — for a missing key, the key that holds it, or the frontmatter's first. Lines come from the source_map that load and parse give a document. A document built from a dict has none, and one changed after parsing no longer fits its map: their diagnostics have no line (None) rather than a stale one. FrontmatterError.line is the line of YAML that cannot be read. render_block() (legaldown.syntax) renders a single block when you are driving your own layout. iter_directives() (legaldown.syntax) lexes the directives in a piece of text by the §11.2 grammar — parameters in any order, quoted values decoded — and is what the validator itself uses.

definition_lookup(collect_definitions(document)) (legaldown.syntax) gives a document's definitions, id to term, without validating it: the same map as result.index.definition_lookup, except for definitions imported from an amended original or an attachment file. A term written empty ("" {{def: x}}) reads as its id (id_term), and an id that is not a valid identifier is left out.

One guarantee worth knowing: the parser is faithful — it never rewrites your input to make it valid, so what you authored is exactly what the validator judges. The serializer, by contrast, normalizes: frontmatter is re-emitted as canonical YAML, setext (underlined) headings are written as # headings, list items are written with - (+ where - would read as a thematic break) or 1., 2., … whichever marker they had, each nested list at its parent item's content and numbered on its own, without blank lines between items, and table rows are written with a pipe at each end and every | in a cell escaped as \|, so expect a formatting-normalized file rather than a byte-for-byte copy.

What it normalizes is formatting, never content: what the validator reports of a document is what it reports of the document written back. So the serializer writes every side, party, representative and attachment the model holds, one not yet filled in too (a side without parties, an attachment without a title), where the validator reports what it lacks. A party's custom fields (§3.4) are the keys of its object that are none of its fields; they are read into custom_fields and written back as keys of their own, a row whose label is still empty as the key ''. When keys cannot hold them all (a label written twice, or naming one of the party's fields), they are written as the party's custom_fields list of label and value entries, which is read too; a key named custom_fields is that list's, never a custom field. A value that is a list or a mapping is not kept: a custom field holds text.

A list block's items are ListItems, each holding the blocks of its content in order: its first paragraph, then later paragraphs, headings, nested lists (a list block of their own kind), code, quotes, tables and raw HTML. An empty item holds none. A paragraph in an item is always of kind paragraph: a directive in it stays in its text. A heading in an item or a quote is a heading block, not a section: its text without the # syntax, and its level (1–6; level is 0 on every other block). Written outside an item, where it would be a section's heading, a heading block becomes paragraph text with a backslash before it. item_text(item) gives an item's first paragraph's text, and render_item(item) its content as written after its marker (both in legaldown.syntax); a string item in Document.from_dict (as models before 0.3 held them) is read as such content. A nested item's condition applies within those of the items it is nested in (§15.3).

A quote block keeps its text: its content, one line per source line without the > marker. The validator and assembly read that content as blocks, as they read an item's, so code and raw HTML in a quote hold no directive.

Tables follow GFM: a table needs a delimiter row with one cell per header, rows take the header's width, and a | inside a cell — a code span's included — is written \|. A table block's headers and rows hold the cell text with those escapes removed, and align holds each column's alignment ("left", "right", "center", or "").

A paragraph's text (a definition's, and a ref's or term's prefix and suffix, too) keeps its lines, joined with \n and without their indentation, as CommonMark reads them: a line ending in \ or in two spaces is a hard line break, and each other line break a soft one, which a renderer shows as a space. Link reference definitions stay on their lines. It is written back as those lines; a line that would, at the margin, end the paragraph (a list item, a heading, a setext underline, a table's delimiter row, and the like — only a lazy line or a model built in code holds one) is written four columns in, where it continues the paragraph and is read back as it was. A blank line in a model's paragraph, which would end it, is left out.

Lines indented four or more columns are an indented code block, as in CommonMark, so a paragraph indented that far is code: its directives and anchors are literal (§11.4). A paragraph that a model holds but that would, written at the margin, open a heading, a fence, or an HTML block is written with a backslash before it.

To write markdown source exactly as typed — an editor's row holding a list, a table or a quote — put it in a source block ({"kind": "source", "text": "* a\n* b"}): the serializer writes its text as it stands, respelling and escaping nothing, and reads it back as the blocks it holds. Only blank lines at its ends are dropped, and a fence it leaves open is closed unless it ends the document. It is written next to its neighbours as it stands, so it means what that text means there: in a list item it is the item's content, and a list in it right after a list with the same bullet joins that list. The parser never produces a source block, and the validator does not read one: validate the written text.

Raw HTML follows CommonMark's HTML blocks: a block of raw HTML, or an HTML comment on lines of its own, is an html block holding its source as written. No heading, directive, or anchor inside it is recognized (§8.6, §11.4), so a clause commented out with <!-- … --> is not a section. As in CommonMark, a comment left unclosed runs to the end of the document.

Tooling API

Tools built on the validator (renderers, editors, importers) need the language itself: its vocabulary, its value formats, and how the validator reads source text. Two modules hand that over. They are part of the supported API and covered by the same versioning as legaldown and legaldown.validator, and they are the supported way to reach these names: every module path below them (legaldown.markdown, legaldown.validator.patterns, …) is internal and may be reorganised in any release (some of the names also still import from legaldown, for compatibility). Their names are the validator's own objects, not copies, so a tool and the validator cannot disagree.

legaldown.grammar is what needs no document: constants and small rules.

from legaldown.grammar import (
    IDENTIFIER_RE, KNOWN_CURRENCIES, MAX_SECTION_LEVEL, PARTY_TYPES, VALID_PLACEHOLDER_TYPES,
    is_valid_iso_date, slugify_identifier,
)

is_valid_iso_date("2026-02-30")                 # False
slugify_identifier("1st Term")                  # "section-1st-term" (§5.3)
slugify_identifier("日本語", fallback="")        # "": nothing usable, as against real text

Names	What they are
SPEC_VERSION, LEGALDOWN_EXTENSIONS	The specification version implemented; the file extensions of LegalDown source (§2.1)
KNOWN_DIRECTIVES, DIRECTIVE_PARAMS, PLACEHOLDER_TYPE_PARAMS	The directive vocabulary and the parameters each defines (§11)
IDENTIFIER_RE, RESERVED_VALUE_TYPES, VALID_DOC_TYPES, PARTY_TYPES, KNOWN_CURRENCIES, DELIMITER_PAIRS	Identifier format (§5.3), reserved value-type names, document types, party types, ISO 4217 codes, accepted definition delimiters (§7.2)
VALID_PLACEHOLDER_TYPES, DURATION_UNITS, VALID_DURATION_UNITS	Placeholder types and duration units (§10); DURATION_UNITS is in the order the specification lists them
LIST_KINDS, MAX_SECTION_LEVEL, MAX_QUOTE_DEPTH, MAX_LIST_DEPTH	Block kinds that are lists; the deepest heading level (5); how deep the parser reads quotes and lists
VALUE_QUESTION_TYPES, DECISION_QUESTION_TYPES, QUESTION_TYPES, DRAFTING_MARKER, FINAL_CHECK_RULES	Template question types (§15.2); the drafting note's first line (§15.6); the rule ids of the final check (§15.9)
slugify_identifier(value, *, fallback="section"), format_section_number	The §5.3 identifier of a heading or term text, and fallback where the text yields none ("" tells that case apart); a dotted section number
is_valid_iso_date, is_valid_numeric, is_valid_money_amount, is_positive_numeric	Value checks (§3.10, §10)
parse_condition → Condition, condition_problem, exclusive, Presence, ALWAYS	Conditions (§15.3, §15.4): parse one, tell why one is invalid, tell whether two units can never appear together
choose_problem(directive, questions)	Why a {{choose:}} is invalid (choose-invalid, §15.5): the first message validate records for it, or None; ValueError for a directive that is not a {{choose:}} or is malformed (validate reports that as directive-malformed)

legaldown.syntax is reading source text the way the validator does.

from legaldown import load
from legaldown.syntax import block_fragments, find_markers, lex

document = load("contract.lgd")
for found in find_markers(document):             # every marker and look-alike in the body
    print(found.source, found.placed(template=False), found.misplaced)
for _section, _index, block in document.iter_blocks():
    for text, _anchor in block_fragments(block):
        print([d.name for d in lex(text).directives])

Names	What they are
lex, Lexed, Directive, iter_directives, is_escaped, format_value	The directive lexer (§11.4) and its inverse. A Directive's positional_span and param_spans are where each value is written, quotes included, so text[start:end] is what to replace to change it; Lexed.literals are the comments and code spans the lexer skipped, as (kind, start, end)
iter_document_directives(document) → DirectiveLocation(section, block, fragment, directive), collect_source_directives	Every directive in a document's body, in order, read as validate reads it: fragment indexes block_fragments(block), and is None for a {{ref:}}, {{term:}} or {{def:}} the parser lifted into a block's fields (a definition's comes before its text); the {{ref:}} and {{term:}} targets of a document
Marker, MARKER_RE, parse_marker, format_marker, is_look_alike	Anchor and condition markers (§5.7, §15.3)
find_markers(document), FoundMarker, is_include_only	Every marker and look-alike in a document's body, placed or not; FoundMarker.placed(template) says which apply (validate gives the placed ones as result.index.placed_markers)
Fragment, ListFragment, block_fragments, list_fragments, text_fragments, list_items, item_text	Where a block's text is, and a list's items
collect_definitions(document) → DefinitionRef, definition_lookup, id_term	The definitions a document declares in its own source (§7), in document order, each with where it is; definition_lookup(refs) is their id-to-term map as validate builds result.index.definition_lookup (but for definitions imported from other files); id_term(id) is the term an empty one reads as
find_definition_anchors(text) → DefinitionAnchor	Every {{def:}} in a text, with the quoted term it anchors and where that term starts (§7.2)
render_block(block), render_item(item)	The other way: a block written as LegalDown source, and a list item's content as written after its marker, as serialize writes them
Quote, block_quotes, is_drafting_note(block)	The block quotes in a block and whether each is a drafting note; whether a quote block is one (§15.6)
quote_blocks(block, *, depth=0), drafting_note_blocks(block, *, depth=0)	The blocks a block quote holds, as the validator reads them, as copies you may change; a drafting note's without its [!DRAFTING] marker (§15.6). depth is how many list items and quotes the quote is in, the count before entering it (a quote in a list item is at 1); past MAX_QUOTE_DEPTH the validator reads a quote as one text, so you get one paragraph of its text as written
code_content(block) → CodeContent(info, text, fenced)	A code block read as CommonMark reads it: the info string, and the code without its fences or indentation
FRONTMATTER_RE, LINE_ENDING_RE, HTML_COMMENT_RE, FENCE_OPEN_RE, closes_fence, fence_end, dedent, indent_width, strip_text	The Markdown rules the reading is built on: frontmatter, line endings, comments, fenced code, indentation

Where things are in the file. document.layout() gives a SourceLayout: the frontmatter's (start, end), the preamble's BlockSpans, and per section a SectionSpan (its HeadingSpan and its blocks' spans), in document.sections order. Lines are file lines counted from 1, end exclusive; a list's BlockSpan has its ItemSpans, each with the spans of its blocks. document.line_of(section, block=None, item=None) is one line: a section's heading, a top-level block (section=None for the preamble), or a list item's marker, items counted in pre-order as PlacedMarker.item counts them. These are the lines diagnostics name (§16.9). Both are None for a document built in code, or changed since it was parsed. A span covers the lines [start, end); the blank lines between parts belong to none (but for those between a list's items, which the item before holds). For text you hold apart from its file — an editor's field — body_layout(text) is the same SourceLayout of that text read as a body alone, with lines counted from 1 at its start and no frontmatter (a --- first line is a thematic break there).

document = load("contract.lgd")
for number, section in enumerate(document.layout().sections):
    print(document.sections[number].title, section.heading.start, [b.start for b in section.blocks])

If you import one of these from a private module, use its public home. The names that left legaldown.__all__ in 0.4.0 are listed too, as guidance only: importing them from legaldown keeps working, without a warning.

From	Use
legaldown: lex, Lexed, Directive, iter_directives, is_escaped, collect_source_directives	legaldown.syntax
legaldown: Fragment, ListFragment, block_fragments, list_fragments, list_items, item_text, is_drafting_note	legaldown.syntax
legaldown: collect_definitions, definition_lookup, id_term, DefinitionRef, find_definition_anchors, DefinitionAnchor	legaldown.syntax
legaldown: render_block, render_item	legaldown.syntax
legaldown: DIRECTIVE_PARAMS, KNOWN_DIRECTIVES, DELIMITER_PAIRS, slugify_identifier	legaldown.grammar
legaldown: document_from_dict(data), document_to_dict(document)	Document.from_dict(data), document.to_dict()
legaldown: empty_document, BLOCK_DEFAULTS, metadata_from_dict, section_from_dict, block_from_dict, side_from_dict, party_from_dict	No other home: they stay importable from legaldown; Document() is a blank document, and Document.from_dict builds every part
legaldown.directives: PLACEHOLDER_TYPE_PARAMS	legaldown.grammar
legaldown.directives: format_value	legaldown.syntax
legaldown.definitions: text_fragments	legaldown.syntax
legaldown.definitions: DELIMITER_PAIRS	legaldown.grammar
legaldown.markdown: FENCE_OPEN_RE, HTML_COMMENT_RE, LINE_ENDING_RE, closes_fence, dedent, fence_end, indent_width, strip_text	legaldown.syntax
legaldown.markers: MARKER_RE, Marker, format_marker, is_look_alike, parse_marker	legaldown.syntax
legaldown.models: LIST_KINDS	legaldown.grammar
legaldown.parser: FRONTMATTER_RE	legaldown.syntax
legaldown.parser: MAX_QUOTE_DEPTH	legaldown.grammar
legaldown.validator.patterns: LEGALDOWN_EXTENSIONS, IDENTIFIER_RE, KNOWN_CURRENCIES, VALID_DOC_TYPES, VALID_DURATION_UNITS, VALID_PLACEHOLDER_TYPES	legaldown.grammar
legaldown.validator.templates: DECISION_QUESTION_TYPES, QUESTION_TYPES	legaldown.grammar
legaldown.validator.templates: block_quotes	legaldown.syntax
legaldown.validator.templates: check_choose	legaldown.grammar.choose_problem
legaldown.validator.units: find_markers, is_include_only	legaldown.syntax
legaldown.parser: quote_content	legaldown.syntax.quote_blocks (and drafting_note_blocks)
legaldown.parser: _layout	legaldown.syntax.body_layout(text) for bare body text (lines from 1 at its start); Document.layout() for a parsed document (file lines)
your own fence stripping (FENCE_OPEN_RE, closes_fence, dedent)	legaldown.syntax.code_content
legaldown.cli: _read_answers	legaldown.load_answers

What gets checked

The full rule set with severities and examples lives in the specification (§16); this is the map:

Area	Checks include
Structure	Heading depth and skipped levels, hardcoded section numbers, missing title, raw HTML other than comments
Directive syntax	Malformed directives, repeated parameters, parameters a directive does not define, unquoted values that begin with a curly quote, stray {{
Cross-references	{{ref:}} targets that do not exist or point at an attachment
Anchors	Duplicate identifiers, malformed identifiers, auto-generated collisions and lost letters, markers outside an anchor position
Definitions	Undefined {{term:}}, duplicate ids, auto-generated ids that lost letters, missing quoted span, ambiguous quoting, unreferenced definitions
Parties and sides	Unknown {{party:}} / {{side:}}, malformed or duplicate names, invalid party types, minimum party and side counts, empty representatives
Values	Invalid dates, money without currency or with an unknown one, invalid durations and units, undeclared or reserved custom field types
Placeholders	Malformed ids, invalid types, one blank with two types, currencies, or units, placeholders in structural and format-checked frontmatter fields
Templates	Malformed question declarations and defaults, placeholders that contradict their declared question, invalid or contradictory conditions, references whose target a condition can remove, identifiers shared by units that can appear together, {{choose:}} that misses or invents an answer, definitions inside drafting notes and mistyped [!DRAFTING] markers, blanks in a defined term or against Markdown punctuation, fragments included twice; with --final, anything left unfilled; on assembly, answers that are invalid, missing, or unknown
Attachments	Undeclared {{attach:}}, duplicate or colliding ids, empty titles, unreferenced attachments
Amendments	Terms the amended original does not define, definition overrides, empty amendment titles
Metadata	No frontmatter at all, invalid document type, invalid dates, missing sides, issuer side requirements, empty supersedes title, a declared legaldown version newer than 0.2

Each check reports at the severity the specification assigns it — Error, Warning, or Info.

Assembly

from legaldown import load_template

template = load_template("nda.lgd")     # reads it, and the files it includes, once
template.questions                      # every question, declared and implicit (§15.2)
template.problems                       # why it cannot be assembled, if it cannot

form = template.form({"forum": "courts", "fee": {"amount": "5000", "currency": "EUR"}})
form.questions     # the questions reached so far, in order: ask these
form.unanswered    # ... of them, with no valid answer and no default
form.blocking      # ... of those, the decisions: assembly cannot run without them
form.diagnostics   # what is wrong with the answers given
form.ready         # assembly can run
form.complete      # ... and nothing is left blank

result = form.assemble()
if result.ok:
    contract = result.output        # the assembled template file
    files = result.files            # assembled fragments and attachment files, by relative path
for diagnostic in result.diagnostics:
    print(diagnostic.level, diagnostic.rule, diagnostic.message)

form.assemble() returns an AssemblyResult: ok (no Error), output, files (an emptied fragment is "", written as zero bytes) and diagnostics; nothing is assembled while an Error stands, and output and files are then empty. AssemblyError is raised, rather than a guess made, if the template's source cannot be mapped onto its parsed structure — never expected for a template the parser reads, so report it if you meet it.

A template is read once; a form is a snapshot of the interview over it, so a front end asks for a new form after every answer — a decision opens or closes the questions under it. parse_template(text) (or Template(text), which takes the same resolve=) is the same for source text; template.path is the file load_template read, and None for a template from its text. load_template raises FrontmatterError for frontmatter that cannot be read, as parse does.

What a form says. questions are those the assembly reaches given the answers so far: a decision question when a condition or {{choose:}} using it lies in a present unit, a value question when one of its placeholders does. A question with a default is among them (the default answers it, and a front end can still offer it). blocking are the decision questions without an answer or default, that a condition or {{choose:}} depends on: an unanswered value question leaves its blank, which the specification allows, but an unanswered decision is not assembled. An answer that is not valid counts as no answer, and does not stand in for the default: it is reported in diagnostics (answer-invalid), and so is an answer to a question the template does not have (answer-unknown, a warning). unused lists the questions that have an answer but are not reached; it is advice, since one answer that changes can make many of them unused. form.problem("fee") says why one answer is not valid, and so does question.problem(answer), before anything is assembled. A form never changes after it is made, and neither do the questions: a Question is a fixed value, a copy of the template's declaration, which every form of the template shares (treat its default and choices as read-only).

Asking a person. A question can read what a person types, so a terminal, a web form or an agent need not know the shapes an answers set holds (money is {amount, currency}, a boolean is a boolean):

answers, skipped = {}, set()
while True:
    form = template.form(answers)
    q = next((q for q in form.questions if answers.get(q.id) is None and q.id not in skipped), None)
    if q is None:
        break
    try:
        answer = q.from_text(input(f"{q.prompt or q.id} — enter {q.hint}: "))
    except ValueError as exc:          # says what to enter
        print(exc)
    else:
        if answer is not None:
            answers[q.id] = answer
        elif q in form.blocking:       # a decision cannot be left open
            print("An answer is needed.")
        else:                          # the default applies, or the blank stays
            skipped.add(q.id)

question.hint is what to type — yes or no, a date, YYYY-MM-DD, an amount and a currency, like 5000 EUR, one of: courts (State courts), arbitration — and question.from_text(text) turns the text into the answer, or raises ValueError saying what to enter; the empty text is None, no answer. What it returns always passes question.problem. It is strict and not locale-aware: yes, no, true, false, y, n for a boolean; a choice by key or label, in any case (a label two choices share is ambiguous, and a key comes before a label); 5000 EUR for money and 30 D for a duration, with the amount alone where every placeholder fixes the currency or the unit; no grouping separators, symbols or 5 000,50.

Answers from a file. load_answers("answers.yaml") reads the YAML mapping (AnswersError if it is not one; a date such as 2026-13-45 is not YAML). Plain YAML gets some shapes wrong for a template: fee: 5000 is a number and not the string money wants, forum: State courts names a choice by its label, a yes is not what a form takes. template.coerce(answers) puts right what it can read without guessing — text as question.from_text reads it, a money amount given as a number — and leaves the rest, a float or an unknown id, for the form to report; it never raises. A form does not coerce by itself, so that its diagnostics stay the specification's; legaldown assemble and legaldown questions do.

answers = template.coerce(load_answers("answers.yaml"))
form = template.form(answers)

The form as data. form.as_dict() is the form as JSON-ready data for a web form, a service or an agent: ready, complete, the template's problems, the diagnostics about the answers (each with the question it is about), and questions — those reached first, in order, then the others — each with id, type, label, prompt, declared, choices, currency, unit, reached, blocking, state (answered, default, invalid, unanswered), answer, default, problem, hint in words and accepts as data: the words of a boolean, the choices, the currency or unit the placeholders fix, the units there are. question.to_text(answer) is what a person would type for an answer (yes, 5000 EUR, 30 D), to show a default or fill in an input; answer_text and default_text hold it in the data.

What stops a template. template.problems are the reasons a template cannot be assembled whatever the answers: a file it includes that cannot be read, a placeholder written across lines, a translation group, and the Errors of the validator's template rules (question-invalid, condition-invalid, choose-invalid, the placeholder-* rules, insertion-boundary, def-term-variable, drafting-note-def, template-fragment-invalid; §15.7.2 gives assembly a template that validates, and condition-reference-unsafe counts too, for a template that includes no fragment: the validator cannot see which sections a fragment holds), and a placeholder, outside a drafting note, that would fill in a directive with a repeated parameter (directive-duplicate-param) or a duration unit §10.5 does not define (duration-invalid-unit). template.validation is the full ValidationResult of the template read alone, as advice: the validator does not read the fragments a template includes, so it cannot see a section or a definition that lives in one. ready is false while there are problems, and form.assemble() then returns them as its diagnostics.

Assembly edits the template as written — "no other byte of the template changes" — so its output is identical to any other conforming implementation's. A template with include fragments or LegalDown attachment files reads them from beside the file (load_template) or through parse_template(text, resolve=...), a function from a relative path to the file's text; without one such a template is refused rather than assembled partially (§17.6). The files it reads are checked as the Full level checks them — no frontmatter, no level 1 heading, and in a template no includes, and no conditions or drafting notes in a fragment — and a template whose files fail is refused. CONFORMANCE.md lists what assembly checks and what it does not.

Deprecated: assemble(text, answers, load_file=), template_questions and needed_questions are replaced by Template and Form. They still work, unchanged (but for Question, below, and that a load_file= loader is never asked for a path that is absolute or leads out of the template's directory, §2.3), and raise a LegaldownDeprecationWarning; they are deprecated since 0.4.0 and will be removed in 0.5.0. Two things differ in the new API: it asks the questions of an included fragment where its {{include:}} is (the functions ask them after the body), and it refuses a template that has Errors in the validator's template rules (the functions assemble it). legaldown assemble follows the new API, and puts the shapes of its answers file right (template.coerce): 5000 EUR is accepted for a money question (and fee: 5000, where its placeholders fix the currency), no for a boolean, and surrounding spaces of a text answer are dropped, which were answer-invalid before. A Question is now a fixed value.

Scope

This implementation claims Level 1 — Core (§17.2) and the Assembly capability (§17.6): everything above applies to a single document. Beyond the file you point it at, validation reads only the definitions declared by the document it amends and by its LegalDown attachment files (§7.5, §12.4), and assembly the include fragments and LegalDown attachment files of a template — each from beside the document, never from outside its directory, or through resolve=. That covers authoring, editing, CI validation, and assembly of individual documents.

It is verified against the specification's own fixtures corpus — every rule it implements passes, bar eight whose fixtures span several files or are marked for the Full level, and are covered by unit tests instead. The specification (§17.5) requires an implementation to be explicit about the checks it does not perform, so those are listed in CONFORMANCE.md rather than left to be discovered.

Development

git clone https://github.com/ForLegalAI/legaldown-validator
cd legaldown-validator
pip install -e ".[dev]"
ruff check .
pytest

CI runs on demand only, as one job on Python 3.12 (the version releases are built with): lint, the tests with the specification fixtures corpus, and the build with its metadata check. Start it with Actions → CI → Run workflow on the branch you pick (the conformance input switches the corpus off for a quicker run), or by adding the ci label to a pull request: it then runs again on each push to that pull request until the label is removed. Nothing else triggers it, so run it before merging and before a release.

Conformance suite

The fixtures corpus lives in the specification repository, so point the harness at a checkout:

git clone https://github.com/ForLegalAI/LegalDown ../LegalDown
LEGALDOWN_FIXTURES_DIR=../LegalDown/fixtures pytest tests/conformance -q

Cases for rules outside Core are skipped and named, so the run doubles as the coverage ledger in CONFORMANCE.md, which accounts for every skip. CI runs it with the rest of the checks, on demand (above).

Bug reports and pull requests are welcome in Issues; questions about the format itself belong in the specification repository's Discussions.

License

MIT — see LICENSE. The LegalDown specification itself is published separately under CC BY 4.0; it permits implementations to choose their own license.