rumdl_lib/rules/

md063_heading_capitalization.rs

/// Rule MD063: Heading capitalization
///
/// See [docs/md063.md](../../docs/md063.md) for full documentation, configuration, and examples.
///
/// This rule enforces consistent capitalization styles for markdown headings.
/// It supports title case, sentence case, and all caps styles.
///
/// **Note:** This rule is disabled by default. Enable it in your configuration:
/// ```toml
/// [MD063]
/// enabled = true
/// style = "title_case"
/// ```
use crate::rule::{Fix, LintError, LintResult, LintWarning, Rule, RuleCategory, Severity};
use crate::utils::header_id_utils::{HTML_TAG_ATTRIBUTES_PATTERN, HTML_TAG_NAME_PATTERN, is_backslash_escaped};
use crate::utils::html_elements::is_void_element;
use crate::utils::mdg;
use crate::utils::range_utils::byte_to_char_count;
use regex::Regex;
use std::collections::HashSet;
use std::sync::LazyLock;

mod md063_config;
pub(super) use md063_config::{HeadingCapStyle, MD063Config};

// Regex to match inline code spans (backticks)
static INLINE_CODE_REGEX: LazyLock<Regex> = LazyLock::new(|| Regex::new(r"`+[^`]+`+").unwrap());

// Regex to match markdown links [text](url) or [text][ref].
// The inline-URL part allows one level of nested parentheses so URLs like
// `https://example.com/docs/v(2)beta` are matched in full; otherwise the URL
// would be truncated at the first ')' and its tail title-cased as prose.
static LINK_REGEX: LazyLock<Regex> =
    LazyLock::new(|| Regex::new(r"\[([^\]]*)\]\((?:[^()]|\([^()]*\))*\)|\[([^\]]*)\]\[[^\]]*\]").unwrap());

// One inline HTML token: a comment, a closing tag (name in group 1), or an open
// tag (name in group 2). `<!-->` and `<!--->` are complete comments, so a later
// `-->` is text. Attributes follow the CommonMark grammar, so a quoted value may
// contain `>`. Elements are paired by name in `html_regions`, since the regex
// engine has no backreferences.
static HTML_TOKEN_REGEX: LazyLock<Regex> = LazyLock::new(|| {
    let pattern = format!(
        r"<!-->|<!--->|<!--.*?-->|</({HTML_TAG_NAME_PATTERN})\s*>|<({HTML_TAG_NAME_PATTERN}){HTML_TAG_ATTRIBUTES_PATTERN}\s*/?>"
    );
    Regex::new(&pattern).unwrap()
});

// Elements that paint content of their own without holding text: the replaced
// elements (images, media, frames, canvases, embedded SVG and MathML) and the
// form controls. An empty one is still visible, so it counts as a word of the
// heading, as a Markdown image does.
const SELF_RENDERING_ELEMENTS: &[&str] = &[
    "img", "video", "audio", "iframe", "embed", "object", "canvas", "svg", "math", "input", "select", "textarea",
    "button", "meter", "progress",
];

// Regex to match custom header IDs {#id}
static CUSTOM_ID_REGEX: LazyLock<Regex> = LazyLock::new(|| Regex::new(r"\s*\{#[^}]+\}\s*$").unwrap());

/// Represents a segment of heading text
#[derive(Debug, Clone)]
enum HeadingSegment {
    /// Regular text that should be capitalized
    Text(String),
    /// Inline code that should be preserved as-is
    Code(String),
    /// Link with text that may be capitalized and URL that's preserved
    Link {
        full: String,
        text_start: usize,
        text_end: usize,
    },
    /// Inline HTML tag that should be preserved as-is
    Html(String),
    /// Image `![alt](url)` preserved as-is, including alt text. Unlike link
    /// text, image alt text is not recased.
    Image(String),
}

impl HeadingSegment {
    /// Whether the segment shows a reader nothing: an empty element such as an
    /// anchor, or a comment. Such a segment does not move the heading's first or
    /// last word. An element that paints content of its own, such as an image or
    /// a form control, is visible even without text.
    fn renders_nothing(&self) -> bool {
        match self {
            HeadingSegment::Html(html) => {
                let paints_something = HTML_TOKEN_REGEX.captures_iter(html).any(|token| {
                    token.get(2).is_some_and(|name| {
                        let name = name.as_str().to_ascii_lowercase();
                        SELF_RENDERING_ELEMENTS.contains(&name.as_str())
                    })
                });
                !paints_something && HTML_TOKEN_REGEX.replace_all(html, "").trim().is_empty()
            }
            _ => false,
        }
    }
}

/// Rule MD063: Heading capitalization
#[derive(Clone)]
pub struct MD063HeadingCapitalization {
    config: MD063Config,
    lowercase_set: HashSet<String>,
    /// Multi-word proper names from MD044 that must survive sentence-case transformation.
    /// Populated via `from_config` when both rules are active.
    proper_names: Vec<String>,
}

impl Default for MD063HeadingCapitalization {
    fn default() -> Self {
        Self::new()
    }
}

impl MD063HeadingCapitalization {
    pub fn new() -> Self {
        let config = MD063Config::default();
        let lowercase_set = config.lowercase_words.iter().cloned().collect();
        Self {
            config,
            lowercase_set,
            proper_names: Vec::new(),
        }
    }

    pub fn from_config_struct(config: MD063Config) -> Self {
        let lowercase_set = config.lowercase_words.iter().cloned().collect();
        Self {
            config,
            lowercase_set,
            proper_names: Vec::new(),
        }
    }

    /// Match `pattern_lower` at `start` in `text` using Unicode-aware lowercasing.
    /// Returns the end byte offset in `text` when the match succeeds.
    ///
    /// This avoids converting the full `text` to lowercase and then reusing those
    /// offsets on the original string, which can panic for case-fold expansions
    /// (e.g. `İ` -> `i̇`).
    fn match_case_insensitive_at(text: &str, start: usize, pattern_lower: &str) -> Option<usize> {
        if start > text.len() || !text.is_char_boundary(start) || pattern_lower.is_empty() {
            return None;
        }

        let mut matched_bytes = 0;

        for (offset, ch) in text[start..].char_indices() {
            if matched_bytes >= pattern_lower.len() {
                break;
            }

            let lowered: String = ch.to_lowercase().collect();
            if !pattern_lower[matched_bytes..].starts_with(&lowered) {
                return None;
            }

            matched_bytes += lowered.len();

            if matched_bytes == pattern_lower.len() {
                return Some(start + offset + ch.len_utf8());
            }
        }

        None
    }

    /// Find the next case-insensitive match of `pattern_lower` in `text`,
    /// returning byte offsets in the ORIGINAL string.
    fn find_case_insensitive_match(text: &str, pattern_lower: &str, search_start: usize) -> Option<(usize, usize)> {
        if pattern_lower.is_empty() || search_start >= text.len() || !text.is_char_boundary(search_start) {
            return None;
        }

        for (offset, _) in text[search_start..].char_indices() {
            let start = search_start + offset;
            if let Some(end) = Self::match_case_insensitive_at(text, start, pattern_lower) {
                return Some((start, end));
            }
        }

        None
    }

    /// Build a map from word byte-position → canonical form for all proper names
    /// that appear in the heading text (case-insensitive phrase match).
    ///
    /// This is used in `apply_sentence_case_from` so that words belonging to a proper
    /// name phrase are never lowercased to begin with.
    fn proper_name_canonical_forms(&self, text: &str) -> std::collections::HashMap<usize, &str> {
        let mut map = std::collections::HashMap::new();

        for name in &self.proper_names {
            if name.is_empty() {
                continue;
            }
            let name_lower = name.to_lowercase();
            let canonical_words: Vec<&str> = name.split_whitespace().collect();
            if canonical_words.is_empty() {
                continue;
            }
            let mut search_start = 0;

            while search_start < text.len() {
                let Some((abs_pos, end_pos)) = Self::find_case_insensitive_match(text, &name_lower, search_start)
                else {
                    break;
                };

                // Require word boundaries
                let before_ok = abs_pos == 0 || !text[..abs_pos].chars().last().is_some_and(char::is_alphanumeric);
                let after_ok =
                    end_pos >= text.len() || !text[end_pos..].chars().next().is_some_and(char::is_alphanumeric);

                if before_ok && after_ok {
                    // Map each word in the matched region to its canonical form.
                    // We zip the words found in the text slice with the words of the
                    // canonical name so that every word gets the right casing.
                    let text_slice = &text[abs_pos..end_pos];
                    let mut word_idx = 0;
                    let mut slice_offset = 0;

                    for text_word in text_slice.split_whitespace() {
                        if let Some(w_rel) = text_slice[slice_offset..].find(text_word) {
                            let word_abs = abs_pos + slice_offset + w_rel;
                            if let Some(&canonical_word) = canonical_words.get(word_idx) {
                                map.insert(word_abs, canonical_word);
                            }
                            slice_offset += w_rel + text_word.len();
                            word_idx += 1;
                        }
                    }
                }

                // Advance by one Unicode scalar value to allow overlapping matches
                // while staying on a UTF-8 char boundary.
                search_start = abs_pos + text[abs_pos..].chars().next().map_or(1, char::len_utf8);
            }
        }

        map
    }

    /// Check if a word has internal capitals (like "iPhone", "macOS", "GitHub", "iOS")
    fn has_internal_capitals(&self, word: &str) -> bool {
        let chars: Vec<char> = word.chars().collect();
        if chars.len() < 2 {
            return false;
        }

        let first = chars[0];
        let rest = &chars[1..];
        let has_upper_in_rest = rest.iter().any(|c| c.is_uppercase());
        let has_lower_in_rest = rest.iter().any(|c| c.is_lowercase());

        // Case 1: Mixed case after first character (like "iPhone", "macOS", "GitHub", "JavaScript")
        if has_upper_in_rest && has_lower_in_rest {
            return true;
        }

        // Case 2: Lowercase first + uppercase in rest (like "iOS", "eBay")
        if first.is_lowercase() && has_upper_in_rest {
            return true;
        }

        false
    }

    /// Check if a word is an all-caps acronym (2+ consecutive uppercase letters)
    /// Examples: "API", "GPU", "HTTP2", "IO" return true
    /// Examples: "A", "iPhone", "npm" return false
    fn is_all_caps_acronym(&self, word: &str) -> bool {
        // Skip single-letter words (handled by title case rules)
        if word.len() < 2 {
            return false;
        }

        let mut consecutive_upper = 0;
        let mut max_consecutive = 0;

        for c in word.chars() {
            if c.is_uppercase() {
                consecutive_upper += 1;
                max_consecutive = max_consecutive.max(consecutive_upper);
            } else if c.is_lowercase() {
                // Any lowercase letter means not all-caps
                return false;
            } else {
                // Non-letter (number, punctuation) - reset counter but don't fail
                consecutive_upper = 0;
            }
        }

        // Must have at least 2 consecutive uppercase letters
        max_consecutive >= 2
    }

    /// Check if a word should be preserved as-is
    fn should_preserve_word(&self, word: &str) -> bool {
        // Check ignore_words list (case-sensitive exact match)
        if self.is_ignored_word(word) {
            return true;
        }

        // Numeric ordinals ("1st", "5th", "21st", ...) must always flow
        // through the normal title-case path so mis-cased forms like
        // "5Th" get normalised back to "5th". Skip the preserve_cased_words
        // heuristics, which would otherwise treat "5Th" as intentionally
        // mixed-case and leave it untouched.
        let is_ordinal = Self::is_numeric_ordinal(word);

        if !is_ordinal {
            // Check if word has internal capitals and preserve_cased_words is enabled
            if self.config.preserve_cased_words && self.has_internal_capitals(word) {
                return true;
            }

            // Check if word is an all-caps acronym (2+ consecutive uppercase)
            if self.config.preserve_cased_words && self.is_all_caps_acronym(word) {
                return true;
            }
        }

        // Preserve caret notation for control characters (^A, ^Z, ^@, etc.)
        if self.is_caret_notation(word) {
            return true;
        }

        false
    }

    /// Whether `word` is an `ignore-words` entry, case-sensitively. A token is
    /// matched as written first (entries like `C++` or `.NET` carry their own
    /// punctuation), then again with trailing punctuation peeled off one character
    /// at a time, so `Kubernetes?`, `Kubernetes),` and `C++,` match the entries
    /// `Kubernetes` and `C++`.
    fn is_ignored_word(&self, word: &str) -> bool {
        let mut candidate = word;
        loop {
            if self.config.ignore_words.iter().any(|ignored| ignored == candidate) {
                return true;
            }
            match candidate.char_indices().next_back() {
                Some((i, c)) if Self::is_trailing_punctuation(c) => candidate = &candidate[..i],
                _ => return false,
            }
        }
    }

    /// Punctuation that can follow a word without becoming part of it: sentence
    /// punctuation and closing brackets or quotes, in ASCII, Latin, CJK and fullwidth
    /// forms. Symbols such as `+` or `#` and combining marks are not in the set, so
    /// `c++` stays distinct from `c` and a decomposed accent stays with its letter.
    fn is_trailing_punctuation(c: char) -> bool {
        matches!(
            c,
            '.' | ','
                | ';'
                | ':'
                | '!'
                | '?'
                | ')'
                | ']'
                | '}'
                | '"'
                | '\''
                | '\u{2026}'
                | '\u{2019}'
                | '\u{201d}'
                | '\u{203a}'
                | '\u{bb}'
                | '\u{3001}'
                | '\u{3002}'
                | '\u{3009}'
                | '\u{300b}'
                | '\u{300d}'
                | '\u{300f}'
                | '\u{3011}'
                | '\u{ff01}'
                | '\u{ff09}'
                | '\u{ff0c}'
                | '\u{ff0e}'
                | '\u{ff1a}'
                | '\u{ff1b}'
                | '\u{ff1f}'
        )
    }

    /// Detect numeric ordinals like `1st`, `2nd`, `3rd`, `4th`, `21st`,
    /// `100th`, ignoring the case of the suffix and any punctuation wrapping
    /// the token (e.g. `5th.`, `1st,`, `(2nd)`, `"3rd"`).
    ///
    /// Such tokens have a fixed lower-case alphabetic suffix in title case
    /// — `21st Century`, never `21St Century` — and must be detected
    /// before applying the generic "capitalise first letter" rule.
    fn is_numeric_ordinal(word: &str) -> bool {
        // The word arrives as a whitespace-split token, punctuation included, so
        // both wrappers come off before the digits are read. Only the wrapping:
        // an interior separator (`2-nd`) leaves a token that is not an ordinal,
        // and a token with no alphanumeric core is rejected by the digit scan.
        let core = word.trim_matches(|c: char| !c.is_alphanumeric());
        let bytes = core.as_bytes();

        // Require at least one leading ASCII digit followed by a letter.
        let alpha_start = match bytes.iter().position(|&b| !b.is_ascii_digit()) {
            Some(pos) if pos > 0 => pos,
            _ => return false,
        };

        // Find where the alphabetic suffix ends (a possessive `'s`, and so on).
        let alpha_end = bytes[alpha_start..]
            .iter()
            .position(|b| !b.is_ascii_alphabetic())
            .map_or(bytes.len(), |p| alpha_start + p);

        let suffix = &core[alpha_start..alpha_end];
        matches!(suffix.to_ascii_lowercase().as_str(), "st" | "nd" | "rd" | "th")
    }

    /// Check if a word is caret notation for control characters (e.g., ^A, ^C, ^Z)
    fn is_caret_notation(&self, word: &str) -> bool {
        let chars: Vec<char> = word.chars().collect();
        // Pattern: ^ followed by uppercase letter or @[\]^_
        if chars.len() >= 2 && chars[0] == '^' {
            let second = chars[1];
            // Control characters: ^@ (NUL) through ^_ (US), which includes ^A-^Z
            if second.is_ascii_uppercase() || "@[\\]^_".contains(second) {
                return true;
            }
        }
        false
    }

    /// Check if a word is a "lowercase word" (articles, prepositions, etc.)
    fn is_lowercase_word(&self, word: &str) -> bool {
        self.lowercase_set.contains(&word.to_lowercase())
    }

    /// Apply title case to a single word
    fn title_case_word(&self, word: &str, is_first: bool, is_last: bool) -> String {
        if word.is_empty() {
            return word.to_string();
        }

        // Preserve words in ignore list or with internal capitals
        if self.should_preserve_word(word) {
            return word.to_string();
        }

        // First and last words are always capitalized
        if is_first || is_last {
            return self.capitalize_first(word);
        }

        // Check if it's a lowercase word (articles, prepositions, etc.)
        if self.is_lowercase_word(word) {
            return Self::lowercase_preserving_composition(word);
        }

        // Regular word - capitalize first letter
        self.capitalize_first(word)
    }

    /// Apply canonical proper-name casing while preserving any trailing punctuation
    /// attached to the original whitespace token (e.g. `javascript,` -> `JavaScript,`).
    fn apply_canonical_form_to_word(word: &str, canonical: &str) -> String {
        let canonical_lower = canonical.to_lowercase();
        if canonical_lower.is_empty() {
            return canonical.to_string();
        }

        if let Some(end_pos) = Self::match_case_insensitive_at(word, 0, &canonical_lower) {
            let mut out = String::with_capacity(canonical.len() + word.len().saturating_sub(end_pos));
            out.push_str(canonical);
            out.push_str(&word[end_pos..]);
            out
        } else {
            canonical.to_string()
        }
    }

    /// Capitalize the first letter of a word, handling Unicode properly
    fn capitalize_first(&self, word: &str) -> String {
        if word.is_empty() {
            return String::new();
        }

        // Find the first alphabetic character to capitalize
        let first_alpha_pos = word.find(|c: char| c.is_alphabetic());
        let Some(pos) = first_alpha_pos else {
            return word.to_string();
        };

        let prefix = &word[..pos];
        let suffix = &word[pos..];

        // Numeric ordinals ("1st", "21st", "5th", ...) keep their
        // alphabetic suffix lower-cased even at title-case positions.
        if Self::is_numeric_ordinal(word) {
            let suffix_lower = Self::lowercase_preserving_composition(suffix);
            return format!("{prefix}{suffix_lower}");
        }

        let mut chars = suffix.chars();
        let first = chars.next().unwrap();
        // Use composition-preserving uppercase to avoid decomposing
        // precomposed characters (e.g., ῷ → Ω + combining marks + Ι)
        let first_upper = Self::uppercase_preserving_composition(&first.to_string());
        let rest: String = chars.collect();
        let rest_lower = Self::lowercase_preserving_composition(&rest);
        format!("{prefix}{first_upper}{rest_lower}")
    }

    /// Lowercase a string character-by-character, preserving precomposed
    /// characters that would decompose during case conversion.
    fn lowercase_preserving_composition(s: &str) -> String {
        let mut result = String::with_capacity(s.len());
        for c in s.chars() {
            let lower: String = c.to_lowercase().collect();
            if lower.chars().count() == 1 {
                result.push_str(&lower);
            } else {
                // Lowercasing would decompose this character; keep original
                result.push(c);
            }
        }
        result
    }

    /// Return the sentence-case spelling of English first-person pronouns found
    /// in `word`, when at least one uses its required uppercase `I`.
    ///
    /// Sentence case normally lowercases every mid-sentence token, but `I` is
    /// uppercase wherever it appears. Recognize its standard contractions as
    /// well, including a typographic apostrophe and adjacent punctuation or
    /// Markdown emphasis. Only already-uppercase instances are recognized: a
    /// lowercase `i` may intentionally be a variable and is outside MD063's
    /// responsibility to reinterpret.
    fn sentence_case_first_person_pronouns(word: &str) -> Option<String> {
        fn is_emphasis_marker(c: char) -> bool {
            matches!(c, '*' | '_' | '~')
        }

        // These characters commonly join identifiers or abbreviations. Treating
        // them as word boundaries would misread technical forms such as `I/O`,
        // `A.I.` or `I-am` as the pronoun.
        fn is_word_connector(c: char) -> bool {
            matches!(c, '/' | '\\' | '-' | '.' | '+' | '&')
        }

        fn is_word_boundary(c: char) -> bool {
            !c.is_alphanumeric() && !is_word_connector(c)
        }

        fn is_pronoun_at(word: &str, pos: usize) -> bool {
            let left = word[..pos].trim_end_matches(is_emphasis_marker);
            if !left.chars().next_back().is_none_or(is_word_boundary) {
                return false;
            }

            let after_i = word[pos + 1..].trim_start_matches(is_emphasis_marker);
            let Some(apostrophe) = after_i.chars().next() else {
                return true;
            };
            if !matches!(apostrophe, '\'' | '’') {
                return is_word_boundary(apostrophe);
            }

            let after_apostrophe = after_i[apostrophe.len_utf8()..].trim_start_matches(is_emphasis_marker);
            let suffix_end = after_apostrophe
                .find(|c: char| !c.is_ascii_alphabetic())
                .unwrap_or(after_apostrophe.len());
            let suffix = &after_apostrophe[..suffix_end];
            if suffix.is_empty() {
                // A trailing apostrophe can be a closing quotation mark around `I`.
                return after_apostrophe.chars().next().is_none_or(is_word_boundary);
            }
            if !matches!(suffix.to_ascii_lowercase().as_str(), "d" | "ll" | "m" | "ve") {
                return false;
            }

            let after_suffix = after_apostrophe[suffix_end..].trim_start_matches(is_emphasis_marker);
            after_suffix.chars().next().is_none_or(is_word_boundary)
        }

        let pronoun_positions: Vec<usize> = word
            .char_indices()
            .filter_map(|(pos, c)| (c == 'I' && is_pronoun_at(word, pos)).then_some(pos))
            .collect();
        if pronoun_positions.is_empty() {
            return None;
        }

        let mut result = String::with_capacity(word.len());
        let mut copied_through = 0;
        for pos in pronoun_positions {
            result.push_str(&Self::lowercase_preserving_composition(&word[copied_through..pos]));
            result.push('I');
            copied_through = pos + 1;
        }
        result.push_str(&Self::lowercase_preserving_composition(&word[copied_through..]));
        Some(result)
    }

    /// Uppercase a string character-by-character, preserving precomposed
    /// characters that would decompose during case conversion.
    /// For example, ῷ (U+1FF7) would decompose into Ω + combining marks + Ι
    /// via to_uppercase(); this function keeps ῷ unchanged instead.
    fn uppercase_preserving_composition(s: &str) -> String {
        let mut result = String::with_capacity(s.len());
        for c in s.chars() {
            let upper: String = c.to_uppercase().collect();
            if upper.chars().count() == 1 {
                result.push_str(&upper);
            } else {
                // Uppercasing would decompose this character; keep original
                result.push(c);
            }
        }
        result
    }

    /// Apply title case to text, using our own title-case logic.
    /// We avoid the external titlecase crate because it decomposes
    /// precomposed Unicode characters during case conversion.
    fn apply_title_case(&self, text: &str) -> String {
        let canonical_forms = self.proper_name_canonical_forms(text);

        let original_words: Vec<&str> = text.split_whitespace().collect();
        let total_words = original_words.len();

        // Pre-compute byte position of each word for canonical form lookup.
        // Use usize::MAX as sentinel for unfound words so canonical_forms.get() returns None.
        let mut word_positions: Vec<usize> = Vec::with_capacity(original_words.len());
        let mut pos = 0;
        for word in &original_words {
            if let Some(rel) = text[pos..].find(word) {
                word_positions.push(pos + rel);
                pos = pos + rel + word.len();
            } else {
                word_positions.push(usize::MAX);
            }
        }

        let result_words: Vec<String> = original_words
            .iter()
            .enumerate()
            .map(|(i, word)| {
                let after_period = i > 0 && original_words[i - 1].ends_with('.');
                let is_first = i == 0 || after_period;
                let is_last = i == total_words - 1;

                // Words that are part of an MD044 proper name use the canonical form directly.
                if let Some(&canonical) = word_positions.get(i).and_then(|&p| canonical_forms.get(&p)) {
                    return Self::apply_canonical_form_to_word(word, canonical);
                }

                // Preserve words in ignore list or with internal capitals
                if self.should_preserve_word(word) {
                    return (*word).to_string();
                }

                // An ignore-words entry is kept whole, hyphens included
                if self.is_ignored_word(word) {
                    return (*word).to_string();
                }

                // Handle hyphenated words
                if word.contains('-') {
                    return self.handle_hyphenated_word(word, is_first, is_last);
                }

                self.title_case_word(word, is_first, is_last)
            })
            .collect();

        result_words.join(" ")
    }

    /// Handle hyphenated words like "self-documenting"
    fn handle_hyphenated_word(&self, word: &str, is_first: bool, is_last: bool) -> String {
        let parts: Vec<&str> = word.split('-').collect();
        let total_parts = parts.len();

        let result_parts: Vec<String> = parts
            .iter()
            .enumerate()
            .map(|(i, part)| {
                // First part of first word and last part of last word get special treatment
                let part_is_first = is_first && i == 0;
                let part_is_last = is_last && i == total_parts - 1;
                self.title_case_word(part, part_is_first, part_is_last)
            })
            .collect();

        result_parts.join("-")
    }

    /// True when a word ends a sentence, so the next word is capitalized.
    ///
    /// The comparison is against the end of the whole word rather than a scan through
    /// it, so `Overview: details` restarts and `a:b` does not. That keeps the boundary
    /// where a reader sees one and leaves `https://example.com` alone.
    fn ends_sentence(&self, word: &str) -> bool {
        self.config
            .sentence_case_restart_after
            .iter()
            .any(|boundary| !boundary.is_empty() && word.ends_with(boundary.as_str()))
    }

    /// Apply sentence case to text, capitalizing the leading word only when it opens
    /// the heading. A segment that follows a code span or link continues the sentence
    /// the earlier segment started, so it begins mid-sentence.
    fn apply_sentence_case_from(&self, text: &str, starts_sentence: bool) -> String {
        if text.is_empty() {
            return text.to_string();
        }

        let canonical_forms = self.proper_name_canonical_forms(text);
        let mut result = String::new();
        let mut current_pos = 0;
        let mut at_sentence_start = starts_sentence;

        // Use original text positions to preserve whitespace correctly
        for word in text.split_whitespace() {
            if let Some(pos) = text[current_pos..].find(word) {
                let abs_pos = current_pos + pos;

                // Preserve whitespace before this word
                result.push_str(&text[current_pos..abs_pos]);

                // Words that are part of an MD044 proper name use the canonical form
                // directly, bypassing sentence-case lowercasing entirely.
                if let Some(&canonical) = canonical_forms.get(&abs_pos) {
                    result.push_str(&Self::apply_canonical_form_to_word(word, canonical));
                } else if self.is_ignored_word(word) {
                    // Explicit ignore-words promise exact byte-for-byte preservation.
                    result.push_str(word);
                } else if let Some(pronoun) = Self::sentence_case_first_person_pronouns(word) {
                    // The English pronoun `I` remains uppercase in every sentence
                    // position; normalize only the contraction suffix.
                    result.push_str(&pronoun);
                } else if at_sentence_start {
                    // Check if word should be preserved BEFORE any capitalization
                    if self.should_preserve_word(word) {
                        // Preserve ignore-words exactly as-is, even at start
                        result.push_str(word);
                    } else {
                        // Sentence-initial word: capitalize first letter, lowercase rest
                        let mut chars = word.chars();
                        if let Some(first) = chars.next() {
                            result.push_str(&Self::uppercase_preserving_composition(&first.to_string()));
                            let rest: String = chars.collect();
                            result.push_str(&Self::lowercase_preserving_composition(&rest));
                        }
                    }
                } else {
                    // Mid-sentence words: preserve if needed, otherwise lowercase
                    if self.should_preserve_word(word) {
                        result.push_str(word);
                    } else {
                        result.push_str(&Self::lowercase_preserving_composition(word));
                    }
                }

                at_sentence_start = self.ends_sentence(word);
                current_pos = abs_pos + word.len();
            }
        }

        // Preserve any trailing whitespace
        if current_pos < text.len() {
            result.push_str(&text[current_pos..]);
        }

        result
    }

    /// Apply all caps to text (preserve whitespace)
    fn apply_all_caps(&self, text: &str) -> String {
        if text.is_empty() {
            return text.to_string();
        }

        let canonical_forms = self.proper_name_canonical_forms(text);
        let mut result = String::new();
        let mut current_pos = 0;

        // Use original text positions to preserve whitespace correctly
        for word in text.split_whitespace() {
            if let Some(pos) = text[current_pos..].find(word) {
                let abs_pos = current_pos + pos;

                // Preserve whitespace before this word
                result.push_str(&text[current_pos..abs_pos]);

                // Words that are part of an MD044 proper name use the canonical form directly.
                // This prevents oscillation with MD044 when all-caps style is active.
                if let Some(&canonical) = canonical_forms.get(&abs_pos) {
                    result.push_str(&Self::apply_canonical_form_to_word(word, canonical));
                } else if self.should_preserve_word(word) {
                    result.push_str(word);
                } else {
                    result.push_str(&Self::uppercase_preserving_composition(word));
                }

                current_pos = abs_pos + word.len();
            }
        }

        // Preserve any trailing whitespace
        if current_pos < text.len() {
            result.push_str(&text[current_pos..]);
        }

        result
    }

    /// Parse heading text into segments
    fn parse_segments(&self, text: &str) -> Vec<HeadingSegment> {
        let mut segments = Vec::new();
        let mut last_end = 0;

        // Collect all special regions (code and links)
        let mut special_regions: Vec<(usize, usize, HeadingSegment)> = Vec::new();

        // Find inline code spans
        for mat in INLINE_CODE_REGEX.find_iter(text) {
            special_regions.push((mat.start(), mat.end(), HeadingSegment::Code(mat.as_str().to_string())));
        }

        // Find links
        for caps in LINK_REGEX.captures_iter(text) {
            let full_match = caps.get(0).unwrap();

            // A '!' immediately before the match makes this an image. Preserve
            // the whole image (including the leading '!' and its alt text)
            // rather than recasing the alt text as if it were link text.
            if full_match.start() >= 1 && text.as_bytes()[full_match.start() - 1] == b'!' {
                let region_start = full_match.start() - 1;
                special_regions.push((
                    region_start,
                    full_match.end(),
                    HeadingSegment::Image(text[region_start..full_match.end()].to_string()),
                ));
                continue;
            }

            let text_match = caps.get(1).or_else(|| caps.get(2));

            if let Some(text_m) = text_match {
                special_regions.push((
                    full_match.start(),
                    full_match.end(),
                    HeadingSegment::Link {
                        full: full_match.as_str().to_string(),
                        text_start: text_m.start() - full_match.start(),
                        text_end: text_m.end() - full_match.start(),
                    },
                ));
            }
        }

        // Find inline HTML: a tag token is never prose, and neither is the content
        // of an element closed on the same line. A tag inside a code span is code.
        let code_ranges: Vec<(usize, usize)> = special_regions
            .iter()
            .filter(|(_, _, segment)| matches!(segment, HeadingSegment::Code(_)))
            .map(|(start, end, _)| (*start, *end))
            .collect();
        for (start, end) in Self::html_regions(text, &code_ranges) {
            special_regions.push((start, end, HeadingSegment::Html(text[start..end].to_string())));
        }

        // Sort by start position
        special_regions.sort_by_key(|(start, _, _)| *start);

        // Drop regions that overlap one already kept. After sorting by start
        // position, the earliest-starting region wins a conflict.
        let mut filtered_regions: Vec<(usize, usize, HeadingSegment)> = Vec::new();
        for region in special_regions {
            let overlaps = filtered_regions.iter().any(|(s, e, _)| region.0 < *e && region.1 > *s);
            if !overlaps {
                filtered_regions.push(region);
            }
        }

        // Build segments
        for (start, end, segment) in filtered_regions {
            // Add text before this special region
            if start > last_end {
                let text_segment = &text[last_end..start];
                if !text_segment.is_empty() {
                    segments.push(HeadingSegment::Text(text_segment.to_string()));
                }
            }
            segments.push(segment);
            last_end = end;
        }

        // Add remaining text
        if last_end < text.len() {
            let remaining = &text[last_end..];
            if !remaining.is_empty() {
                segments.push(HeadingSegment::Text(remaining.to_string()));
            }
        }

        // If no segments were found, treat the whole thing as text
        if segments.is_empty() && !text.is_empty() {
            segments.push(HeadingSegment::Text(text.to_string()));
        }

        segments
    }

    /// Byte ranges of `text` that are HTML and therefore never recased.
    ///
    /// Every tag token and comment is one region. When a closing tag pairs with
    /// the nearest earlier open tag of the same name, the whole element becomes
    /// one region, so `<b>bold <i>inner</i> more</b>` is preserved verbatim
    /// while the prose after an unpaired tag (`<br>`) stays prose. A self-closing
    /// token (`<span/>`) and a void element (`<br>`) hold no content, so neither
    /// opens an element and a later closing tag of the same name belongs to the
    /// enclosing one. A token that starts inside a code span is code, and one
    /// whose `<` is backslash-escaped is text, so neither is markup; the scan
    /// resumes just past its `<`, since that text may hold a real tag of its
    /// own. Tokens are read one after another as a browser tokenizes them, so a
    /// tag written inside another tag's attribute value is part of that value.
    fn html_regions(text: &str, code_ranges: &[(usize, usize)]) -> Vec<(usize, usize)> {
        let mut regions: Vec<(usize, usize)> = Vec::new();
        let mut open_elements: Vec<(String, usize)> = Vec::new();

        let mut pos = 0;
        while let Some(token) = HTML_TOKEN_REGEX.captures_at(text, pos) {
            let whole = token.get(0).unwrap();
            if code_ranges
                .iter()
                .any(|&(start, end)| start <= whole.start() && whole.start() < end)
                || is_backslash_escaped(text, whole.start())
            {
                // The token is text, and text may hold a tag of its own past its `<`.
                pos = whole.start() + 1;
                continue;
            }
            pos = whole.end();

            if let Some(closing) = token.get(1) {
                let name = closing.as_str().to_ascii_lowercase();
                if let Some(depth) = open_elements.iter().rposition(|(open_name, _)| *open_name == name) {
                    let element_start = open_elements[depth].1;
                    open_elements.truncate(depth);
                    regions.retain(|&(start, _)| start < element_start);
                    regions.push((element_start, whole.end()));
                    continue;
                }
            } else if let Some(opening) = token.get(2) {
                let name = opening.as_str().to_ascii_lowercase();
                if !whole.as_str().ends_with("/>") && !is_void_element(&name) {
                    open_elements.push((name, whole.start()));
                }
            }

            regions.push((whole.start(), whole.end()));
        }

        regions
    }

    /// Preserve paired JSX elements, like the existing inline HTML regions.
    /// Only examine tags intersecting this heading, rather than scanning every
    /// document tag again for each heading.
    fn mdx_html_regions(
        ctx: &crate::lint_context::LintContext,
        start: usize,
        end: usize,
    ) -> (Vec<(usize, usize)>, bool) {
        let html = ctx.html_tags();
        let components = ctx.jsx_component_tags();
        let mut tags = Vec::new();
        for cached in [html.as_slice(), components.as_slice()] {
            let first = cached.partition_point(|tag| tag.byte_end <= start);
            tags.extend(cached[first..].iter().take_while(|tag| tag.byte_offset < end));
        }
        tags.sort_unstable_by_key(|tag| tag.byte_offset);
        let mut regions = Vec::new();
        let mut open = Vec::new();
        let mut requires_source_map = false;
        for tag in tags {
            let source = &ctx.content[tag.byte_offset..tag.byte_end];
            requires_source_map |= !HTML_TOKEN_REGEX
                .find(source)
                .is_some_and(|token| token.start() == 0 && token.end() == source.len());
            if tag.is_closing {
                if let Some(depth) = open.iter().rposition(|(name, _)| *name == tag.tag_name.as_str()) {
                    let (_, element_start) = open[depth];
                    open.truncate(depth);
                    regions.push((element_start, tag.byte_end));
                }
            } else if !tag.is_self_closing && !is_void_element(&tag.tag_name) {
                open.push((tag.tag_name.as_str(), tag.byte_offset));
            }
            regions.push((tag.byte_offset, tag.byte_end));
        }
        (regions, requires_source_map)
    }

    /// Preserve cached MDX code when heading text maps directly to its source.
    fn apply_heading_capitalization(
        &self,
        ctx: &crate::lint_context::LintContext,
        line_index: usize,
        heading: &crate::lint_context::HeadingInfo,
    ) -> String {
        if ctx.flavor != crate::config::MarkdownFlavor::MDX {
            return self.apply_capitalization(&heading.raw_text, ctx.flavor);
        }
        let line_info = &ctx.lines[line_index];
        if heading.text_lines > 1 {
            // Joined Setext text has a different byte map. Keep code-bearing
            // headings intact rather than applying source offsets to that map.
            let first = line_index.saturating_sub(heading.text_lines - 1);
            let start = ctx.lines[first].byte_offset;
            let end = line_info.byte_offset + line_info.byte_len;
            let (_, requires_source_map) = Self::mdx_html_regions(ctx, start, end);
            if ctx.overlaps_mdx_inline_code(start, end) || requires_source_map {
                return heading.raw_text.clone();
            }
            return self.apply_capitalization(&heading.raw_text, ctx.flavor);
        }
        let line = line_info.content(ctx.content);
        let Some(relative_start) = line.find(&heading.raw_text) else {
            return heading.raw_text.clone();
        };
        let start = line_info.byte_offset + relative_start;
        let end = start + heading.raw_text.len();
        let (mut ranges, _) = Self::mdx_html_regions(ctx, start, end);
        ranges.extend(ctx.mdx_inline_code_ranges(start, end));
        let ranges = ranges
            .into_iter()
            .map(|(range_start, range_end)| (range_start.max(start) - start, range_end.min(end) - start))
            .collect();
        self.apply_capitalization_with_ranges(&heading.raw_text, ctx.flavor, ranges)
    }

    /// Preserve template identifiers and string values while recasing prose.
    fn apply_capitalization(&self, text: &str, flavor: crate::config::MarkdownFlavor) -> String {
        self.apply_capitalization_with_ranges(text, flavor, Vec::new())
    }

    fn apply_capitalization_with_ranges(
        &self,
        text: &str,
        flavor: crate::config::MarkdownFlavor,
        mut ranges: Vec<(usize, usize)>,
    ) -> String {
        ranges.extend(crate::utils::jinja_utils::find_jinja_ranges(text));
        ranges.extend(crate::utils::shortcode_utils::shortcode_ranges(text));
        ranges.sort_unstable();
        if ranges.is_empty() {
            return self.apply_capitalization_to_segments(text, flavor);
        }

        // Recovery ranges for malformed markup can overlap. Protect their union
        // so every slice below remains ordered and each tag is copied intact.
        let mut merged: Vec<(usize, usize)> = Vec::with_capacity(ranges.len());
        for (start, end) in ranges {
            if let Some(last) = merged.last_mut()
                && start <= last.1
            {
                last.1 = last.1.max(end);
            } else {
                merged.push((start, end));
            }
        }

        // Template tokens contain no letters, so every style preserves them.
        // JSX tokens remain HTML segments so their visible-word context stays
        // the same as the original element. A space inside each template token keeps
        // words on opposite sides of a tag separate during casing. Choose a
        // delimiter absent from the input so literal private-use characters
        // cannot collide with them.
        let mut delimiter = String::from("\u{e000}");
        while text.contains(&delimiter) {
            delimiter.push('\u{e000}');
        }
        let mut masked = String::with_capacity(text.len());
        let mut templates = Vec::with_capacity(merged.len());
        let mut last_end = 0;
        for (index, (start, end)) in merged.into_iter().enumerate() {
            let source = &text[start..end];
            let token = if flavor == crate::config::MarkdownFlavor::MDX && source.starts_with('<') {
                let body = if HeadingSegment::Html(source.to_string()).renders_nothing() {
                    ""
                } else {
                    "0"
                };
                format!("<span data-md063=\"{delimiter}{index}{delimiter}\">{body}</span>")
            } else if flavor == crate::config::MarkdownFlavor::MDX
                && crate::utils::mdx_comments::is_comment_only_expression(&text[start..end])
            {
                // A comment renders nothing, so it must not consume the first
                // visible word when applying sentence case.
                format!("<!--{delimiter}{index}{delimiter}-->")
            } else {
                format!("{delimiter}{index}{delimiter} {delimiter}{index}{delimiter}")
            };
            masked.push_str(&text[last_end..start]);
            masked.push_str(&token);
            templates.push((token, &text[start..end]));
            last_end = end;
        }
        masked.push_str(&text[last_end..]);

        let capitalized = self.apply_capitalization_to_segments(&masked, flavor);
        let mut result = String::with_capacity(capitalized.len());
        let mut cursor = 0;
        // Casing preserves token order. Walk forward so restoration does not
        // repeatedly scan a heading containing many template tags.
        for (token, template) in templates {
            let Some(offset) = capitalized[cursor..].find(&token) else {
                // Preserve the source if a configured transformation consumed a
                // token; an internal placeholder must never become an autofix.
                return text.to_string();
            };
            let start = cursor + offset;
            result.push_str(&capitalized[cursor..start]);
            result.push_str(template);
            cursor = start + token.len();
        }
        result.push_str(&capitalized[cursor..]);
        result
    }

    /// Apply capitalization to heading text after protecting template tags.
    fn apply_capitalization_to_segments(&self, text: &str, flavor: crate::config::MarkdownFlavor) -> String {
        // Strip custom ID if present and re-add later
        let (main_text, custom_id) = if let Some(mat) = CUSTOM_ID_REGEX.find(text) {
            (&text[..mat.start()], Some(mat.as_str()))
        } else {
            (text, None)
        };

        // Markdown with Gherkin spells every structure as a `Keyword: name` heading, and
        // a keyword names a structure only when spelled exactly, so recasing starts after
        // the colon and the keyword is copied through verbatim. The split precedes segment
        // parsing so the keyword keeps its own spacing and never counts as the heading's
        // first or last word, and so a code span the split declines stays visible to the
        // parser below instead of having its contents recased.
        let (keyword, main_text) = if flavor == crate::config::MarkdownFlavor::MDG {
            mdg::keyword_split(main_text).unwrap_or(("", main_text))
        } else {
            ("", main_text)
        };

        // Parse into segments
        let segments = self.parse_segments(main_text);

        // Count text segments to determine first/last word context
        let text_segments: Vec<usize> = segments
            .iter()
            .enumerate()
            .filter_map(|(i, s)| matches!(s, HeadingSegment::Text(_)).then_some(i))
            .collect();

        // Sentence case starts with visible prose. A link label is prose too, even
        // though its Markdown destination is kept opaque. Invisible HTML, such as
        // an empty anchor or a comment, is looked past.
        let first_segment_starts_sentence = segments
            .iter()
            .find(|s| !s.renders_nothing())
            .is_some_and(|s| matches!(s, HeadingSegment::Text(_) | HeadingSegment::Link { .. }));

        // If the last visible segment is Code or Link, then the last text segment should
        // NOT treat its last word as the heading's last word (for lowercase-words respect)
        let last_segment_is_text = segments
            .iter()
            .rev()
            .find(|s| !s.renders_nothing())
            .is_some_and(|s| matches!(s, HeadingSegment::Text(_)));

        // Apply capitalization to each segment
        let mut result_parts: Vec<String> = Vec::new();

        // Where the heading's sentence currently stands, carried across segments so a
        // boundary in one segment governs the next.
        let mut at_sentence_start = first_segment_starts_sentence;

        for (i, segment) in segments.iter().enumerate() {
            // Whether this segment closes a sentence, which decides how the next one
            // starts. Only prose can, and the prose of a heading is what this rule
            // capitalizes: plain text and link text. Code, HTML and images are opaque,
            // so a boundary inside them is not one a reader is offered.
            at_sentence_start = match segment {
                HeadingSegment::Text(t) => {
                    let is_first_text = text_segments.first() == Some(&i);
                    // A text segment is "last" only if it's the last text segment AND
                    // the last segment overall is also text. If there's Code/Link after,
                    // the last word should respect lowercase-words.
                    let is_last_text = text_segments.last() == Some(&i) && last_segment_is_text;

                    let capitalized = match self.config.style {
                        HeadingCapStyle::TitleCase => self.apply_title_case_segment(t, is_first_text, is_last_text),
                        HeadingCapStyle::SentenceCase => self.apply_sentence_case_from(t, at_sentence_start),
                        HeadingCapStyle::AllCaps => self.apply_all_caps(t),
                    };
                    let ends_sentence = self.ends_sentence(capitalized.trim_end());
                    result_parts.push(capitalized);
                    ends_sentence
                }
                HeadingSegment::Code(c) => {
                    result_parts.push(c.clone());
                    false
                }
                HeadingSegment::Link {
                    full,
                    text_start,
                    text_end,
                } => {
                    // Apply capitalization to link text only
                    let link_text = &full[*text_start..*text_end];
                    let capitalized_text = match self.config.style {
                        HeadingCapStyle::TitleCase => self.apply_title_case(link_text),
                        // For sentence case, apply same preservation logic as text
                        // This preserves acronyms (API), brand names (iPhone), etc.
                        HeadingCapStyle::SentenceCase => self.apply_sentence_case_from(link_text, at_sentence_start),
                        HeadingCapStyle::AllCaps => self.apply_all_caps(link_text),
                    };
                    // The link's own text ends the sentence, not its destination: a reader
                    // sees `[see:](url)` as `see:`, so the boundary is where they read it.
                    let ends_sentence = self.ends_sentence(capitalized_text.trim_end());

                    let mut new_link = String::new();
                    new_link.push_str(&full[..*text_start]);
                    new_link.push_str(&capitalized_text);
                    new_link.push_str(&full[*text_end..]);
                    result_parts.push(new_link);
                    ends_sentence
                }
                HeadingSegment::Html(h) => {
                    // Preserve HTML tags as-is (like code). Markup that renders
                    // nothing is invisible to the reader, so the sentence stands
                    // where it stood before it.
                    result_parts.push(h.clone());
                    segment.renders_nothing() && at_sentence_start
                }
                HeadingSegment::Image(img) => {
                    // Preserve images as-is, including alt text.
                    result_parts.push(img.clone());
                    false
                }
            };
        }

        let mut result = String::with_capacity(text.len());
        result.push_str(keyword);
        result.push_str(&result_parts.join(""));

        // Re-add custom ID if present
        if let Some(id) = custom_id {
            result.push_str(id);
        }

        result
    }

    /// Apply title case to a text segment with first/last awareness
    fn apply_title_case_segment(&self, text: &str, is_first_segment: bool, is_last_segment: bool) -> String {
        let canonical_forms = self.proper_name_canonical_forms(text);
        let words: Vec<&str> = text.split_whitespace().collect();
        let total_words = words.len();

        if total_words == 0 {
            return text.to_string();
        }

        // Pre-compute byte position of each word so we can look up canonical forms.
        // Use usize::MAX as sentinel for unfound words so canonical_forms.get() returns None.
        let mut word_positions: Vec<usize> = Vec::with_capacity(words.len());
        let mut pos = 0;
        for word in &words {
            if let Some(rel) = text[pos..].find(word) {
                word_positions.push(pos + rel);
                pos = pos + rel + word.len();
            } else {
                word_positions.push(usize::MAX);
            }
        }

        let result_words: Vec<String> = words
            .iter()
            .enumerate()
            .map(|(i, word)| {
                let after_period = i > 0 && words[i - 1].ends_with('.');
                let is_first = (is_first_segment && i == 0) || after_period;
                let is_last = is_last_segment && i == total_words - 1;

                // Words that are part of an MD044 proper name use the canonical form directly.
                if let Some(&canonical) = word_positions.get(i).and_then(|&p| canonical_forms.get(&p)) {
                    return Self::apply_canonical_form_to_word(word, canonical);
                }

                // An ignore-words entry is kept whole, hyphens included
                if self.is_ignored_word(word) {
                    return (*word).to_string();
                }

                // Handle hyphenated words
                if word.contains('-') {
                    return self.handle_hyphenated_word(word, is_first, is_last);
                }

                self.title_case_word(word, is_first, is_last)
            })
            .collect();

        // Preserve original spacing
        let mut result = String::new();
        let mut word_iter = result_words.iter();
        let mut in_word = false;

        for c in text.chars() {
            if c.is_whitespace() {
                if in_word {
                    in_word = false;
                }
                result.push(c);
            } else if !in_word {
                if let Some(word) = word_iter.next() {
                    result.push_str(word);
                }
                in_word = true;
            }
        }

        result
    }

    /// Fix an ATX heading line
    fn fix_atx_heading(&self, line: &str, heading: &crate::lint_context::HeadingInfo, fixed_text: &str) -> String {
        // Everything before the `#`s stays: the indentation and, for a heading
        // that opens a list item, the list marker
        let indent = &line[..heading.marker_column];
        let hashes = "#".repeat(heading.level as usize);

        // Reconstruct with closing sequence if present
        let closing = &heading.closing_sequence;
        if heading.has_closing_sequence {
            format!("{indent}{hashes} {fixed_text} {closing}")
        } else {
            format!("{indent}{hashes} {fixed_text}")
        }
    }

    /// Fix a Setext heading line
    fn fix_setext_heading(&self, line: &str, heading: &crate::lint_context::HeadingInfo, fixed_text: &str) -> String {
        // Everything before the text stays: the indentation and, for a heading
        // that opens a list item, the list marker
        let leading_ws = &line[..heading.content_column];

        format!("{leading_ws}{fixed_text}")
    }

    /// Rewrite `line` word by word, taking each word's replacement from `words`
    /// and keeping the line's own whitespace runs.
    fn rewrite_words<'w>(line: &str, words: &mut impl Iterator<Item = &'w str>) -> String {
        let mut result = String::with_capacity(line.len());
        let mut in_word = false;
        for c in line.chars() {
            if c.is_whitespace() {
                in_word = false;
                result.push(c);
            } else if !in_word {
                if let Some(word) = words.next() {
                    result.push_str(word);
                }
                in_word = true;
            }
        }
        result
    }

    /// Capitalize a setext heading whose text spans several source lines.
    ///
    /// The heading's text is the whole paragraph its underline ends, so the
    /// capitalization runs over the joined text and the result is written back
    /// word by word onto the original lines, keeping their line breaks and
    /// their indentation. A hard break's backslash goes with the line ending
    /// it marks and is left out of the joined text, so it is put back after
    /// the words; the last line ends with no break, and a backslash inside a
    /// code span is code, so those are in the text and come back with it.
    ///
    /// Returns `None` when the capitalized text does not hold one word per
    /// original word. Prose keeps its whitespace through the capitalization,
    /// but link text is rebuilt from its words alone, so padding inside a link
    /// label changes the count. The heading is then reported without a fix
    /// rather than rewritten from a mapping whose words do not line up.
    fn fix_setext_heading_span(
        &self,
        ctx: &crate::lint_context::LintContext,
        first_idx: usize,
        last_idx: usize,
        heading: &crate::lint_context::HeadingInfo,
        quoted: bool,
        flavor: crate::config::MarkdownFlavor,
    ) -> Option<Vec<String>> {
        let quote_depth = if quoted {
            ctx.lines
                .get(last_idx + 1)
                .and_then(|underline| crate::utils::blockquote::parse_blockquote_prefix(underline.content(ctx.content)))
                .map_or(0, |prefix| prefix.nesting_level)
        } else {
            0
        };
        let bodies: Vec<(&str, &str, &str)> = (first_idx..=last_idx)
            .map(|idx| {
                let line = ctx.lines[idx].content(ctx.content);
                // The first line may open with a list marker, which is not
                // one of the heading's words; a line inside a blockquote
                // carries the quote markers, which are not either
                let (prefix, line) = if idx == first_idx {
                    line.split_at(heading.content_column)
                } else if quoted {
                    // Only the heading's own quote markers are the quote's; a
                    // `>` further in is text of the heading
                    let margin = crate::utils::blockquote::quote_margin_len(line, quote_depth).unwrap_or(0);
                    line.split_at(margin)
                } else {
                    ("", line)
                };
                if idx < last_idx && ctx.line_ends_with_hard_break(idx + 1) {
                    let (body, hard_break) = line.split_at(line.len() - 1);
                    (prefix, body, hard_break)
                } else {
                    (prefix, line, "")
                }
            })
            .collect();
        let fixed_text = self.apply_capitalization(&heading.raw_text, flavor);
        let fixed_words: Vec<&str> = fixed_text.split_whitespace().collect();
        let original_words: usize = bodies.iter().map(|(_, body, _)| body.split_whitespace().count()).sum();
        if fixed_words.len() != original_words {
            return None;
        }

        let mut words = fixed_words.into_iter();
        Some(
            bodies
                .iter()
                .map(|(prefix, body, hard_break)| {
                    format!("{prefix}{}{hard_break}", Self::rewrite_words(body, &mut words))
                })
                .collect(),
        )
    }
}

impl Rule for MD063HeadingCapitalization {
    fn name(&self) -> &'static str {
        "MD063"
    }

    fn description(&self) -> &'static str {
        "Heading capitalization"
    }

    fn category(&self) -> RuleCategory {
        RuleCategory::Heading
    }

    fn should_skip(&self, ctx: &crate::lint_context::LintContext) -> bool {
        !ctx.likely_has_headings() || ctx.headings().next().is_none()
    }

    fn check(&self, ctx: &crate::lint_context::LintContext) -> LintResult {
        let content = ctx.content;

        if content.is_empty() {
            return Ok(Vec::new());
        }

        let mut warnings = Vec::new();

        let edges = crate::utils::blockquote::container_edges(ctx);

        for parsed in ctx.headings() {
            let line_num = parsed.line_num - 1;
            let line_info = parsed.line_info;
            let heading = parsed.heading;
            // Check level filter
            if heading.level < self.config.min_level || heading.level > self.config.max_level {
                continue;
            }

            // Skip headings in code blocks (indented headings)
            if !parsed.is_blockquote()
                && crate::utils::blockquote::is_indented_past_edge(&edges, line_num, line_info.content(ctx.content))
                && matches!(heading.style, crate::lint_context::HeadingStyle::ATX)
            {
                continue;
            }

            // Apply capitalization and compare
            let original_text = &heading.raw_text;
            let fixed_text = self.apply_heading_capitalization(ctx, line_num, heading);

            if original_text != &fixed_text {
                let line = line_info.content(ctx.content);
                let style_name = match self.config.style {
                    HeadingCapStyle::TitleCase => "title case",
                    HeadingCapStyle::SentenceCase => "sentence case",
                    HeadingCapStyle::AllCaps => "ALL CAPS",
                };

                // A setext heading's text is the whole paragraph its
                // underline ends, so the warning covers every one of those
                // lines and the fix rewrites them where they stand.
                if heading.text_lines > 1 {
                    let first_idx = line_num + 1 - heading.text_lines;
                    let first_line = ctx.lines[first_idx].content(ctx.content);
                    let fix = self
                        .fix_setext_heading_span(ctx, first_idx, line_num, heading, parsed.is_blockquote(), ctx.flavor)
                        .map(|rewritten| {
                            let range = ctx.line_content_byte_range(first_idx + 1).start
                                ..ctx.line_content_byte_range(line_num + 1).end;
                            Fix::new(range, rewritten.join("\n"))
                        });
                    warnings.push(LintWarning {
                        rule_name: Some(self.name().to_string()),
                        line: first_idx + 1,
                        column: byte_to_char_count(first_line, heading.content_column),
                        end_line: line_num + 1,
                        end_column: line.trim_end().chars().count() + 1,
                        message: format!("Heading should use {style_name}: '{original_text}' -> '{fixed_text}'"),
                        severity: Severity::Warning,
                        fix,
                    });
                    continue;
                }

                warnings.push(LintWarning {
                    rule_name: Some(self.name().to_string()),
                    line: line_num + 1,
                    column: byte_to_char_count(line, heading.content_column),
                    end_line: line_num + 1,
                    end_column: byte_to_char_count(line, heading.content_column) + original_text.chars().count(),
                    message: format!("Heading should use {style_name}: '{original_text}' -> '{fixed_text}'"),
                    severity: Severity::Warning,
                    fix: Some(Fix::new(
                        ctx.line_content_byte_range(line_num + 1),
                        match heading.style {
                            crate::lint_context::HeadingStyle::ATX => self.fix_atx_heading(line, heading, &fixed_text),
                            _ => self.fix_setext_heading(line, heading, &fixed_text),
                        },
                    )),
                });
            }
        }

        Ok(warnings)
    }

    fn fix(&self, ctx: &crate::lint_context::LintContext) -> Result<String, LintError> {
        let content = ctx.content;

        if content.is_empty() {
            return Ok(content.to_string());
        }

        let lines = ctx.raw_lines();
        let mut fixed_lines: Vec<String> = lines.iter().map(|&s| s.to_string()).collect();

        let edges = crate::utils::blockquote::container_edges(ctx);

        for parsed in ctx.headings() {
            let line_num = parsed.line_num - 1;
            let line_info = parsed.line_info;
            let heading = parsed.heading;
            // Skip lines where the rule is disabled via inline config
            if ctx.is_rule_disabled(self.name(), line_num + 1) {
                continue;
            }

            // Check level filter
            if heading.level < self.config.min_level || heading.level > self.config.max_level {
                continue;
            }

            // Skip headings in code blocks
            if !parsed.is_blockquote()
                && crate::utils::blockquote::is_indented_past_edge(&edges, line_num, line_info.content(ctx.content))
                && matches!(heading.style, crate::lint_context::HeadingStyle::ATX)
            {
                continue;
            }

            let original_text = &heading.raw_text;
            let fixed_text = self.apply_heading_capitalization(ctx, line_num, heading);

            if original_text != &fixed_text {
                let line = line_info.content(ctx.content);
                // A setext heading's text is the whole paragraph its
                // underline ends, so each of those lines keeps its own words.
                if heading.text_lines > 1 {
                    let first_idx = line_num + 1 - heading.text_lines;
                    // The warning is dropped when any of the heading's
                    // lines is disabled, and the rewrite goes with it.
                    if (first_idx..line_num).any(|idx| ctx.is_rule_disabled(self.name(), idx + 1)) {
                        continue;
                    }
                    if let Some(rewritten) = self.fix_setext_heading_span(
                        ctx,
                        first_idx,
                        line_num,
                        heading,
                        parsed.is_blockquote(),
                        ctx.flavor,
                    ) {
                        fixed_lines[first_idx..=line_num].clone_from_slice(&rewritten);
                    }
                    continue;
                }
                fixed_lines[line_num] = match heading.style {
                    crate::lint_context::HeadingStyle::ATX => self.fix_atx_heading(line, heading, &fixed_text),
                    _ => self.fix_setext_heading(line, heading, &fixed_text),
                };
            }
        }

        // Reconstruct content preserving line endings
        let mut result = String::with_capacity(content.len());
        for (i, line) in fixed_lines.iter().enumerate() {
            result.push_str(line);
            if i < fixed_lines.len() - 1 || content.ends_with('\n') {
                result.push('\n');
            }
        }

        Ok(result)
    }

    fn as_any(&self) -> &dyn std::any::Any {
        self
    }

    crate::impl_rule_config_sections!(MD063Config);

    fn from_config(config: &crate::config::Config) -> Box<dyn Rule>
    where
        Self: Sized,
    {
        let rule_config = crate::rule_config_serde::load_rule_config::<MD063Config>(config);
        let md044_config =
            crate::rule_config_serde::load_rule_config::<crate::rules::md044_proper_names::MD044Config>(config);
        let mut rule = Self::from_config_struct(rule_config);
        rule.proper_names = md044_config.names;
        Box::new(rule)
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::lint_context::LintContext;

    fn create_rule() -> MD063HeadingCapitalization {
        let config = MD063Config {
            enabled: true,
            ..Default::default()
        };
        MD063HeadingCapitalization::from_config_struct(config)
    }

    fn create_rule_with_style(style: HeadingCapStyle) -> MD063HeadingCapitalization {
        let config = MD063Config {
            enabled: true,
            style,
            ..Default::default()
        };
        MD063HeadingCapitalization::from_config_struct(config)
    }

    #[test]
    fn test_heading_indent_counts_from_the_list_item_content_column() {
        let rule = create_rule();
        for (source, expected) in [
            ("- x\n\n    ## hello world\n", "- x\n\n    ## Hello World\n"),
            ("- x\n\n     ## hello world\n", "- x\n\n     ## Hello World\n"),
            ("1. x\n\n    ## hello world\n", "1. x\n\n    ## Hello World\n"),
            ("10. x\n\n    ## hello world\n", "10. x\n\n    ## Hello World\n"),
            ("10. x\n\n     ## hello world\n", "10. x\n\n     ## Hello World\n"),
            ("- x\n\n\t## hello world\n", "- x\n\n\t## Hello World\n"),
        ] {
            let ctx = LintContext::new(source, crate::config::MarkdownFlavor::Standard, None);
            assert_eq!(rule.check(&ctx).unwrap().len(), 1, "{source:?}");
            assert_eq!(rule.fix(&ctx).unwrap(), expected, "{source:?}");
        }
    }

    #[test]
    fn test_heading_four_columns_past_the_list_item_content_column_is_code() {
        let rule = create_rule();
        for source in [
            "- x\n\n      ## hello world\n",
            "1. x\n\n       ## hello world\n",
            "10. x\n\n        ## hello world\n",
            "- x\n\n\t\t## hello world\n",
            "    ## hello world\n",
        ] {
            let ctx = LintContext::new(source, crate::config::MarkdownFlavor::Standard, None);
            assert!(rule.check(&ctx).unwrap().is_empty(), "{source:?}");
            assert_eq!(rule.fix(&ctx).unwrap(), source, "{source:?}");
        }
    }

    // Title case tests

    #[test]
    fn test_mdx_sentence_case_distinguishes_comments_from_comments_plus_code() {
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        for (source, expected) in [
            ("# { /* hidden */ } hello world\n", "# { /* hidden */ } Hello world\n"),
            (
                "export const name = \"Visible\"\n\n# {/* c */ name} hello world\n",
                "export const name = \"Visible\"\n\n# {/* c */ name} hello world\n",
            ),
        ] {
            let ctx = LintContext::new(source, crate::config::MarkdownFlavor::MDX, None);
            assert_eq!(rule.fix(&ctx).unwrap(), expected);
        }
    }

    #[test]
    fn test_mdx_leading_comment_does_not_consume_sentence_start() {
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        for (source, expected) in [
            (
                "# {/* literal comment */} hello world\n",
                "# {/* literal comment */} Hello world\n",
            ),
            (
                "{/* literal comment */} hello world\n===\n",
                "{/* literal comment */} Hello world\n===\n",
            ),
        ] {
            let ctx = LintContext::new(source, crate::config::MarkdownFlavor::MDX, None);
            assert_eq!(rule.fix(&ctx).unwrap(), expected);
        }
    }

    #[test]
    fn test_mdx_heading_expression_code_is_preserved() {
        for (style, before, after) in [
            (HeadingCapStyle::TitleCase, "Hello", "World"),
            (HeadingCapStyle::SentenceCase, "Hello", "world"),
            (HeadingCapStyle::AllCaps, "HELLO", "WORLD"),
        ] {
            let rule = create_rule_with_style(style);
            for expression in ["name", "data.text", r#""literal value""#, "name.toLowerCase()"] {
                let source = format!("# hello {{{expression}}} world\n");
                let expected = format!("# {before} {{{expression}}} {after}\n");
                let ctx = LintContext::new(&source, crate::config::MarkdownFlavor::MDX, None);
                assert_eq!(rule.check(&ctx).unwrap().len(), 1);
                assert_eq!(
                    rule.fix(&ctx).unwrap(),
                    expected,
                    "expression code is literal: {source}"
                );
                let fixed_ctx = LintContext::new(&expected, crate::config::MarkdownFlavor::MDX, None);
                assert!(rule.check(&fixed_ctx).unwrap().is_empty());
                assert_eq!(rule.fix(&fixed_ctx).unwrap(), expected);
            }
        }
    }

    #[test]
    fn test_mdx_heading_link_labels_and_adjacent_expressions_preserve_code() {
        let rule = create_rule();
        for (source, expected) in [
            (
                "# hello [{name} world](target) friend\n",
                "# Hello [{name} World](target) Friend\n",
            ),
            ("# hello {name}{data.text} world\n", "# Hello {name}{data.text} World\n"),
            (
                "# hello {/* literal comment */} world\n",
                "# Hello {/* literal comment */} World\n",
            ),
            ("# 日本語 {name} world\n", "# 日本語 {name} World\n"),
        ] {
            let ctx = LintContext::new(source, crate::config::MarkdownFlavor::MDX, None);
            assert_eq!(rule.fix(&ctx).unwrap(), expected);
        }
    }

    #[test]
    fn test_mdx_expression_protection_is_flavor_specific() {
        let rule = create_rule();
        let source = "# hello {name} world\n";
        let ctx = LintContext::new(source, crate::config::MarkdownFlavor::Standard, None);
        assert_eq!(rule.fix(&ctx).unwrap(), "# Hello {Name} World\n");
    }

    #[test]
    fn test_mdx_heading_attribute_expressions_keep_existing_html_protection() {
        let rule = create_rule_with_style(HeadingCapStyle::AllCaps);
        let source = "# hello <span title={name}>world</span> friend\n";
        let expected = "# HELLO <span title={name}>world</span> FRIEND\n";
        let ctx = LintContext::new(source, crate::config::MarkdownFlavor::MDX, None);
        assert_eq!(rule.fix(&ctx).unwrap(), expected);
    }

    #[test]
    fn test_heading_template_tags_are_preserved() {
        for (style, before, after) in [
            (HeadingCapStyle::TitleCase, "Hello", "World"),
            (HeadingCapStyle::SentenceCase, "Hello", "world"),
            (HeadingCapStyle::AllCaps, "HELLO", "WORLD"),
        ] {
            let rule = create_rule_with_style(style);
            for template in [
                "{{ name }}",
                "{{ data['display_name'] }}",
                "{{< note title=\"literal value\" >}}",
                "{{% note title=\"literal value\" %}}",
            ] {
                let source = format!("# hello {template} world\n");
                let expected = format!("# {before} {template} {after}\n");
                let ctx = LintContext::new(&source, crate::config::MarkdownFlavor::Standard, None);
                assert_eq!(rule.check(&ctx).unwrap().len(), 1);
                assert_eq!(rule.fix(&ctx).unwrap(), expected, "template code is literal: {source}");
                let fixed_ctx = LintContext::new(&expected, crate::config::MarkdownFlavor::Standard, None);
                assert!(rule.check(&fixed_ctx).unwrap().is_empty());
                assert_eq!(rule.fix(&fixed_ctx).unwrap(), expected);
            }
        }
    }

    #[test]
    fn test_heading_template_control_tags_and_link_labels_are_preserved() {
        let rule = create_rule();
        for (source, expected) in [
            (
                "# hello {% if enabled %}yes{% else %}no{% endif %} world\n",
                "# Hello {% if enabled %}Yes{% else %}No{% endif %} World\n",
            ),
            (
                "# hello [{{ name }} world](target) friend\n",
                "# Hello [{{ name }} World](target) Friend\n",
            ),
            (
                "# hello {{ name }} and {{ data['display_name'] }} world\n",
                "# Hello {{ name }} and {{ data['display_name'] }} World\n",
            ),
        ] {
            let ctx = LintContext::new(source, crate::config::MarkdownFlavor::Standard, None);
            assert_eq!(rule.fix(&ctx).unwrap(), expected);
        }
    }

    #[test]
    fn test_heading_template_tokens_do_not_collide_with_literal_text() {
        let rule = create_rule_with_style(HeadingCapStyle::AllCaps);
        let source = "# 日本語 \u{e000}0\u{e000} {{ name }} world\n";
        let expected = "# 日本語 \u{e000}0\u{e000} {{ name }} WORLD\n";
        let ctx = LintContext::new(source, crate::config::MarkdownFlavor::Standard, None);
        assert_eq!(rule.fix(&ctx).unwrap(), expected);
    }

    #[test]
    fn test_an_escaped_tag_does_not_hide_the_element_written_inside_it() {
        // `\<span` is text, so the `<a>` where its attribute value would be is a
        // real element and only its bytes are HTML.
        let text = r#"\<span title='<a id="x"></a>'>foo"#;
        assert_eq!(MD063HeadingCapitalization::html_regions(text, &[]), vec![(14, 28)]);
    }

    #[test]
    fn test_title_case_basic() {
        let rule = create_rule();
        let content = "# hello world\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        assert!(result[0].message.contains("Hello World"));
    }

    #[test]
    fn test_title_case_lowercase_words() {
        let rule = create_rule();
        let content = "# the quick brown fox\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        // "The" should be capitalized (first word), "quick", "brown", "fox" should be capitalized
        assert!(result[0].message.contains("The Quick Brown Fox"));
    }

    #[test]
    fn test_title_case_already_correct() {
        let rule = create_rule();
        let content = "# The Quick Brown Fox\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert!(result.is_empty(), "Already correct heading should not be flagged");
    }

    #[test]
    fn test_title_case_hyphenated() {
        let rule = create_rule();
        let content = "# self-documenting code\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        assert!(result[0].message.contains("Self-Documenting Code"));
    }

    #[test]
    fn test_title_case_preserves_url_with_nested_parens() {
        let rule = create_rule();
        // The URL contains a parenthesised segment followed by more URL text.
        let content = "# guide for [the api](https://example.com/docs/v(2)beta)\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let fixed = rule.fix(&ctx).unwrap();
        // The whole URL, including the lowercase "beta" after the nested
        // parens, must be preserved exactly and never title-cased.
        assert!(
            fixed.contains("https://example.com/docs/v(2)beta"),
            "URL with nested parens was corrupted: {fixed:?}"
        );
    }

    #[test]
    fn test_title_case_does_not_recase_image_alt() {
        let rule = create_rule();
        let content = "# overview ![a small icon](icon.png)\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let fixed = rule.fix(&ctx).unwrap();
        // Image (alt text and all) is preserved as-is; only prose is recased.
        assert!(
            fixed.contains("![a small icon](icon.png)"),
            "image alt text was modified: {fixed:?}"
        );
        assert!(
            fixed.contains("# Overview"),
            "surrounding prose should still be title-cased: {fixed:?}"
        );
    }

    // Sentence case tests
    #[test]
    fn test_sentence_case_basic() {
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "# The Quick Brown Fox\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        assert!(result[0].message.contains("The quick brown fox"));
    }

    #[test]
    fn test_sentence_case_already_correct() {
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "# The quick brown fox\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert!(result.is_empty());
    }

    #[test]
    fn test_sentence_case_preserves_first_person_pronoun() {
        // Regression test for issue #845.
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "# How do I debug playbooks?\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);

        assert!(rule.check(&ctx).unwrap().is_empty());
        assert_eq!(rule.fix(&ctx).unwrap(), content);
    }

    #[test]
    fn test_sentence_case_preserves_first_person_contractions() {
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);

        for content in [
            "# What I'd change\n",
            "# Where I'll look\n",
            "# Why I'm here\n",
            "# What I've learned\n",
            "# What I’d change\n",
            "# Where I’ll look\n",
            "# Why I’m here\n",
            "# What I’ve learned\n",
        ] {
            let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
            assert!(rule.check(&ctx).unwrap().is_empty(), "{content:?}");
            assert_eq!(rule.fix(&ctx).unwrap(), content);
        }
    }

    #[test]
    fn test_sentence_case_pronoun_handles_markup_punctuation_and_suffix_case() {
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let cases = [
            ("# What (**I**) would change\n", "# What (**I**) would change\n"),
            ("# Why **I**'M changing it\n", "# Why **I**'m changing it\n"),
            ("# What I’LL change\n", "# What I’ll change\n"),
            ("# What I—really want\n", "# What I—really want\n"),
            ("# What I’d—reluctantly change\n", "# What I’d—reluctantly change\n"),
            ("# What I—yes—I—would do\n", "# What I—yes—I—would do\n"),
            ("# What [I'll change](plan.md)\n", "# What [I'll change](plan.md)\n"),
            ("# [What I—really want](plan.md)\n", "# [What I—really want](plan.md)\n"),
        ];

        for (content, expected) in cases {
            let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
            assert_eq!(rule.fix(&ctx).unwrap(), expected, "{content:?}");
        }
    }

    #[test]
    fn test_sentence_case_pronoun_is_independent_of_cased_word_preservation() {
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::SentenceCase,
            preserve_cased_words: false,
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);
        let content = "# How do I debug what I’LL change?\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);

        assert_eq!(rule.fix(&ctx).unwrap(), "# How do I debug what I’ll change?\n");
    }

    #[test]
    fn test_sentence_case_explicit_ignore_wins_over_pronoun_normalization() {
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::SentenceCase,
            ignore_words: vec!["I'LL".to_string(), "I’LL".to_string()],
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);
        let content = "# Why I'LL stay and why I’LL leave\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);

        assert!(rule.check(&ctx).unwrap().is_empty());
        assert_eq!(rule.fix(&ctx).unwrap(), content);
    }

    #[test]
    fn test_sentence_case_pronoun_does_not_preserve_other_single_letters_or_compounds() {
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "# Compare i, A, I/O, and A.I. values\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);

        assert_eq!(rule.fix(&ctx).unwrap(), "# Compare i, a, i/o, and a.i. values\n");
    }

    // All caps tests
    #[test]
    fn test_all_caps_basic() {
        let rule = create_rule_with_style(HeadingCapStyle::AllCaps);
        let content = "# hello world\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        assert!(result[0].message.contains("HELLO WORLD"));
    }

    // Preserve tests
    #[test]
    fn test_preserve_ignore_words() {
        let config = MD063Config {
            enabled: true,
            ignore_words: vec!["iPhone".to_string(), "macOS".to_string()],
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);

        let content = "# using iPhone on macOS\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        // iPhone and macOS should be preserved
        assert!(result[0].message.contains("iPhone"));
        assert!(result[0].message.contains("macOS"));
    }

    #[test]
    fn test_preserve_cased_words() {
        let rule = create_rule();
        let content = "# using GitHub actions\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        // GitHub should be preserved (has internal capital)
        assert!(result[0].message.contains("GitHub"));
    }

    // Inline code tests
    #[test]
    fn test_inline_code_preserved() {
        let rule = create_rule();
        let content = "# using `const` in javascript\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        // `const` should be preserved, rest capitalized
        assert!(result[0].message.contains("`const`"));
        assert!(result[0].message.contains("Javascript") || result[0].message.contains("JavaScript"));
    }

    // Level filter tests
    #[test]
    fn test_level_filter() {
        let config = MD063Config {
            enabled: true,
            min_level: 2,
            max_level: 4,
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);

        let content = "# h1 heading\n## h2 heading\n### h3 heading\n##### h5 heading\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();

        // Only h2 and h3 should be flagged (h1 < min_level, h5 > max_level)
        assert_eq!(result.len(), 2);
        assert_eq!(result[0].line, 2); // h2
        assert_eq!(result[1].line, 3); // h3
    }

    // Fix tests
    #[test]
    fn test_fix_atx_heading() {
        let rule = create_rule();
        let content = "# hello world\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let fixed = rule.fix(&ctx).unwrap();
        assert_eq!(fixed, "# Hello World\n");
    }

    #[test]
    fn test_fix_multiple_headings() {
        let rule = create_rule();
        let content = "# first heading\n\n## second heading\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let fixed = rule.fix(&ctx).unwrap();
        assert_eq!(fixed, "# First Heading\n\n## Second Heading\n");
    }

    // Setext heading tests
    #[test]
    fn test_setext_heading() {
        let rule = create_rule();
        let content = "hello world\n============\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        assert!(result[0].message.contains("Hello World"));
    }

    #[test]
    fn test_multi_line_setext_heading_is_capitalized_in_place() {
        // A setext heading's text is the whole paragraph its underline ends. The
        // capitalization runs on that joined text so the position rules see the
        // whole heading, and the result goes back onto the author's own lines.
        let rule = create_rule();
        let content = "hello world\nand more words\n==============\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let fixed = rule.fix(&ctx).unwrap();
        assert_eq!(fixed, "Hello World\nand More Words\n==============\n");

        let ctx = LintContext::new(&fixed, crate::config::MarkdownFlavor::Standard, None);
        assert_eq!(rule.fix(&ctx).unwrap(), fixed, "fix is not idempotent");
    }

    #[test]
    fn test_multi_line_setext_heading_keeps_a_hard_break_backslash() {
        // A hard line break's backslash goes with its line ending and is left out
        // of the joined text, so writing the text back puts it back.
        let rule = create_rule();
        let content = "foo bar\\\nbaz qux\n=======\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let fixed = rule.fix(&ctx).unwrap();
        assert_eq!(fixed, "Foo Bar\\\nBaz Qux\n=======\n");

        let ctx = LintContext::new(&fixed, crate::config::MarkdownFlavor::Standard, None);
        assert_eq!(rule.fix(&ctx).unwrap(), fixed, "fix is not idempotent");
    }

    #[test]
    fn test_multi_line_setext_heading_keeps_a_backslash_that_is_no_hard_break() {
        // A backslash inside a code span is code, and one ending the last line
        // is text, so both are in the joined text already and are written back
        // once.
        let rule = create_rule();
        for (content, expected) in [
            ("foo `a\\\nb` tail\n===\n", "Foo `a\\\nb` Tail\n===\n"),
            ("foo bar\nbaz\\\n===\n", "Foo Bar\nBaz\\\n===\n"),
        ] {
            let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
            let fixed = rule.fix(&ctx).unwrap();
            assert_eq!(fixed, expected, "{content:?}");

            let ctx = LintContext::new(&fixed, crate::config::MarkdownFlavor::Standard, None);
            assert_eq!(rule.fix(&ctx).unwrap(), fixed, "fix is not idempotent: {content:?}");
        }
    }

    #[test]
    fn test_multi_line_setext_heading_warning_covers_the_whole_span() {
        // The warning starts on the heading's first text line and runs to the end
        // of the text on its last.
        let rule = create_rule();
        let content = "hello world\nand more\n=====\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1, "got: {result:?}");
        assert_eq!(result[0].line, 1);
        assert_eq!(result[0].column, 1);
        assert_eq!(result[0].end_line, 2);
        assert_eq!(result[0].end_column, 9);
    }

    #[test]
    fn test_multi_line_setext_heading_with_a_multi_byte_word_is_idempotent() {
        // The shrunk input from the idempotency proptest: three text lines under
        // one dash underline, with a multi-byte word in the middle line. The fix
        // may only recase the words where they stand.
        let rule = create_rule();
        let content = "`A`\n| à |  |\n| --- | --- |\n---";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let fixed = rule.fix(&ctx).unwrap();
        assert_eq!(fixed, "`A`\n| À |  |\n| --- | --- |\n---");

        let ctx = LintContext::new(&fixed, crate::config::MarkdownFlavor::Standard, None);
        assert_eq!(rule.fix(&ctx).unwrap(), fixed, "fix is not idempotent");
    }

    #[test]
    fn test_multi_line_setext_heading_whose_words_cannot_be_mapped_back_is_reported_without_a_fix() {
        // Link text is rebuilt from its words, so the padding inside this label
        // is gone from the capitalized text and the words no longer map back
        // one to one onto the author's lines. The heading is still reported;
        // only the rewrite is withheld, and the document is left as written.
        let rule = create_rule();
        let content = "see [ the guide ](guide.md) first\nand then more\n=====\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1, "got: {result:?}");
        assert_eq!(result[0].line, 1);
        assert_eq!(result[0].end_line, 2);
        assert!(
            result[0].fix.is_none(),
            "no rewrite is offered, got: {:?}",
            result[0].fix
        );
        assert_eq!(rule.fix(&ctx).unwrap(), content, "the heading is left as written");

        // Control: the same heading with an unpadded label is rewritten in place.
        let content = "see [the guide](guide.md) first\nand then more\n=====\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1, "got: {result:?}");
        assert!(result[0].fix.is_some(), "the control heading is fixable");
        assert_eq!(
            rule.fix(&ctx).unwrap(),
            "See [The Guide](guide.md) First\nand Then More\n=====\n"
        );
    }

    #[test]
    fn test_fix_honours_a_suppression_on_any_line_of_a_multi_line_setext_heading() {
        // The warning is dropped when any of the heading's lines is disabled,
        // and the rewrite goes with it: the comment disables the first text
        // line, and the heading is recorded on the last.
        let rule = create_rule();
        let content = "<!-- rumdl-disable-next-line MD063 -->\nhello\nworld\n===\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        assert_eq!(
            rule.fix(&ctx).unwrap(),
            content,
            "the suppressed heading is left as written"
        );

        // Control: without the comment the same heading is rewritten in place.
        let content = "hello\nworld\n===\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        assert_eq!(rule.fix(&ctx).unwrap(), "Hello\nWorld\n===\n");
    }

    // Custom ID tests
    #[test]
    fn test_custom_id_preserved() {
        let rule = create_rule();
        let content = "# getting started {#intro}\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        // Custom ID should be preserved
        assert!(result[0].message.contains("{#intro}"));
    }

    // Acronym preservation tests
    #[test]
    fn test_skip_obsidian_tags_not_headings() {
        let rule = create_rule();

        // #tag (no space after #) is an Obsidian tag, not a heading
        let content = "# H1\n\n#tag\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Obsidian, None);
        let result = rule.check(&ctx).unwrap();
        assert!(
            result.is_empty() || result.iter().all(|w| w.line != 3),
            "Obsidian tag #tag should not be treated as a heading: {result:?}"
        );
    }

    #[test]
    fn test_skip_invalid_atx_headings_no_space() {
        let rule = create_rule();

        // #NoSpace is not a valid ATX heading (requires space after #)
        let content = "#notaheading\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert!(
            result.is_empty(),
            "Invalid ATX heading without space should not be flagged: {result:?}"
        );
    }

    #[test]
    fn test_fix_skips_obsidian_tags() {
        let rule = create_rule();

        let content = "# hello world\n\n#tag\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Obsidian, None);
        let fixed = rule.fix(&ctx).unwrap();
        // Should fix the real heading but leave the tag alone
        assert!(fixed.contains("#tag"), "Fix should not modify Obsidian tag #tag");
        assert!(fixed.contains("# Hello World"), "Fix should still fix real headings");
    }

    #[test]
    fn test_preserve_all_caps_acronyms() {
        let rule = create_rule();
        let ctx = |c| LintContext::new(c, crate::config::MarkdownFlavor::Standard, None);

        // Basic acronyms should be preserved
        let fixed = rule.fix(&ctx("# using API in production\n")).unwrap();
        assert_eq!(fixed, "# Using API in Production\n");

        // Multiple acronyms
        let fixed = rule.fix(&ctx("# API and GPU integration\n")).unwrap();
        assert_eq!(fixed, "# API and GPU Integration\n");

        // Two-letter acronyms
        let fixed = rule.fix(&ctx("# IO performance guide\n")).unwrap();
        assert_eq!(fixed, "# IO Performance Guide\n");

        // Acronyms with numbers
        let fixed = rule.fix(&ctx("# HTTP2 and MD5 hashing\n")).unwrap();
        assert_eq!(fixed, "# HTTP2 and MD5 Hashing\n");
    }

    #[test]
    fn test_preserve_acronyms_in_hyphenated_words() {
        let rule = create_rule();
        let ctx = |c| LintContext::new(c, crate::config::MarkdownFlavor::Standard, None);

        // Acronyms at start of hyphenated word
        let fixed = rule.fix(&ctx("# API-driven architecture\n")).unwrap();
        assert_eq!(fixed, "# API-Driven Architecture\n");

        // Multiple acronyms with hyphens
        let fixed = rule.fix(&ctx("# GPU-accelerated CPU-intensive tasks\n")).unwrap();
        assert_eq!(fixed, "# GPU-Accelerated CPU-Intensive Tasks\n");
    }

    #[test]
    fn test_single_letters_not_treated_as_acronyms() {
        let rule = create_rule();
        let ctx = |c| LintContext::new(c, crate::config::MarkdownFlavor::Standard, None);

        // Single uppercase letters should follow title case rules, not be preserved
        let fixed = rule.fix(&ctx("# i am a heading\n")).unwrap();
        assert_eq!(fixed, "# I Am a Heading\n");
    }

    #[test]
    fn test_lowercase_terms_need_ignore_words() {
        let ctx = |c| LintContext::new(c, crate::config::MarkdownFlavor::Standard, None);

        // Without ignore_words: npm gets capitalized
        let rule = create_rule();
        let fixed = rule.fix(&ctx("# using npm packages\n")).unwrap();
        assert_eq!(fixed, "# Using Npm Packages\n");

        // With ignore_words: npm preserved
        let config = MD063Config {
            enabled: true,
            ignore_words: vec!["npm".to_string()],
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);
        let fixed = rule.fix(&ctx("# using npm packages\n")).unwrap();
        assert_eq!(fixed, "# Using npm Packages\n");
    }

    #[test]
    fn test_acronyms_with_mixed_case_preserved() {
        let rule = create_rule();
        let ctx = |c| LintContext::new(c, crate::config::MarkdownFlavor::Standard, None);

        // Both acronyms (API, GPU) and mixed-case (GitHub) should be preserved
        let fixed = rule.fix(&ctx("# using API with GitHub\n")).unwrap();
        assert_eq!(fixed, "# Using API with GitHub\n");
    }

    #[test]
    fn test_real_world_acronyms() {
        let rule = create_rule();
        let ctx = |c| LintContext::new(c, crate::config::MarkdownFlavor::Standard, None);

        // Common technical acronyms from tested repositories
        let content = "# FFI bindings for CPU optimization\n";
        let fixed = rule.fix(&ctx(content)).unwrap();
        assert_eq!(fixed, "# FFI Bindings for CPU Optimization\n");

        let content = "# DOM manipulation and SSR rendering\n";
        let fixed = rule.fix(&ctx(content)).unwrap();
        assert_eq!(fixed, "# DOM Manipulation and SSR Rendering\n");

        let content = "# CVE security and RNN models\n";
        let fixed = rule.fix(&ctx(content)).unwrap();
        assert_eq!(fixed, "# CVE Security and RNN Models\n");
    }

    #[test]
    fn test_is_all_caps_acronym() {
        let rule = create_rule();

        // Should return true for all-caps with 2+ letters
        assert!(rule.is_all_caps_acronym("API"));
        assert!(rule.is_all_caps_acronym("IO"));
        assert!(rule.is_all_caps_acronym("GPU"));
        assert!(rule.is_all_caps_acronym("HTTP2")); // Numbers don't break it

        // Should return false for single letters
        assert!(!rule.is_all_caps_acronym("A"));
        assert!(!rule.is_all_caps_acronym("I"));

        // Should return false for words with lowercase
        assert!(!rule.is_all_caps_acronym("Api"));
        assert!(!rule.is_all_caps_acronym("npm"));
        assert!(!rule.is_all_caps_acronym("iPhone"));
    }

    #[test]
    fn test_sentence_case_starts_after_a_leading_empty_anchor() {
        // The anchor renders nothing, so the sentence still starts at `the`.
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::SentenceCase,
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);

        let content = "# <a id=\"top\"></a>the beginning\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        assert_eq!(rule.check(&ctx).unwrap().len(), 1);
        assert_eq!(rule.fix(&ctx).unwrap(), "# <a id=\"top\"></a>The beginning\n");

        // Visible HTML is an element of its own, so the prose after it is mid-sentence.
        let content = "# <kbd>ctrl</kbd> the key\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        assert!(rule.check(&ctx).unwrap().is_empty());

        // An image paints something without holding text, so it is visible too,
        // whether written as HTML or as Markdown.
        for content in ["# <img src=\"x.png\"> the picture\n", "# ![x](x.png) the picture\n"] {
            let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
            assert!(rule.check(&ctx).unwrap().is_empty(), "{content:?}");
        }
    }

    #[test]
    fn test_sentence_case_ignore_words_first_word() {
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::SentenceCase,
            ignore_words: vec!["nvim".to_string()],
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);

        // "nvim" as first word should be preserved exactly
        let content = "# nvim config\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert!(
            result.is_empty(),
            "nvim in ignore-words should not be flagged. Got: {result:?}"
        );

        // Verify fix also preserves it
        let fixed = rule.fix(&ctx).unwrap();
        assert_eq!(fixed, "# nvim config\n");
    }

    #[test]
    fn test_sentence_case_ignore_words_not_first() {
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::SentenceCase,
            ignore_words: vec!["nvim".to_string()],
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);

        // "nvim" in middle should also be preserved
        let content = "# Using nvim editor\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert!(
            result.is_empty(),
            "nvim in ignore-words should be preserved. Got: {result:?}"
        );
    }

    #[test]
    fn test_preserve_cased_words_ios() {
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::SentenceCase,
            preserve_cased_words: true,
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);

        // "iOS" should be preserved (has mixed case: lowercase 'i' + uppercase 'OS')
        let content = "## This is iOS\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert!(
            result.is_empty(),
            "iOS should be preserved with preserve-cased-words. Got: {result:?}"
        );

        // Verify fix also preserves it
        let fixed = rule.fix(&ctx).unwrap();
        assert_eq!(fixed, "## This is iOS\n");
    }

    #[test]
    fn test_preserve_cased_words_ios_title_case() {
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::TitleCase,
            preserve_cased_words: true,
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);

        // "iOS" should be preserved in title case too
        let content = "# developing for iOS\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let fixed = rule.fix(&ctx).unwrap();
        assert_eq!(fixed, "# Developing for iOS\n");
    }

    #[test]
    fn test_has_internal_capitals_ios() {
        let rule = create_rule();

        // iOS should be detected as having internal capitals
        assert!(
            rule.has_internal_capitals("iOS"),
            "iOS has mixed case (lowercase i, uppercase OS)"
        );

        // Other mixed-case words
        assert!(rule.has_internal_capitals("iPhone"));
        assert!(rule.has_internal_capitals("macOS"));
        assert!(rule.has_internal_capitals("GitHub"));
        assert!(rule.has_internal_capitals("JavaScript"));
        assert!(rule.has_internal_capitals("eBay"));

        // All-caps should NOT be detected (handled by is_all_caps_acronym)
        assert!(!rule.has_internal_capitals("API"));
        assert!(!rule.has_internal_capitals("GPU"));

        // All-lowercase should NOT be detected
        assert!(!rule.has_internal_capitals("npm"));
        assert!(!rule.has_internal_capitals("config"));

        // Regular capitalized words should NOT be detected
        assert!(!rule.has_internal_capitals("The"));
        assert!(!rule.has_internal_capitals("Hello"));
    }

    #[test]
    fn test_lowercase_words_before_trailing_code() {
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::TitleCase,
            lowercase_words: vec![
                "a".to_string(),
                "an".to_string(),
                "and".to_string(),
                "at".to_string(),
                "but".to_string(),
                "by".to_string(),
                "for".to_string(),
                "from".to_string(),
                "into".to_string(),
                "nor".to_string(),
                "on".to_string(),
                "onto".to_string(),
                "or".to_string(),
                "the".to_string(),
                "to".to_string(),
                "upon".to_string(),
                "via".to_string(),
                "vs".to_string(),
                "with".to_string(),
                "without".to_string(),
            ],
            preserve_cased_words: true,
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);

        // Test: "subtitle with a `app`" (all lowercase input)
        // Expected fix: "Subtitle With a `app`" - capitalize "Subtitle" and "With",
        // but keep "a" lowercase (it's in lowercase-words and not the last word)
        // Incorrect: "Subtitle with A `app`" (would incorrectly capitalize "a")
        let content = "## subtitle with a `app`\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();

        // Should flag it
        assert!(!result.is_empty(), "Should flag incorrect capitalization");
        let fixed = rule.fix(&ctx).unwrap();
        // "a" should remain lowercase (not "A") because inline code at end doesn't change lowercase-words behavior
        assert!(
            fixed.contains("with a `app`"),
            "Expected 'with a `app`' but got: {fixed:?}"
        );
        assert!(
            !fixed.contains("with A `app`"),
            "Should not capitalize 'a' to 'A'. Got: {fixed:?}"
        );
        // "Subtitle" should be capitalized, "with" and "a" should remain lowercase (they're in lowercase-words)
        assert!(
            fixed.contains("Subtitle with a `app`"),
            "Expected 'Subtitle with a `app`' but got: {fixed:?}"
        );
    }

    #[test]
    fn test_lowercase_words_preserved_before_trailing_code_variant() {
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::TitleCase,
            lowercase_words: vec!["a".to_string(), "the".to_string(), "with".to_string()],
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);

        // Another variant: "Title with the `code`"
        let content = "## Title with the `code`\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let fixed = rule.fix(&ctx).unwrap();
        // "the" should remain lowercase
        assert!(
            fixed.contains("with the `code`"),
            "Expected 'with the `code`' but got: {fixed:?}"
        );
        assert!(
            !fixed.contains("with The `code`"),
            "Should not capitalize 'the' to 'The'. Got: {fixed:?}"
        );
    }

    #[test]
    fn test_last_word_capitalized_when_no_trailing_code() {
        // Verify that when there's NO trailing code, the last word IS capitalized
        // (even if it's in lowercase-words) - this is the normal title case behavior
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::TitleCase,
            lowercase_words: vec!["a".to_string(), "the".to_string()],
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);

        // "title with a word" - "word" is last, should be capitalized
        // "a" is in lowercase-words and not last, so should be lowercase
        let content = "## title with a word\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let fixed = rule.fix(&ctx).unwrap();
        // "a" should be lowercase, "word" should be capitalized (it's last)
        assert!(
            fixed.contains("With a Word"),
            "Expected 'With a Word' but got: {fixed:?}"
        );
    }

    #[test]
    fn test_multiple_lowercase_words_before_code() {
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::TitleCase,
            lowercase_words: vec![
                "a".to_string(),
                "the".to_string(),
                "with".to_string(),
                "for".to_string(),
            ],
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);

        // Multiple lowercase words before code - all should remain lowercase
        let content = "## Guide for the `user`\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let fixed = rule.fix(&ctx).unwrap();
        assert!(
            fixed.contains("for the `user`"),
            "Expected 'for the `user`' but got: {fixed:?}"
        );
        assert!(
            !fixed.contains("For The `user`"),
            "Should not capitalize lowercase words before code. Got: {fixed:?}"
        );
    }

    #[test]
    fn test_code_in_middle_normal_rules_apply() {
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::TitleCase,
            lowercase_words: vec!["a".to_string(), "the".to_string(), "for".to_string()],
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);

        // Code in the middle - normal title case rules apply (last word capitalized)
        let content = "## Using `const` for the code\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let fixed = rule.fix(&ctx).unwrap();
        // "for" and "the" should be lowercase (middle), "code" should be capitalized (last)
        assert!(
            fixed.contains("for the Code"),
            "Expected 'for the Code' but got: {fixed:?}"
        );
    }

    #[test]
    fn test_link_at_end_same_as_code() {
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::TitleCase,
            lowercase_words: vec!["a".to_string(), "the".to_string(), "for".to_string()],
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);

        // Link at the end - same behavior as code (lowercase words before should remain lowercase)
        let content = "## Guide for the [link](./page.md)\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let fixed = rule.fix(&ctx).unwrap();
        // "for" and "the" should remain lowercase (not last word because link follows)
        assert!(
            fixed.contains("for the [Link]"),
            "Expected 'for the [Link]' but got: {fixed:?}"
        );
        assert!(
            !fixed.contains("for The [Link]"),
            "Should not capitalize 'the' before link. Got: {fixed:?}"
        );
    }

    #[test]
    fn test_multiple_code_segments() {
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::TitleCase,
            lowercase_words: vec!["a".to_string(), "the".to_string(), "with".to_string()],
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);

        // Multiple code segments - last segment is code, so lowercase words before should remain lowercase
        let content = "## Using `const` with a `variable`\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let fixed = rule.fix(&ctx).unwrap();
        // "a" should remain lowercase (not last word because code follows)
        assert!(
            fixed.contains("with a `variable`"),
            "Expected 'with a `variable`' but got: {fixed:?}"
        );
        assert!(
            !fixed.contains("with A `variable`"),
            "Should not capitalize 'a' before trailing code. Got: {fixed:?}"
        );
    }

    #[test]
    fn test_code_and_link_combination() {
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::TitleCase,
            lowercase_words: vec!["a".to_string(), "the".to_string(), "for".to_string()],
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);

        // Code then link - last segment is link, so lowercase words before code should remain lowercase
        let content = "## Guide for the `code` [link](./page.md)\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let fixed = rule.fix(&ctx).unwrap();
        // "for" and "the" should remain lowercase (not last word because link follows)
        assert!(
            fixed.contains("for the `code`"),
            "Expected 'for the `code`' but got: {fixed:?}"
        );
    }

    #[test]
    fn test_text_after_code_capitalizes_last() {
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::TitleCase,
            lowercase_words: vec!["a".to_string(), "the".to_string(), "for".to_string()],
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);

        // Code in middle, text after - last word should be capitalized
        let content = "## Using `const` for the code\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let fixed = rule.fix(&ctx).unwrap();
        // "for" and "the" should be lowercase, "code" is last word, should be capitalized
        assert!(
            fixed.contains("for the Code"),
            "Expected 'for the Code' but got: {fixed:?}"
        );
    }

    #[test]
    fn test_preserve_cased_words_with_trailing_code() {
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::TitleCase,
            lowercase_words: vec!["a".to_string(), "the".to_string(), "for".to_string()],
            preserve_cased_words: true,
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);

        // Preserve-cased words should still work with trailing code
        let content = "## Guide for iOS `app`\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let fixed = rule.fix(&ctx).unwrap();
        // "iOS" should be preserved, "for" should be lowercase
        assert!(
            fixed.contains("for iOS `app`"),
            "Expected 'for iOS `app`' but got: {fixed:?}"
        );
        assert!(
            !fixed.contains("For iOS `app`"),
            "Should not capitalize 'for' before trailing code. Got: {fixed:?}"
        );
    }

    #[test]
    fn test_ignore_words_with_trailing_code() {
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::TitleCase,
            lowercase_words: vec!["a".to_string(), "the".to_string(), "with".to_string()],
            ignore_words: vec!["npm".to_string()],
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);

        // Ignore-words should still work with trailing code
        let content = "## Using npm with a `script`\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let fixed = rule.fix(&ctx).unwrap();
        // "npm" should be preserved, "with" and "a" should be lowercase
        assert!(
            fixed.contains("npm with a `script`"),
            "Expected 'npm with a `script`' but got: {fixed:?}"
        );
        assert!(
            !fixed.contains("with A `script`"),
            "Should not capitalize 'a' before trailing code. Got: {fixed:?}"
        );
    }

    #[test]
    fn test_empty_text_segment_edge_case() {
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::TitleCase,
            lowercase_words: vec!["a".to_string(), "with".to_string()],
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);

        // Edge case: code at start, then text with lowercase word, then code at end
        let content = "## `start` with a `end`\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let fixed = rule.fix(&ctx).unwrap();
        // "with" is first word in text segment, so capitalized (correct)
        // "a" should remain lowercase (not last word because code follows) - this is the key test
        assert!(fixed.contains("a `end`"), "Expected 'a `end`' but got: {fixed:?}");
        assert!(
            !fixed.contains("A `end`"),
            "Should not capitalize 'a' before trailing code. Got: {fixed:?}"
        );
    }

    #[test]
    fn test_sentence_case_with_trailing_code() {
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::SentenceCase,
            lowercase_words: vec!["a".to_string(), "the".to_string()],
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);

        // Sentence case should also respect lowercase words before code
        let content = "## guide for the `user`\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let fixed = rule.fix(&ctx).unwrap();
        // First word capitalized, rest lowercase including "the" before code
        assert!(
            fixed.contains("Guide for the `user`"),
            "Expected 'Guide for the `user`' but got: {fixed:?}"
        );
    }

    #[test]
    fn test_hyphenated_word_before_code() {
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::TitleCase,
            lowercase_words: vec!["a".to_string(), "the".to_string(), "with".to_string()],
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);

        // Hyphenated word before code - last part should respect lowercase-words
        let content = "## Self-contained with a `feature`\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let fixed = rule.fix(&ctx).unwrap();
        // "with" and "a" should remain lowercase (not last word because code follows)
        assert!(
            fixed.contains("with a `feature`"),
            "Expected 'with a `feature`' but got: {fixed:?}"
        );
    }

    // Issue #228: Sentence case with inline code at heading start
    // When a heading starts with inline code, the first word after the code
    // should NOT be capitalized because the heading already has a "first element"

    #[test]
    fn test_sentence_case_code_at_start_basic() {
        // The exact case from issue #228
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "# `rumdl` is a linter\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        // Should be correct as-is: code is first, "is" stays lowercase
        assert!(
            result.is_empty(),
            "Heading with code at start should not flag 'is' for capitalization. Got: {:?}",
            result.iter().map(|w| &w.message).collect::<Vec<_>>()
        );
    }

    #[test]
    fn test_sentence_case_code_at_start_incorrect_capitalization() {
        // Verify we detect incorrect capitalization after code at start
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "# `rumdl` Is a Linter\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        // Should flag: "Is" and "Linter" should be lowercase
        assert_eq!(result.len(), 1, "Should detect incorrect capitalization");
        assert!(
            result[0].message.contains("`rumdl` is a linter"),
            "Should suggest lowercase after code. Got: {:?}",
            result[0].message
        );
    }

    #[test]
    fn test_sentence_case_code_at_start_fix() {
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "# `rumdl` Is A Linter\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let fixed = rule.fix(&ctx).unwrap();
        assert!(
            fixed.contains("# `rumdl` is a linter"),
            "Should fix to lowercase after code. Got: {fixed:?}"
        );
    }

    #[test]
    fn test_sentence_case_text_at_start_still_capitalizes() {
        // Ensure normal headings still capitalize first word
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "# the quick brown fox\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        assert!(
            result[0].message.contains("The quick brown fox"),
            "Text-first heading should capitalize first word. Got: {:?}",
            result[0].message
        );
    }

    #[test]
    fn test_sentence_case_link_at_start() {
        // Link labels are visible prose, so an opening label starts the sentence.
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "# [api](api.md) reference guide\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        assert!(result[0].message.contains("[Api](api.md) reference guide"));
    }

    #[test]
    fn test_sentence_case_link_preserves_acronyms() {
        // Acronyms in link text should be preserved (API, HTTP, etc.)
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "# [API](api.md) Reference Guide\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        // "API" should be preserved (acronym), "Reference Guide" should be lowercased
        assert!(
            result[0].message.contains("[API](api.md) reference guide"),
            "Should preserve acronym 'API' but lowercase following text. Got: {:?}",
            result[0].message
        );
    }

    #[test]
    fn test_sentence_case_link_preserves_brand_names() {
        // Brand names with internal capitals should be preserved
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::SentenceCase,
            preserve_cased_words: true,
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);
        let content = "# [iPhone](iphone.md) Features Guide\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        // "iPhone" should be preserved, "Features Guide" should be lowercased
        assert!(
            result[0].message.contains("[iPhone](iphone.md) features guide"),
            "Should preserve 'iPhone' but lowercase following text. Got: {:?}",
            result[0].message
        );
    }

    #[test]
    fn test_sentence_case_link_lowercases_regular_words() {
        // The first link word is capitalized and later words are lowercased.
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "# [Documentation](docs.md) Reference\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        assert!(
            result[0].message.contains("[Documentation](docs.md) reference"),
            "Should preserve the sentence-initial capital. Got: {:?}",
            result[0].message
        );
    }

    #[test]
    fn test_sentence_case_opening_link_label_is_sentence_initial() {
        // Regression test for issue #844.
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "# [Foo bar](https://example.com)\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);

        assert!(rule.check(&ctx).unwrap().is_empty());
        assert_eq!(rule.fix(&ctx).unwrap(), content);
    }

    #[test]
    fn test_sentence_case_link_at_start_correct_already() {
        // Link with correct casing should not be flagged
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "# [API](api.md) reference guide\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert!(
            result.is_empty(),
            "Correctly cased heading with link should not be flagged. Got: {:?}",
            result.iter().map(|w| &w.message).collect::<Vec<_>>()
        );
    }

    #[test]
    fn test_sentence_case_link_github_preserved() {
        // GitHub should be preserved (internal capitals)
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::SentenceCase,
            preserve_cased_words: true,
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);
        let content = "# [GitHub](gh.md) Repository Setup\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        assert!(
            result[0].message.contains("[GitHub](gh.md) repository setup"),
            "Should preserve 'GitHub'. Got: {:?}",
            result[0].message
        );
    }

    #[test]
    fn test_sentence_case_multiple_code_spans() {
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "# `foo` and `bar` are methods\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        // All text after first code should be lowercase
        assert!(
            result.is_empty(),
            "Should not capitalize words between/after code spans. Got: {:?}",
            result.iter().map(|w| &w.message).collect::<Vec<_>>()
        );
    }

    #[test]
    fn test_sentence_case_code_only_heading() {
        // Heading with only code, no text
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "# `rumdl`\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert!(
            result.is_empty(),
            "Code-only heading should be fine. Got: {:?}",
            result.iter().map(|w| &w.message).collect::<Vec<_>>()
        );
    }

    #[test]
    fn test_sentence_case_code_at_end() {
        // Heading ending with code, text before should still capitalize first word
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "# install the `rumdl` tool\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        // "install" should be capitalized (first word), rest lowercase
        assert_eq!(result.len(), 1);
        assert!(
            result[0].message.contains("Install the `rumdl` tool"),
            "First word should still be capitalized when text comes first. Got: {:?}",
            result[0].message
        );
    }

    #[test]
    fn test_sentence_case_code_in_middle() {
        // Code in middle, text at start should capitalize first word
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "# using the `rumdl` linter for markdown\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        // "using" should be capitalized, rest lowercase
        assert_eq!(result.len(), 1);
        assert!(
            result[0].message.contains("Using the `rumdl` linter for markdown"),
            "First word should be capitalized. Got: {:?}",
            result[0].message
        );
    }

    #[test]
    fn test_sentence_case_preserved_word_after_code() {
        // Preserved words (like iPhone) should stay preserved even after code
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::SentenceCase,
            preserve_cased_words: true,
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);
        let content = "# `swift` iPhone development\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        // "iPhone" should be preserved, "development" lowercase
        assert!(
            result.is_empty(),
            "Preserved words after code should stay. Got: {:?}",
            result.iter().map(|w| &w.message).collect::<Vec<_>>()
        );
    }

    #[test]
    fn test_title_case_code_at_start_still_capitalizes() {
        // Title case should still capitalize words even after code at start
        let rule = create_rule_with_style(HeadingCapStyle::TitleCase);
        let content = "# `api` quick start guide\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        // Title case: all major words capitalized
        assert_eq!(result.len(), 1);
        assert!(
            result[0].message.contains("Quick Start Guide") || result[0].message.contains("quick Start Guide"),
            "Title case should capitalize major words after code. Got: {:?}",
            result[0].message
        );
    }

    // ======== HTML TAG TESTS ========

    #[test]
    fn test_sentence_case_html_tag_at_start() {
        // HTML tag at start: text after should NOT capitalize first word
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "# <kbd>Ctrl</kbd> is a Modifier Key\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        // "is", "a", "Modifier", "Key" should all be lowercase (except preserved words)
        assert_eq!(result.len(), 1);
        let fixed = rule.fix(&ctx).unwrap();
        assert_eq!(
            fixed, "# <kbd>Ctrl</kbd> is a modifier key\n",
            "Text after HTML at start should be lowercase"
        );
    }

    #[test]
    fn test_sentence_case_html_tag_preserves_content() {
        // Content inside HTML tags should be preserved as-is
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "# The <abbr>API</abbr> documentation guide\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        // "The" is first, "API" inside tag preserved, rest lowercase
        assert!(
            result.is_empty(),
            "HTML tag content should be preserved. Got: {:?}",
            result.iter().map(|w| &w.message).collect::<Vec<_>>()
        );
    }

    #[test]
    fn test_sentence_case_html_tag_at_start_with_acronym() {
        // HTML tag at start with acronym content
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "# <abbr>API</abbr> Documentation Guide\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        let fixed = rule.fix(&ctx).unwrap();
        assert_eq!(
            fixed, "# <abbr>API</abbr> documentation guide\n",
            "Text after HTML at start should be lowercase, HTML content preserved"
        );
    }

    #[test]
    fn test_sentence_case_html_tag_in_middle() {
        // HTML tag in middle: first word still capitalized
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "# using the <code>config</code> File\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        let fixed = rule.fix(&ctx).unwrap();
        assert_eq!(
            fixed, "# Using the <code>config</code> file\n",
            "First word capitalized, HTML preserved, rest lowercase"
        );
    }

    #[test]
    fn test_html_tag_strong_emphasis() {
        // <strong> tag handling
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "# The <strong>Bold</strong> Way\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        let fixed = rule.fix(&ctx).unwrap();
        assert_eq!(
            fixed, "# The <strong>Bold</strong> way\n",
            "<strong> tag content should be preserved"
        );
    }

    #[test]
    fn test_html_tag_with_attributes() {
        // HTML tags with attributes should still be detected
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "# <span class=\"highlight\">Important</span> Notice Here\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        let fixed = rule.fix(&ctx).unwrap();
        assert_eq!(
            fixed, "# <span class=\"highlight\">Important</span> notice here\n",
            "HTML tag with attributes should be preserved"
        );
    }

    #[test]
    fn test_multiple_html_tags() {
        // Multiple HTML tags in heading
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "# <kbd>Ctrl</kbd>+<kbd>C</kbd> to Copy Text\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        let fixed = rule.fix(&ctx).unwrap();
        assert_eq!(
            fixed, "# <kbd>Ctrl</kbd>+<kbd>C</kbd> to copy text\n",
            "Multiple HTML tags should all be preserved"
        );
    }

    #[test]
    fn test_html_and_code_mixed() {
        // Mix of HTML tags and inline code
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "# <kbd>Ctrl</kbd>+`v` Paste command\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        let fixed = rule.fix(&ctx).unwrap();
        assert_eq!(
            fixed, "# <kbd>Ctrl</kbd>+`v` paste command\n",
            "HTML and code should both be preserved"
        );
    }

    #[test]
    fn test_self_closing_html_tag() {
        // Self-closing tags like <br/>
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "# Line one<br/>Line Two Here\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        let fixed = rule.fix(&ctx).unwrap();
        assert_eq!(
            fixed, "# Line one<br/>line two here\n",
            "Self-closing HTML tags should be preserved"
        );
    }

    #[test]
    fn test_title_case_with_html_tags() {
        // Title case with HTML tags
        let rule = create_rule_with_style(HeadingCapStyle::TitleCase);
        let content = "# the <kbd>ctrl</kbd> key is a modifier\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        let fixed = rule.fix(&ctx).unwrap();
        // "the" as first word should be "The", content inside <kbd> preserved
        assert!(
            fixed.contains("<kbd>ctrl</kbd>"),
            "HTML tag content should be preserved in title case. Got: {fixed}"
        );
        assert!(
            fixed.starts_with("# The ") || fixed.starts_with("# the "),
            "Title case should work with HTML. Got: {fixed}"
        );
    }

    // ======== CARET NOTATION TESTS ========

    #[test]
    fn test_sentence_case_preserves_caret_notation() {
        // Caret notation for control characters should be preserved
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "## Ctrl+A, Ctrl+R output ^A, ^R on zsh\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        // Should not flag - ^A and ^R are preserved
        assert!(
            result.is_empty(),
            "Caret notation should be preserved. Got: {:?}",
            result.iter().map(|w| &w.message).collect::<Vec<_>>()
        );
    }

    #[test]
    fn test_sentence_case_caret_notation_various() {
        // Various caret notation patterns
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);

        // ^C for interrupt
        let content = "## Press ^C to cancel\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert!(
            result.is_empty(),
            "^C should be preserved. Got: {:?}",
            result.iter().map(|w| &w.message).collect::<Vec<_>>()
        );

        // ^Z for suspend
        let content = "## Use ^Z for background\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert!(
            result.is_empty(),
            "^Z should be preserved. Got: {:?}",
            result.iter().map(|w| &w.message).collect::<Vec<_>>()
        );

        // ^[ for escape
        let content = "## Press ^[ for escape\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert!(
            result.is_empty(),
            "^[ should be preserved. Got: {:?}",
            result.iter().map(|w| &w.message).collect::<Vec<_>>()
        );
    }

    #[test]
    fn test_caret_notation_detection() {
        let rule = create_rule();

        // Valid caret notation
        assert!(rule.is_caret_notation("^A"));
        assert!(rule.is_caret_notation("^Z"));
        assert!(rule.is_caret_notation("^C"));
        assert!(rule.is_caret_notation("^@")); // NUL
        assert!(rule.is_caret_notation("^[")); // ESC
        assert!(rule.is_caret_notation("^]")); // GS
        assert!(rule.is_caret_notation("^^")); // RS
        assert!(rule.is_caret_notation("^_")); // US

        // Not caret notation
        assert!(!rule.is_caret_notation("^a")); // lowercase
        assert!(!rule.is_caret_notation("A")); // no caret
        assert!(!rule.is_caret_notation("^")); // caret alone
        assert!(!rule.is_caret_notation("^1")); // digit
    }

    // MD044 proper names integration tests
    //
    // When MD063 (sentence case) and MD044 (proper names) are both active, MD063 must
    // preserve the exact capitalization of MD044 proper names rather than lowercasing them.
    // Without this, the two rules oscillate: MD044 re-capitalizes what MD063 lowercases.

    fn create_sentence_case_rule_with_proper_names(names: Vec<String>) -> MD063HeadingCapitalization {
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::SentenceCase,
            ..Default::default()
        };
        let mut rule = MD063HeadingCapitalization::from_config_struct(config);
        rule.proper_names = names;
        rule
    }

    #[test]
    fn test_sentence_case_preserves_single_word_proper_name() {
        let rule = create_sentence_case_rule_with_proper_names(vec!["JavaScript".to_string()]);
        // "javascript" in non-first position should become "JavaScript", not "javascript"
        let content = "# installing javascript\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1, "Should flag the heading");
        let fix_text = result[0].fix.as_ref().unwrap().replacement.as_str();
        assert!(
            fix_text.contains("JavaScript"),
            "Fix should preserve proper name 'JavaScript', got: {fix_text:?}"
        );
        assert!(
            !fix_text.contains("javascript"),
            "Fix should not have lowercase 'javascript', got: {fix_text:?}"
        );
    }

    #[test]
    fn test_sentence_case_preserves_multi_word_proper_name() {
        let rule = create_sentence_case_rule_with_proper_names(vec!["Good Application".to_string()]);
        // "Good Application" is a proper name; sentence case must not lowercase "Application"
        let content = "# using good application features\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1, "Should flag the heading");
        let fix_text = result[0].fix.as_ref().unwrap().replacement.as_str();
        assert!(
            fix_text.contains("Good Application"),
            "Fix should preserve 'Good Application' as a phrase, got: {fix_text:?}"
        );
    }

    #[test]
    fn test_sentence_case_proper_name_at_start_of_heading() {
        let rule = create_sentence_case_rule_with_proper_names(vec!["Good Application".to_string()]);
        // The proper name "Good Application" starts the heading; both words must be canonical
        let content = "# good application overview\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1, "Should flag the heading");
        let fix_text = result[0].fix.as_ref().unwrap().replacement.as_str();
        assert!(
            fix_text.contains("Good Application"),
            "Fix should produce 'Good Application' at start of heading, got: {fix_text:?}"
        );
        assert!(
            fix_text.contains("overview"),
            "Non-proper-name word 'overview' should be lowercase, got: {fix_text:?}"
        );
    }

    #[test]
    fn test_sentence_case_with_proper_names_no_oscillation() {
        // This is the core convergence test: applying the fix once must produce
        // output that is already correct (no further changes needed).
        let rule = create_sentence_case_rule_with_proper_names(vec!["Good Application".to_string()]);

        // First application of fix
        let content = "# installing good application on your system\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        let fixed_heading = result[0].fix.as_ref().unwrap().replacement.as_str();

        // The fixed heading should contain the proper name preserved
        assert!(
            fixed_heading.contains("Good Application"),
            "After fix, proper name must be preserved: {fixed_heading:?}"
        );

        // Second application: must produce no further warnings (convergence)
        let fixed_line = format!("{fixed_heading}\n");
        let ctx2 = LintContext::new(&fixed_line, crate::config::MarkdownFlavor::Standard, None);
        let result2 = rule.check(&ctx2).unwrap();
        assert!(
            result2.is_empty(),
            "After one fix, heading must already satisfy both MD063 and MD044 - no oscillation. \
             Second pass warnings: {result2:?}"
        );
    }

    #[test]
    fn test_sentence_case_proper_names_already_correct() {
        let rule = create_sentence_case_rule_with_proper_names(vec!["Good Application".to_string()]);
        // Heading already has correct sentence case with proper name preserved
        let content = "# Installing Good Application\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert!(
            result.is_empty(),
            "Correct sentence-case heading with proper name should not be flagged, got: {result:?}"
        );
    }

    #[test]
    fn test_sentence_case_multiple_proper_names_in_heading() {
        let rule = create_sentence_case_rule_with_proper_names(vec!["TypeScript".to_string(), "React".to_string()]);
        let content = "# using typescript with react\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        let fix_text = result[0].fix.as_ref().unwrap().replacement.as_str();
        assert!(
            fix_text.contains("TypeScript"),
            "Fix should preserve 'TypeScript', got: {fix_text:?}"
        );
        assert!(
            fix_text.contains("React"),
            "Fix should preserve 'React', got: {fix_text:?}"
        );
    }

    #[test]
    fn test_sentence_case_unicode_casefold_expansion_before_proper_name() {
        // Regression for Unicode case-fold expansion: `İ` lowercases to `i̇` (2 code points),
        // so matching offsets must be computed from the original text, not from a lowercased copy.
        let rule = create_sentence_case_rule_with_proper_names(vec!["Österreich".to_string()]);
        let content = "# İ österreich guide\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);

        // Should not panic and should preserve canonical proper-name casing.
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1, "Should flag heading for canonical proper-name casing");
        let fix_text = result[0].fix.as_ref().unwrap().replacement.as_str();
        assert!(
            fix_text.contains("Österreich"),
            "Fix should preserve canonical 'Österreich', got: {fix_text:?}"
        );
    }

    #[test]
    fn test_sentence_case_preserves_trailing_punctuation_on_proper_name() {
        let rule = create_sentence_case_rule_with_proper_names(vec!["JavaScript".to_string()]);
        let content = "# using javascript, today\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1, "Should flag heading");
        let fix_text = result[0].fix.as_ref().unwrap().replacement.as_str();
        assert!(
            fix_text.contains("JavaScript,"),
            "Fix should preserve trailing punctuation, got: {fix_text:?}"
        );
    }

    // Title case + MD044 conflict tests
    //
    // In title case, short words like "the", "a", "of" are kept lowercase by MD063.
    // If those words are part of an MD044 proper name (e.g. "The Rolling Stones"),
    // the same oscillation problem occurs.  The fix must extend to title case too.

    fn create_title_case_rule_with_proper_names(names: Vec<String>) -> MD063HeadingCapitalization {
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::TitleCase,
            ..Default::default()
        };
        let mut rule = MD063HeadingCapitalization::from_config_struct(config);
        rule.proper_names = names;
        rule
    }

    #[test]
    fn test_title_case_preserves_proper_name_with_lowercase_article() {
        // "The" is in the lowercase_words list for title case, so "the" in the middle
        // of a heading would normally stay lowercase.  But "The Rolling Stones" is a
        // proper name that must be capitalised exactly.
        let rule = create_title_case_rule_with_proper_names(vec!["The Rolling Stones".to_string()]);
        let content = "# listening to the rolling stones today\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1, "Should flag the heading");
        let fix_text = result[0].fix.as_ref().unwrap().replacement.as_str();
        assert!(
            fix_text.contains("The Rolling Stones"),
            "Fix should preserve proper name 'The Rolling Stones', got: {fix_text:?}"
        );
    }

    #[test]
    fn test_title_case_proper_name_no_oscillation() {
        // One fix pass must produce output that title case already accepts.
        let rule = create_title_case_rule_with_proper_names(vec!["The Rolling Stones".to_string()]);
        let content = "# listening to the rolling stones today\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        let fixed_heading = result[0].fix.as_ref().unwrap().replacement.as_str();

        let fixed_line = format!("{fixed_heading}\n");
        let ctx2 = LintContext::new(&fixed_line, crate::config::MarkdownFlavor::Standard, None);
        let result2 = rule.check(&ctx2).unwrap();
        assert!(
            result2.is_empty(),
            "After one title-case fix, heading must already satisfy both rules. \
             Second pass warnings: {result2:?}"
        );
    }

    #[test]
    fn test_title_case_unicode_casefold_expansion_before_proper_name() {
        let rule = create_title_case_rule_with_proper_names(vec!["Österreich".to_string()]);
        let content = "# İ österreich guide\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1, "Should flag the heading");
        let fix_text = result[0].fix.as_ref().unwrap().replacement.as_str();
        assert!(
            fix_text.contains("Österreich"),
            "Fix should preserve canonical proper-name casing, got: {fix_text:?}"
        );
    }

    // End-to-end integration test: from_config wires MD044 names into MD063
    //
    // This tests the actual code path used in production, where both rules are
    // configured in a rumdl.toml and the rule registry calls from_config.

    #[test]
    fn test_from_config_loads_md044_names_into_md063() {
        use crate::config::{Config, RuleConfig};
        use crate::rule::Rule;
        use std::collections::BTreeMap;

        let mut config = Config::default();

        // Configure MD063 with sentence_case
        let mut md063_values = BTreeMap::new();
        md063_values.insert("style".to_string(), toml::Value::String("sentence_case".to_string()));
        md063_values.insert("enabled".to_string(), toml::Value::Boolean(true));
        config.rules.insert(
            "MD063".to_string(),
            RuleConfig {
                values: md063_values,
                severity: None,
            },
        );

        // Configure MD044 with a proper name
        let mut md044_values = BTreeMap::new();
        md044_values.insert(
            "names".to_string(),
            toml::Value::Array(vec![toml::Value::String("Good Application".to_string())]),
        );
        config.rules.insert(
            "MD044".to_string(),
            RuleConfig {
                values: md044_values,
                severity: None,
            },
        );

        // Build MD063 via the production code path
        let rule = MD063HeadingCapitalization::from_config(&config);

        // Verify MD044 names were loaded: the fix must preserve "Good Application"
        let content = "# using good application features\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1, "Should flag the heading");
        let fix_text = result[0].fix.as_ref().unwrap().replacement.as_str();
        assert!(
            fix_text.contains("Good Application"),
            "from_config should wire MD044 names into MD063; fix should preserve \
             'Good Application', got: {fix_text:?}"
        );
    }

    #[test]
    fn test_title_case_short_word_not_confused_with_substring() {
        // Verify that short preposition matching ("in") does not trigger on
        // substrings of longer words ("insert"). Title case must capitalize
        // "insert" while keeping "in" lowercase.
        let rule = create_rule_with_style(HeadingCapStyle::TitleCase);

        // "in" is a short preposition (should be lowercase in title case)
        // "insert" contains "in" as substring but is a regular word (should be capitalized)
        let content = "# in the insert\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1, "Should flag the heading");
        let fix = result[0].fix.as_ref().expect("Fix should be present");
        // "In" capitalized as first word, "the" lowercase as article, "Insert" capitalized
        assert!(
            fix.replacement.contains("In the Insert"),
            "Expected 'In the Insert', got: {:?}",
            fix.replacement
        );
    }

    #[test]
    fn test_title_case_or_not_confused_with_orchestra() {
        let rule = create_rule_with_style(HeadingCapStyle::TitleCase);

        // "or" is a conjunction (should be lowercase in title case)
        // "orchestra" contains "or" as substring but is a regular word
        let content = "# or the orchestra\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1, "Should flag the heading");
        let fix = result[0].fix.as_ref().expect("Fix should be present");
        // "Or" capitalized as first word, "the" lowercase, "Orchestra" capitalized
        assert!(
            fix.replacement.contains("Or the Orchestra"),
            "Expected 'Or the Orchestra', got: {:?}",
            fix.replacement
        );
    }

    #[test]
    fn test_all_caps_preserves_all_words() {
        let rule = create_rule_with_style(HeadingCapStyle::AllCaps);

        let content = "# in the insert\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1, "Should flag the heading");
        let fix = result[0].fix.as_ref().expect("Fix should be present");
        assert!(
            fix.replacement.contains("IN THE INSERT"),
            "All caps should uppercase all words, got: {:?}",
            fix.replacement
        );
    }

    // Numbered prefix tests — words following a period-terminated token must be capitalized
    #[test]
    fn test_title_case_numbered_prefix_lowercase_word() {
        // "to" follows "1." and must be treated as the start of a new phrase
        let rule = create_rule();
        let content = "## 1. To Be a Thing\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert!(
            result.is_empty(),
            "Should not flag '## 1. To Be a Thing', got: {result:?}"
        );

        let content_lower = "## 1. to be a thing\n";
        let ctx2 = LintContext::new(content_lower, crate::config::MarkdownFlavor::Standard, None);
        let result2 = rule.check(&ctx2).unwrap();
        assert!(!result2.is_empty(), "Should flag '## 1. to be a thing'");
        let fix = result2[0].fix.as_ref().expect("Should have a fix");
        assert!(
            fix.replacement.contains("1. To Be a Thing"),
            "Fix should capitalize 'To', got: {:?}",
            fix.replacement
        );
    }

    #[test]
    fn test_title_case_numbered_prefix_article() {
        // "a" follows "2." and must be capitalized as the first word of the phrase
        let rule = create_rule();
        let content = "## 2. A Guide to the Galaxy\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert!(
            result.is_empty(),
            "Should not flag '## 2. A Guide to the Galaxy', got: {result:?}"
        );

        let content_lower = "## 2. a guide to the galaxy\n";
        let ctx2 = LintContext::new(content_lower, crate::config::MarkdownFlavor::Standard, None);
        let result2 = rule.check(&ctx2).unwrap();
        assert!(!result2.is_empty(), "Should flag '## 2. a guide to the galaxy'");
        let fix = result2[0].fix.as_ref().expect("Should have a fix");
        assert!(
            fix.replacement.contains("2. A Guide to the Galaxy"),
            "Fix should capitalize 'A', got: {:?}",
            fix.replacement
        );
    }

    #[test]
    fn test_title_case_mid_sentence_period_word() {
        // "introduction" follows "1." embedded in a phrase — must be capitalized
        let rule = create_rule();
        let content = "## Step 1. Introduction to the Problem\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert!(
            result.is_empty(),
            "Should not flag '## Step 1. Introduction to the Problem', got: {result:?}"
        );

        let content_lower = "## Step 1. introduction to the problem\n";
        let ctx2 = LintContext::new(content_lower, crate::config::MarkdownFlavor::Standard, None);
        let result2 = rule.check(&ctx2).unwrap();
        assert!(
            !result2.is_empty(),
            "Should flag '## Step 1. introduction to the problem'"
        );
        let fix = result2[0].fix.as_ref().expect("Should have a fix");
        assert!(
            fix.replacement.contains("Step 1. Introduction to the Problem"),
            "Fix should capitalize 'Introduction', got: {:?}",
            fix.replacement
        );
    }

    #[test]
    fn test_title_case_numbered_prefix_in_link_text() {
        // apply_title_case (link text path) must also respect after_period.
        // A heading whose only content is a link: ## [1. to be a thing](url)
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::TitleCase,
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);

        // Correct heading — link text already title-cased after numbered prefix
        let content = "## [1. To Be a Thing](https://example.com)\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert!(
            result.is_empty(),
            "Should not flag '## [1. To Be a Thing](url)', got: {result:?}"
        );

        // Incorrect heading — "to" in link text must be capitalized after "1."
        let content_lower = "## [1. to be a thing](https://example.com)\n";
        let ctx2 = LintContext::new(content_lower, crate::config::MarkdownFlavor::Standard, None);
        let result2 = rule.check(&ctx2).unwrap();
        assert!(!result2.is_empty(), "Should flag '## [1. to be a thing](url)'");
        let fix = result2[0].fix.as_ref().expect("Should have a fix");
        assert!(
            fix.replacement.contains("1. To Be a Thing"),
            "Fix should capitalize 'To' in link text, got: {:?}",
            fix.replacement
        );
    }

    // Numeric-ordinal tests (issue #608): "1st", "2nd", "3rd", "4th", "21st"
    // and so on must keep their alphabetic suffix lower-cased in title case
    // and must be normalised back from mis-cased forms like "5Th".

    #[test]
    fn test_is_numeric_ordinal_recognises_canonical_forms() {
        for word in &[
            "1st", "2nd", "3rd", "4th", "5th", "11th", "21st", "22nd", "23rd", "100th", "1ST", "5Th", "21St", "21sT",
        ] {
            assert!(
                MD063HeadingCapitalization::is_numeric_ordinal(word),
                "expected `{word}` to be detected as a numeric ordinal"
            );
        }
    }

    #[test]
    fn test_is_numeric_ordinal_rejects_non_ordinals() {
        // Words without a digit prefix, an unrecognised alphabetic suffix,
        // or a non-ordinal alpha tail are all rejected. Compound forms with
        // hyphens are handled by `handle_hyphenated_word` so the helper's
        // behaviour on them is intentionally unconstrained.
        for word in &[
            "first", "1stop", "ist", "5", "th", "abc", "4G", "4K", "30s", "100k", "5x", "1.5", "iPhone6S",
        ] {
            assert!(
                !MD063HeadingCapitalization::is_numeric_ordinal(word),
                "expected `{word}` NOT to be detected as a numeric ordinal"
            );
        }
    }

    #[test]
    fn test_is_numeric_ordinal_strips_trailing_punctuation() {
        for word in &["5th.", "1st,", "21st!", "3rd:", "4th)", "5th's"] {
            assert!(
                MD063HeadingCapitalization::is_numeric_ordinal(word),
                "expected `{word}` to be detected as a numeric ordinal (with punctuation)"
            );
        }
    }

    #[test]
    fn test_is_numeric_ordinal_ignores_wrapping_punctuation() {
        // A word reaches this check as a whitespace-split token, so it still
        // carries whatever punctuation wraps it. A leading wrapper hid the
        // digits, and the capitaliser then uppercased the first letter it could
        // find, turning `(2nd` into `(2Nd`.
        for word in &[
            "(2nd", "[2nd", "\"2nd", "'2nd", "*2nd", "(21st)", "\"3rd\"", "**5th**", "_1st_",
        ] {
            assert!(
                MD063HeadingCapitalization::is_numeric_ordinal(word),
                "expected `{word}` to be detected as a numeric ordinal"
            );
        }

        // Only the wrapping comes off. An interior separator still makes the
        // token something other than an ordinal, and a token with no core at
        // all must be rejected rather than read as an empty ordinal.
        for word in &["2-nd", "2 nd", "(", "\"\"", "()", "(nd", "(2"] {
            assert!(
                !MD063HeadingCapitalization::is_numeric_ordinal(word),
                "expected `{word}` NOT to be detected as a numeric ordinal"
            );
        }
    }

    #[test]
    fn test_ordinal_wrapped_in_punctuation_survives_a_fix() {
        // Already correct in each style: the wrapped ordinal is preserved, so
        // nothing is reported and the fix leaves the heading byte-identical.
        for (style, content) in [
            (HeadingCapStyle::SentenceCase, "# The second (2nd) attempt\n"),
            (HeadingCapStyle::SentenceCase, "# Ranked \"3rd\" overall\n"),
            (HeadingCapStyle::SentenceCase, "# Plain 2nd place\n"),
            (HeadingCapStyle::TitleCase, "# The Second (2nd) Attempt\n"),
            (HeadingCapStyle::TitleCase, "# Ranked \"3rd\" Overall\n"),
        ] {
            let rule = create_rule_with_style(style);
            let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
            let result = rule.check(&ctx).unwrap();
            assert!(
                result.is_empty(),
                "{style:?} should not flag {content:?}, got: {result:?}"
            );
            assert_eq!(
                rule.fix(&ctx).unwrap(),
                content,
                "{style:?} must leave {content:?} alone"
            );
        }

        // Positive control: on a heading that does need recasing the fix runs,
        // recases the prose around the ordinal and still leaves the ordinal
        // itself untouched. A second pass changes nothing more.
        for (style, content, expected) in [
            (
                HeadingCapStyle::SentenceCase,
                "# The Second (2nd) Attempt\n",
                "# The second (2nd) attempt\n",
            ),
            (
                HeadingCapStyle::TitleCase,
                "# the second (2nd) attempt\n",
                "# The Second (2nd) Attempt\n",
            ),
        ] {
            let rule = create_rule_with_style(style);
            let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
            assert!(
                !rule.check(&ctx).unwrap().is_empty(),
                "{style:?} should flag {content:?}"
            );
            let fixed = rule.fix(&ctx).unwrap();
            assert_eq!(fixed, expected, "{style:?} fix of {content:?}");

            let refixed = rule
                .fix(&LintContext::new(&fixed, crate::config::MarkdownFlavor::Standard, None))
                .unwrap();
            assert_eq!(refixed, expected, "{style:?} second pass over {fixed:?}");
        }
    }

    #[test]
    fn test_wrapped_ordinal_corrupted_by_the_old_fix_is_repaired() {
        // A document already rewritten by the buggy capitaliser has to come back,
        // not stay wrong: `(2Nd)` was not recognised as an ordinal either, so the
        // old behaviour was stable and `check` reported nothing about it.
        for (style, content, expected) in [
            (
                HeadingCapStyle::SentenceCase,
                "# The second (2Nd) attempt\n",
                "# The second (2nd) attempt\n",
            ),
            (
                HeadingCapStyle::TitleCase,
                "# The Second (2Nd) Attempt\n",
                "# The Second (2nd) Attempt\n",
            ),
            (
                HeadingCapStyle::SentenceCase,
                "# Ranked \"3Rd\" overall\n",
                "# Ranked \"3rd\" overall\n",
            ),
        ] {
            let rule = create_rule_with_style(style);
            let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
            assert!(
                !rule.check(&ctx).unwrap().is_empty(),
                "{style:?} should flag {content:?}"
            );
            assert_eq!(rule.fix(&ctx).unwrap(), expected, "{style:?} fix of {content:?}");
        }
    }

    #[test]
    fn test_title_case_ordinal_first_word_not_flagged() {
        let rule = create_rule();
        for content in &[
            "# 1st Place\n",
            "# 2nd Edition\n",
            "# 3rd Time\n",
            "# 5th Avenue\n",
            "# 21st Century Skills\n",
            "# 100th Customer\n",
        ] {
            let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
            let result = rule.check(&ctx).unwrap();
            assert!(result.is_empty(), "Should not flag {content:?}, got: {result:?}");
        }
    }

    #[test]
    fn test_title_case_ordinal_mid_heading_not_flagged() {
        let rule = create_rule();
        for content in &[
            "# May 3rd Notes\n",
            "# Top 100th Customer\n",
            "# Notes for the 5th of May\n",
        ] {
            let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
            let result = rule.check(&ctx).unwrap();
            assert!(result.is_empty(), "Should not flag {content:?}, got: {result:?}");
        }
    }

    #[test]
    fn test_title_case_ordinal_corrupted_form_is_fixed() {
        // The "sticky" case: a heading already mangled by the buggy
        // capitaliser must be flagged and corrected back, not left alone.
        let rule = create_rule();
        for (input, expected) in &[
            ("# 1St Place\n", "1st Place"),
            ("# 5Th Avenue\n", "5th Avenue"),
            ("# 21St Century Skills\n", "21st Century Skills"),
            ("# May 3Rd Notes\n", "May 3rd Notes"),
            ("# 22Nd Edition\n", "22nd Edition"),
        ] {
            let ctx = LintContext::new(input, crate::config::MarkdownFlavor::Standard, None);
            let result = rule.check(&ctx).unwrap();
            assert!(!result.is_empty(), "Should flag {input:?}");
            let fix = result[0].fix.as_ref().expect("should have a fix");
            assert!(
                fix.replacement.contains(expected),
                "Fix for {input:?} should contain {expected:?}, got: {:?}",
                fix.replacement
            );
        }
    }

    #[test]
    fn test_title_case_ordinal_lowercase_other_words_capitalised() {
        // Non-ordinal words around an ordinal still need title-casing.
        let rule = create_rule();
        let content = "# 5th avenue\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        let fix = result[0].fix.as_ref().expect("should have a fix");
        assert!(
            fix.replacement.contains("5th Avenue"),
            "Fix should produce '5th Avenue', got: {:?}",
            fix.replacement
        );
    }

    #[test]
    fn test_title_case_ordinal_with_trailing_punctuation() {
        let rule = create_rule();
        let content = "# Released on the 5th.\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert!(result.is_empty(), "Should not flag {content:?}, got: {result:?}");
    }

    #[test]
    fn test_title_case_ordinal_hyphenated() {
        let rule = create_rule();
        for content in &["# 21st-Century Skills\n", "# A 19th-Century Novel\n"] {
            let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
            let result = rule.check(&ctx).unwrap();
            assert!(result.is_empty(), "Should not flag {content:?}, got: {result:?}");
        }
    }

    #[test]
    fn test_sentence_case_ordinal_corrupted_form_is_fixed() {
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        let content = "# 5Th avenue\n";
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        let fix = result[0].fix.as_ref().expect("should have a fix");
        assert!(
            fix.replacement.contains("5th avenue"),
            "Fix should produce '5th avenue', got: {:?}",
            fix.replacement
        );
    }

    #[test]
    fn test_title_case_digit_acronym_unchanged() {
        // Non-ordinal digit-prefixed tokens (4G, 4K) must still be preserved
        // as all-caps acronyms — the ordinal carve-out must not catch them.
        let rule = create_rule();
        for content in &["# 4G Networks\n", "# 4K Streaming\n"] {
            let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
            let result = rule.check(&ctx).unwrap();
            assert!(result.is_empty(), "Should not flag {content:?}, got: {result:?}");
        }
    }

    // --- sentence-case-restart-after ---

    fn restart_rule(boundaries: &[&str]) -> MD063HeadingCapitalization {
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::SentenceCase,
            sentence_case_restart_after: boundaries.iter().copied().map(String::from).collect(),
            ..Default::default()
        };
        MD063HeadingCapitalization::from_config_struct(config)
    }

    /// The heading text MD063 would rewrite this content to, or `None` when it is
    /// already compliant.
    fn suggested(rule: &MD063HeadingCapitalization, content: &str) -> Option<String> {
        let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
        let warnings = rule.check(&ctx).unwrap();
        let fixed = rule.fix(&ctx).unwrap();
        assert_eq!(
            warnings.is_empty(),
            fixed == content,
            "a warning and a rewrite must agree for {content:?}"
        );
        (!warnings.is_empty()).then(|| fixed.trim_start_matches('#').trim().to_string())
    }

    #[test]
    fn test_restart_after_capitalizes_the_word_following_a_boundary() {
        let rule = restart_rule(&[":"]);
        assert_eq!(
            suggested(&rule, "# Requirement 1: Struct to Logger Slice Conversion\n").as_deref(),
            Some("Requirement 1: Struct to logger slice conversion")
        );
    }

    #[test]
    fn test_restart_after_defaults_to_no_boundaries() {
        // The empty default must leave sentence case as it was: first word only.
        let rule = create_rule_with_style(HeadingCapStyle::SentenceCase);
        assert_eq!(
            suggested(&rule, "# Requirement 1: Struct to Logger Slice Conversion\n").as_deref(),
            Some("Requirement 1: struct to logger slice conversion")
        );
    }

    #[test]
    fn test_restart_after_only_honors_configured_punctuation() {
        // A colon-only configuration must not drag in the dash and semicolon cases.
        let rule = restart_rule(&[":"]);
        assert_eq!(
            suggested(&rule, "# Design - Data Model Overview\n").as_deref(),
            Some("Design - data model overview")
        );
        assert_eq!(
            suggested(&rule, "# Setup; Then Run\n").as_deref(),
            Some("Setup; then run")
        );

        let rule = restart_rule(&[";", "\u{2014}"]);
        assert_eq!(
            suggested(&rule, "# Setup; Then Run\n").as_deref(),
            Some("Setup; Then run")
        );
        assert_eq!(
            suggested(&rule, "# Part One \u{2014} The Big Idea\n").as_deref(),
            Some("Part one \u{2014} The big idea")
        );
    }

    #[test]
    fn test_restart_after_matches_only_at_the_end_of_a_word() {
        // An intra-word hyphen is not a sentence boundary, so a configured dash must
        // not restart inside `Well-Known`, and a URL's punctuation must not either.
        let rule = restart_rule(&["-", ":"]);
        assert_eq!(
            suggested(&rule, "# Ports: Well-Known Ports Explained\n").as_deref(),
            Some("Ports: Well-Known ports explained")
        );
        assert_eq!(
            suggested(&rule, "# See https://example.com/A/B For Details\n").as_deref(),
            Some("See https://example.com/A/B for details")
        );
    }

    #[test]
    fn test_restart_after_a_trailing_boundary_is_a_no_op() {
        let rule = restart_rule(&[":"]);
        assert_eq!(suggested(&rule, "# Setup:\n"), None);
    }

    #[test]
    fn test_restart_after_does_not_override_preserved_words() {
        // The likeliest regression: a preserved brand name landing right after a
        // boundary must not be re-capitalized into `IPhone`.
        let rule = restart_rule(&[":"]);
        assert_eq!(
            suggested(&rule, "# Devices: iPhone And Android\n").as_deref(),
            Some("Devices: iPhone and android")
        );

        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::SentenceCase,
            sentence_case_restart_after: vec![":".to_string()],
            ignore_words: vec!["kubectl".to_string()],
            preserve_cased_words: false,
            ..Default::default()
        };
        let rule = MD063HeadingCapitalization::from_config_struct(config);
        assert_eq!(
            suggested(&rule, "# Tools: kubectl And Helm\n").as_deref(),
            Some("Tools: kubectl and helm")
        );
    }

    #[test]
    fn test_restart_after_keeps_md044_canonical_forms() {
        let config = MD063Config {
            enabled: true,
            style: HeadingCapStyle::SentenceCase,
            sentence_case_restart_after: vec![":".to_string()],
            ..Default::default()
        };
        let mut rule = MD063HeadingCapitalization::from_config_struct(config);
        rule.proper_names = vec!["GitHub".to_string()];

        // The canonical form wins over the restart, so this is `GitHub`, not `Github`.
        assert_eq!(
            suggested(&rule, "# Docs: github Actions Guide\n").as_deref(),
            Some("Docs: GitHub actions guide")
        );
        assert_eq!(suggested(&rule, "# Docs: GitHub actions guide\n"), None);
    }

    #[test]
    fn test_restart_after_carries_across_segments() {
        // A boundary in one segment governs the next, so a heading with a link behaves
        // the same as one without.
        let rule = restart_rule(&[":"]);
        assert_eq!(
            suggested(
                &rule,
                "# Overview: [Some Link Here](https://example.com) Trailing Words\n"
            )
            .as_deref(),
            Some("Overview: [Some link here](https://example.com) trailing words")
        );
        assert_eq!(
            suggested(&rule, "# Overview: `code` Then More Words\n").as_deref(),
            Some("Overview: `code` then more words")
        );
    }

    #[test]
    fn test_restart_after_ends_a_sentence_at_the_end_of_link_text() {
        // A reader sees `[see:](url)` as `see:`, so the boundary is where they read it,
        // not at the closing paren of the destination. This is the complement of a
        // boundary before a link carrying into its text.
        let rule = restart_rule(&[":"]);
        assert_eq!(
            suggested(&rule, "# Topic [See:](https://example.com) More Words\n").as_deref(),
            Some("Topic [see:](https://example.com) More words")
        );

        // A boundary that only appears in the destination is not visible prose.
        assert_eq!(
            suggested(&rule, "# Topic [See](https://example.com) More Words\n").as_deref(),
            Some("Topic [see](https://example.com) more words")
        );
    }

    #[test]
    fn test_restart_after_ignores_boundaries_inside_opaque_segments() {
        // Code, HTML and image alt text are preserved verbatim rather than capitalized,
        // so a boundary inside them is not one this rule offers the reader.
        let rule = restart_rule(&[":"]);
        for content in [
            "# Topic `see:` More Words\n",
            "# Topic ![alt:](image.png) More Words\n",
            "# Topic <span title=\"x:\">y</span> More Words\n",
        ] {
            let fixed = suggested(&rule, content).expect("heading should be rewritten");
            assert!(
                fixed.ends_with("more words"),
                "opaque segment restarted the sentence in {content:?}: {fixed}"
            );
        }
    }

    #[test]
    fn test_restart_after_treats_a_leading_link_as_sentence_initial() {
        // A heading opening with visible link text starts the sentence, whether or not
        // sentence restarts are configured.
        for rule in [restart_rule(&[]), restart_rule(&[":"])] {
            assert_eq!(
                suggested(&rule, "# [Some Link Here](https://example.com) Trailing Words\n").as_deref(),
                Some("[Some link here](https://example.com) trailing words")
            );
        }
    }

    #[test]
    fn test_restart_after_fix_is_idempotent() {
        let rule = restart_rule(&[":", ";", "-", "\u{2014}"]);
        for content in [
            "# Requirement 1: Struct to Logger Slice Conversion\n",
            "# Ports: Well-Known Ports Explained\n",
            "# Devices: iPhone And Android\n",
            "# Overview: [Some Link Here](https://example.com) Trailing Words\n",
            "# Setup:\n",
        ] {
            let ctx = LintContext::new(content, crate::config::MarkdownFlavor::Standard, None);
            let once = rule.fix(&ctx).unwrap();
            let ctx = LintContext::new(&once, crate::config::MarkdownFlavor::Standard, None);
            assert_eq!(rule.fix(&ctx).unwrap(), once, "fix is not idempotent for {content:?}");
        }
    }

    #[test]
    fn test_restart_after_ignores_empty_boundary_entries() {
        // An empty string would otherwise end every word, capitalizing the whole heading.
        let rule = restart_rule(&[""]);
        assert_eq!(
            suggested(&rule, "# Requirement 1: Struct to Logger Slice Conversion\n").as_deref(),
            Some("Requirement 1: struct to logger slice conversion")
        );
    }

    // Markdown with Gherkin

    const STYLES: [HeadingCapStyle; 3] = [
        HeadingCapStyle::TitleCase,
        HeadingCapStyle::SentenceCase,
        HeadingCapStyle::AllCaps,
    ];

    /// Every Gherkin structure keyword, each with a name all three styles rewrite:
    /// (heading, title case, sentence case, all caps).
    const GHERKIN_STRUCTURES: [(&str, &str, &str, &str); 6] = [
        (
            "# Feature: the system under test",
            "# Feature: The System Under Test",
            "# Feature: The system under test",
            "# Feature: THE SYSTEM UNDER TEST",
        ),
        (
            "## Background: a shared setup",
            "## Background: A Shared Setup",
            "## Background: A shared setup",
            "## Background: A SHARED SETUP",
        ),
        (
            "## Rule: money is never lost",
            "## Rule: Money Is Never Lost",
            "## Rule: Money is never lost",
            "## Rule: MONEY IS NEVER LOST",
        ),
        (
            "### Scenario: add two numbers",
            "### Scenario: Add Two Numbers",
            "### Scenario: Add two numbers",
            "### Scenario: ADD TWO NUMBERS",
        ),
        (
            "### Scenario Outline: add two numbers",
            "### Scenario Outline: Add Two Numbers",
            "### Scenario Outline: Add two numbers",
            "### Scenario Outline: ADD TWO NUMBERS",
        ),
        (
            "#### Examples: happy path",
            "#### Examples: Happy Path",
            "#### Examples: Happy path",
            "#### Examples: HAPPY PATH",
        ),
    ];

    /// The heading MD063 leaves behind under `flavor`, rewritten or not.
    fn recased(style: HeadingCapStyle, heading: &str, flavor: crate::config::MarkdownFlavor) -> String {
        let rule = create_rule_with_style(style);
        let content = format!("{heading}\n");
        let ctx = LintContext::new(&content, flavor, None);
        let warnings = rule.check(&ctx).unwrap();
        let fixed = rule.fix(&ctx).unwrap();
        assert_eq!(
            warnings.is_empty(),
            fixed == content,
            "a warning and a rewrite must agree for {content:?} under {flavor:?}"
        );
        fixed.trim_end().to_string()
    }

    #[test]
    fn test_mdg_keeps_the_keyword_of_every_structure() {
        // A keyword only names a structure when spelled exactly, so a recased one
        // silently turns the structure into prose.
        for (heading, ..) in GHERKIN_STRUCTURES {
            let keyword = &heading[..=heading.find(':').unwrap()];
            for style in STYLES {
                let fixed = recased(style, heading, crate::config::MarkdownFlavor::MDG);
                assert!(
                    fixed.starts_with(keyword),
                    "{style:?} lost the keyword of {heading:?}: {fixed}"
                );
            }
        }
    }

    #[test]
    fn test_mdg_recases_only_the_name_of_a_structure() {
        for (heading, title, sentence, caps) in GHERKIN_STRUCTURES {
            let mdg = crate::config::MarkdownFlavor::MDG;
            assert_eq!(recased(HeadingCapStyle::TitleCase, heading, mdg), title);
            assert_eq!(recased(HeadingCapStyle::SentenceCase, heading, mdg), sentence);
            assert_eq!(recased(HeadingCapStyle::AllCaps, heading, mdg), caps);
        }
    }

    #[test]
    fn test_standard_flavor_recases_a_keyword_like_any_other_word() {
        // The exemption belongs to the flavor, not to the rule.
        let standard = crate::config::MarkdownFlavor::Standard;
        assert_eq!(
            recased(HeadingCapStyle::TitleCase, "# Feature: the system under test", standard),
            "# Feature: the System Under Test"
        );
        assert_eq!(
            recased(
                HeadingCapStyle::SentenceCase,
                "### Scenario Outline: add two numbers",
                standard
            ),
            "### Scenario outline: add two numbers"
        );
        assert_eq!(
            recased(HeadingCapStyle::AllCaps, "# Feature: the system under test", standard),
            "# FEATURE: THE SYSTEM UNDER TEST"
        );
    }

    #[test]
    fn test_mdg_leaves_a_heading_without_a_colon_to_the_normal_rule() {
        for heading in ["## notes about the system", "## Notes", "# THE SYSTEM"] {
            for style in STYLES {
                assert_eq!(
                    recased(style, heading, crate::config::MarkdownFlavor::MDG),
                    recased(style, heading, crate::config::MarkdownFlavor::Standard),
                    "{style:?} treated {heading:?} as a Gherkin structure"
                );
            }
        }
    }

    #[test]
    fn test_mdg_splits_at_the_first_colon_only() {
        // A later colon belongs to the name, which is prose this rule still owns.
        let mdg = crate::config::MarkdownFlavor::MDG;
        let heading = "## Scenario: ratio: two to one";
        assert_eq!(
            recased(HeadingCapStyle::TitleCase, heading, mdg),
            "## Scenario: Ratio: Two to One"
        );
        assert_eq!(
            recased(HeadingCapStyle::SentenceCase, heading, mdg),
            "## Scenario: Ratio: two to one"
        );
        assert_eq!(
            recased(HeadingCapStyle::AllCaps, heading, mdg),
            "## Scenario: RATIO: TWO TO ONE"
        );
    }

    #[test]
    fn test_mdg_leaves_a_colon_behind_a_backtick_to_the_normal_rule() {
        // Dialect keywords are plain words, so such a colon is inside a code span rather
        // than after a keyword. Splitting there would hide the span from the segment
        // parser and recase what a reader sees as code.
        for heading in [
            "# See `x: y` Notes",
            "# `a: b`",
            "# `code` Feature: a name",
            "# `x: y` Feature: a name",
        ] {
            for style in STYLES {
                assert_eq!(
                    recased(style, heading, crate::config::MarkdownFlavor::MDG),
                    recased(style, heading, crate::config::MarkdownFlavor::Standard),
                    "{style:?} split {heading:?} at a colon inside a code span"
                );
            }
        }
    }

    #[test]
    fn test_mdg_splits_at_a_keyword_colon_that_precedes_a_code_span() {
        // The backtick is in the name, so the keyword colon still governs.
        let mdg = crate::config::MarkdownFlavor::MDG;
        let heading = "# Scenario: use `a: b` here";
        assert_eq!(
            recased(HeadingCapStyle::TitleCase, heading, mdg),
            "# Scenario: Use `a: b` Here"
        );
        assert_eq!(
            recased(HeadingCapStyle::SentenceCase, heading, mdg),
            "# Scenario: Use `a: b` here"
        );
        assert_eq!(
            recased(HeadingCapStyle::AllCaps, heading, mdg),
            "# Scenario: USE `a: b` HERE"
        );
    }

    #[test]
    fn test_mdg_splits_at_a_keyword_colon_before_an_unbalanced_backtick() {
        // An unclosed backtick opens no code span for either flavor, so the tail stays
        // prose and only the keyword is held back.
        let mdg = crate::config::MarkdownFlavor::MDG;
        let heading = "# Scenario: a ` b";
        assert_eq!(recased(HeadingCapStyle::TitleCase, heading, mdg), "# Scenario: A ` B");
        assert_eq!(
            recased(HeadingCapStyle::SentenceCase, heading, mdg),
            "# Scenario: A ` b"
        );
        assert_eq!(recased(HeadingCapStyle::AllCaps, heading, mdg), "# Scenario: A ` B");
    }

    #[test]
    fn test_mdg_keeps_a_keyword_with_nothing_left_to_recase() {
        for style in STYLES {
            assert_eq!(
                recased(style, "# Feature:", crate::config::MarkdownFlavor::MDG),
                "# Feature:"
            );
        }
    }

    #[test]
    fn test_mdg_keeps_a_custom_id_after_the_name() {
        assert_eq!(
            recased(
                HeadingCapStyle::TitleCase,
                "# Feature: the system {#overview}",
                crate::config::MarkdownFlavor::MDG
            ),
            "# Feature: The System {#overview}"
        );
    }

    #[test]
    fn test_mdg_fix_is_idempotent() {
        for (heading, ..) in GHERKIN_STRUCTURES {
            for style in STYLES {
                let rule = create_rule_with_style(style);
                let content = format!("{heading}\n");
                let ctx = LintContext::new(&content, crate::config::MarkdownFlavor::MDG, None);
                let once = rule.fix(&ctx).unwrap();
                let ctx = LintContext::new(&once, crate::config::MarkdownFlavor::MDG, None);
                assert_eq!(
                    rule.fix(&ctx).unwrap(),
                    once,
                    "fix is not idempotent for {heading:?} ({style:?})"
                );
            }
        }
    }
}

#[cfg(test)]
mod shortcode_delimiter_tests {
    use super::*;

    #[test]
    fn heading_capitalization_preserves_shortcode_raw_arguments() {
        let rule = MD063HeadingCapitalization::from_config_struct(MD063Config {
            enabled: true,
            style: HeadingCapStyle::TitleCase,
            ..Default::default()
        });
        for (open, close) in [("{{<", ">}}"), ("{{%", "%}}")] {
            let tag = format!("{open} note title=`literal {close} friend` {close}");
            assert_eq!(
                rule.apply_capitalization(&format!("hello {tag} world"), crate::config::MarkdownFlavor::Hugo),
                format!("Hello {tag} World")
            );
        }
    }
}

#[cfg(test)]
mod mdx_tag_capitalization_tests {
    use super::*;
    use crate::rule::Rule;

    fn fixed(style: HeadingCapStyle, source: &str) -> String {
        let rule = MD063HeadingCapitalization::from_config_struct(MD063Config {
            enabled: true,
            style,
            ..Default::default()
        });
        let ctx = crate::lint_context::LintContext::new(source, crate::config::MarkdownFlavor::MDX, None);
        rule.fix(&ctx).unwrap()
    }

    #[test]
    fn mdx_heading_capitalization_preserves_native_jsx_elements() {
        for name in ["span", "Card", "ui.Card", "svg:path"] {
            let element = format!("<{name} onClick={{() => null}} title=\"literal friend\">world</{name}>");
            let source = format!("# hello {element} after\n");
            for (style, before, after) in [
                (HeadingCapStyle::TitleCase, "Hello", "After"),
                (HeadingCapStyle::SentenceCase, "Hello", "after"),
                (HeadingCapStyle::AllCaps, "HELLO", "AFTER"),
            ] {
                assert_eq!(fixed(style, &source), format!("# {before} {element} {after}\n"));
            }
        }
    }

    #[test]
    fn mdx_heading_capitalization_preserves_nested_member_elements() {
        let element = "<ui.Card title='literal friend'>world <span>inner</span></ui.Card>";
        let source = format!("# hello {element} after\n");
        assert_eq!(
            fixed(HeadingCapStyle::TitleCase, &source),
            format!("# Hello {element} After\n")
        );
    }

    #[test]
    fn mdx_heading_tag_protection_preserves_existing_visible_word_context() {
        assert_eq!(
            fixed(HeadingCapStyle::SentenceCase, "# <span></span> hello world\n"),
            "# <span></span> Hello world\n"
        );
        assert_eq!(
            fixed(HeadingCapStyle::SentenceCase, "# <img alt='friend'/> hello world\n"),
            "# <img alt='friend'/> hello world\n"
        );
        assert_eq!(
            fixed(HeadingCapStyle::TitleCase, "# <span>visible</span> and the world\n"),
            "# <span>visible</span> And the World\n"
        );
    }

    #[test]
    fn mdx_setext_member_attributes_preserve_their_source_map() {
        let member = "hello <ui.Card title=\"literal\nfriend\">world</ui.Card> after\n===\n";
        assert_eq!(fixed(HeadingCapStyle::TitleCase, member), member);
        let html = "hello <span title=\"literal\nfriend\">world</span> after\n===\n";
        assert_eq!(
            fixed(HeadingCapStyle::TitleCase, html),
            "Hello <span title=\"literal\nfriend\">world</span> After\n===\n"
        );
    }
}