<?php
/**
 * The content analysis behind the AI audit (docs/05 §3.2) and the editor panel (§3.3): passages,
 * quotability, and the page-level checks from config/checks-ai.yaml. Pure functions over HTML.
 *
 * The same analysis runs in JavaScript (assets/analyzer.js) for the live editor panel. Both are
 * tested against tests/analyzer-fixtures.json, so the audit table and the panel agree.
 *
 * @package GetReport
 */

defined( 'ABSPATH' ) || exit;

/**
 * Content analysis.
 */
class GetReport_Analyzer {

	/** Words that start a passage which only makes sense after the one before (per language). */
	const ANAPHORA = array(
		'en' => array( 'this', 'that', 'these', 'those', 'as mentioned', 'as noted', 'as we', 'as you can see', 'however', 'above' ),
		'hr' => array( 'to', 'ovo', 'ono', 'oni', 'kao što je spomenuto', 'kao što smo', 'međutim', 'također' ),
		'de' => array( 'dies', 'diese', 'dieser', 'dieses', 'wie erwähnt', 'wie oben', 'außerdem', 'jedoch' ),
	);

	/** Verbs that make a first sentence definitional ("X is …", "The best … are …"). */
	const DEFINITIONAL = array(
		'en' => '/(?<![\p{L}\p{N}])(is|are|was|means|refers to|stands for|is called|costs?|takes?|lasts?|opens?)(?![\p{L}\p{N}])/iu',
		'hr' => '/(?<![\p{L}\p{N}])(je|su|znači|označava|predstavlja|košta|traje|radi)(?![\p{L}\p{N}])/iu',
		'de' => '/(?<![\p{L}\p{N}])(ist|sind|bedeutet|bezeichnet|kostet|dauert|öffnet)(?![\p{L}\p{N}])/iu',
	);

	/** Question words that make a heading a question even without "?". */
	const QUESTION_WORDS = array(
		'en' => '/^(what|why|how|when|where|who|which|can|do|does|is|are|should|will)(?![\p{L}\p{N}])/iu',
		'hr' => '/^(što|sto|zašto|kako|kada|gdje|tko|koji|koja|koje|može|mogu|je li|jesu li|trebam|treba)(?![\p{L}\p{N}])/iu',
		'de' => '/^(was|warum|wie|wann|wo|wer|welche|welcher|welches|kann|können|ist|sind|soll)(?![\p{L}\p{N}])/iu',
	);

	/** Language that words compare things in, so a table would help. */
	const COMPARISON = array(
		'en' => '/(?<![\p{L}\p{N}])(vs|versus|compared (to|with)|comparison|better than|difference between|pros and cons)(?![\p{L}\p{N}])/iu',
		'hr' => '/(?<![\p{L}\p{N}])(vs|u usporedbi|usporedba|nasuprot|razlika između|bolje od|prednosti i nedostaci)(?![\p{L}\p{N}])/iu',
		'de' => '/(?<![\p{L}\p{N}])(vs|im vergleich|vergleich|gegenüber|unterschied zwischen|besser als|vor- und nachteile)(?![\p{L}\p{N}])/iu',
	);

	/** Freshness cues: a year, a month name, or "updated" (month names match as word starts). */
	const FRESHNESS = '/(?<![\p{L}\p{N}])(19|20)\d{2}(?![\p{L}\p{N}])|(?<![\p{L}\p{N}])(updated|last updated|ažurirano|aktualisiert|january|february|march|april|may|june|july|august|september|october|november|december|siječ|velja|ožuj|trav|svib|lip|srp|kolov|ruj|listop|stud|prosin|januar|februar|märz|juni|juli|oktober|dezember)/iu';

	/** Words ignored when matching a question to passages. */
	const STOP = array( 'a', 'an', 'the', 'is', 'are', 'do', 'does', 'i', 'you', 'we', 'my', 'your', 'our', 'to', 'of', 'in', 'on', 'for', 'and', 'or', 'can', 'what', 'how', 'why', 'where', 'when', 'who', 'which', 'with', 'it', 'be', 'u', 'na', 'i', 'je', 'li', 'se', 'za', 'od', 'do', 'što', 'kako', 'gdje', 'der', 'die', 'das', 'und', 'ist', 'ein', 'eine', 'wie', 'was', 'wo', 'für', 'mit', 'von', 'zu' );

	/**
	 * The language family used for rules: en, hr or de (others use the English rules).
	 *
	 * @param string $locale Locale such as hr_HR or de-DE.
	 */
	public static function lang( $locale ) {
		$l = strtolower( substr( (string) $locale, 0, 2 ) );
		if ( in_array( $l, array( 'hr', 'bs', 'sr', 'sl' ), true ) ) {
			return 'hr';
		}
		return in_array( $l, array( 'de' ), true ) ? 'de' : 'en';
	}

	/**
	 * Plain text of an HTML fragment.
	 *
	 * @param string $html HTML.
	 */
	public static function text( $html ) {
		$t = preg_replace( '/<(script|style)\b[^>]*>.*?<\/\1>/is', ' ', (string) $html );
		$t = preg_replace( '/<br\s*\/?>/i', ' ', $t );
		$t = html_entity_decode( wp_strip_all_tags( $t ), ENT_QUOTES | ENT_HTML5, 'UTF-8' );
		return trim( preg_replace( '/\s+/u', ' ', $t ) );
	}

	/**
	 * Words of a text.
	 *
	 * @param string $text Text.
	 * @return string[]
	 */
	public static function words( $text ) {
		preg_match_all( "/[\p{L}\p{N}][\p{L}\p{N}'’\-]*/u", (string) $text, $m );
		return $m[0];
	}

	/**
	 * Sentences of a text.
	 *
	 * @param string $text Text.
	 * @return string[]
	 */
	public static function sentences( $text ) {
		$s = preg_split( '/(?<=[.!?])\s+(?=[\p{Lu}\p{N}"„“])/u', trim( (string) $text ) );
		return array_values( array_filter( (array) $s, 'strlen' ) );
	}

	/**
	 * The blocks of a page: headings and text passages, in order.
	 *
	 * @param string $html HTML.
	 * @return array<int, array{type:string, level:int, text:string}>
	 */
	public static function blocks( $html ) {
		preg_match_all( '/<(h[1-6]|p|li|blockquote|dd|dt|figcaption|td)\b[^>]*>(.*?)<\/\1>/is', (string) $html, $m, PREG_SET_ORDER );
		$out = array();
		foreach ( $m as $b ) {
			$tag  = strtolower( $b[1] );
			$text = self::text( $b[2] );
			if ( '' === $text ) {
				continue;
			}
			$heading = 'h' === $tag[0];
			$out[]   = array(
				'type'  => $heading ? 'heading' : 'text',
				'level' => $heading ? (int) $tag[1] : 0,
				'text'  => $text,
			);
		}
		return $out;
	}

	/**
	 * Whether a heading is a question.
	 *
	 * @param string $text Heading.
	 * @param string $lang en | hr | de.
	 */
	public static function is_question( $text, $lang ) {
		return '?' === substr( rtrim( $text ), -1 ) || (bool) preg_match( self::QUESTION_WORDS[ $lang ], ltrim( $text ) );
	}

	/**
	 * Specific facts in a text: numbers and proper nouns (capitalised words that don't start a
	 * sentence), plus currency and unit symbols.
	 *
	 * @param string $text Text.
	 */
	public static function facts( $text ) {
		$n = 0;
		foreach ( self::sentences( $text ) as $s ) {
			$w = self::words( $s );
			foreach ( $w as $i => $word ) {
				if ( preg_match( '/\p{N}/u', $word ) ) {
					++$n;
				} elseif ( $i > 0 && preg_match( '/^\p{Lu}/u', $word ) ) {
					++$n;
				}
			}
		}
		return $n + preg_match_all( '/[€$£%]|(?<![\p{L}\p{N}])(km|kg|cm|mm|m²|ml|mg|min|h|kWh|GB|MB)(?![\p{L}\p{N}])/u', $text );
	}

	/**
	 * Reading grade: Flesch-Kincaid for English, LIX (as a grade) for other languages.
	 *
	 * @param string $text Text.
	 * @param string $lang en | hr | de.
	 */
	public static function grade( $text, $lang ) {
		$words = self::words( $text );
		$nw    = count( $words );
		if ( $nw < 5 ) {
			return 0.0;
		}
		$ns = max( 1, count( self::sentences( $text ) ) );
		if ( 'en' === $lang ) {
			$syl = 0;
			foreach ( $words as $w ) {
				$syl += max( 1, preg_match_all( '/[aeiouy]+/i', preg_replace( '/e$/i', '', strtolower( $w ) ) ) );
			}
			return round( max( 0, 0.39 * $nw / $ns + 11.8 * $syl / $nw - 15.59 ), 1 );
		}
		$long = 0;
		foreach ( $words as $w ) {
			if ( mb_strlen( $w ) > 6 ) {
				++$long;
			}
		}
		$lix = $nw / $ns + 100 * $long / $nw;
		return round( max( 0, ( $lix - 10 ) / 3 ), 1 );
	}

	/**
	 * Quotability of one passage, 0-100, from the spec §4.2 weights.
	 *
	 * @param array  $p Passage: text, pos (0-1), under_question.
	 * @param array  $w Weights (GetReport_Registry::quotability()).
	 * @param string $lang en | hr | de.
	 * @return array{score:int, features:array<string, float>}
	 */
	public static function quotability( $p, $w, $lang ) {
		$words  = self::words( $p['text'] );
		$nw     = max( 1, count( $words ) );
		$first  = self::sentences( $p['text'] )[0] ?? $p['text'];
		$grade  = self::grade( $p['text'], $lang );
		$best   = $w['readability_best_grade'] ?? array( 12, 16 );
		$worst  = (float) ( $w['readability_penalty_above'] ?? 19 );
		$target = (float) ( $w['entity_density_target'] ?? 0.2 );
		$lower  = mb_strtolower( $p['text'] );
		$ana    = false;
		foreach ( self::ANAPHORA[ $lang ] as $a ) {
			if ( 0 === strpos( $lower, $a . ' ' ) || 0 === strpos( $lower, $a . ',' ) ) {
				$ana = true;
				break;
			}
		}
		$f     = array(
			'position_first_third'   => $p['pos'] < 0.33 ? 1.0 : 0.0,
			'definitional_form'      => ( count( self::words( $first ) ) <= 40 && preg_match( self::DEFINITIONAL[ $lang ], $first ) ) ? 1.0 : 0.0,
			'under_question_heading' => ! empty( $p['under_question'] ) ? 1.0 : 0.0,
			'fact_density'           => min( 1.0, self::facts( $p['text'] ) / $nw / $target ),
			'self_contained'         => $ana ? 0.0 : 1.0,
			'readability'            => $grade <= $best[1] ? 1.0 : ( $grade >= $worst ? 0.0 : ( $worst - $grade ) / ( $worst - $best[1] ) ),
			'freshness_cue'          => preg_match( self::FRESHNESS, $p['text'] ) ? 1.0 : 0.0,
		);
		$score = 0.0;
		foreach ( $f as $k => $v ) {
			$score += ( $w[ $k ] ?? 0 ) * $v;
		}
		return array(
			'score'    => (int) round( 100 * $score ),
			'features' => $f,
		);
	}

	/**
	 * How well a passage matches a question: the share of the question's content words in it.
	 *
	 * @param string   $text Passage.
	 * @param string[] $terms Question terms (lowercase, stop words removed).
	 */
	public static function match( $text, $terms ) {
		if ( ! $terms ) {
			return 0.0;
		}
		$have = array_map( 'mb_strtolower', self::words( $text ) );
		$hit  = 0;
		foreach ( $terms as $t ) {
			// A shared stem counts, so "ceramics" matches "ceramic" and "keramiku" matches "keramika".
			$stem = mb_substr( $t, 0, max( 4, mb_strlen( $t ) - 2 ) );
			foreach ( $have as $w ) {
				if ( $w === $t || ( mb_strlen( $stem ) >= 4 && 0 === strpos( $w, $stem ) ) ) {
					++$hit;
					break;
				}
			}
		}
		return $hit / count( $terms );
	}

	/**
	 * Content words of a question.
	 *
	 * @param string $q Question.
	 * @return string[]
	 */
	public static function terms( $q ) {
		$out = array();
		foreach ( self::words( $q ) as $w ) {
			$w = mb_strtolower( $w );
			if ( mb_strlen( $w ) > 2 && ! in_array( $w, self::STOP, true ) ) {
				$out[] = $w;
			}
		}
		return array_values( array_unique( $out ) );
	}

	/**
	 * The full analysis of one page.
	 *
	 * @param array $doc html, title, slug, question, lang, post_type, modified (timestamp), now,
	 *                   stale_months, and site-level facts in 'site' (see GetReport_Audit::site_facts()).
	 * @return array{score:int, subs:array<string, int|null>, fails:string[], passed:string[], passages:array, quotability:int, words:int, grade:float, answer_pos:float|null, details:array<string, string>}
	 */
	public static function analyze( $doc ) {
		$lang   = self::lang( $doc['lang'] ?? 'en' );
		$w      = GetReport_Registry::quotability();
		$blocks = self::blocks( $doc['html'] );
		$texts  = array_values(
			array_filter(
				$blocks,
				function ( $b ) {
					return 'text' === $b['type'];
				}
			)
		);
		$all    = implode( ' ', array_column( $texts, 'text' ) );
		$total  = max( 1, count( self::words( $all ) ) );

		// Passages with their position (share of words before them) and heading context.
		$passages = array();
		$before   = 0;
		$question = false;
		$headings = array();
		foreach ( $blocks as $b ) {
			if ( 'heading' === $b['type'] ) {
				$question = self::is_question( $b['text'], $lang );
				if ( $b['level'] >= 2 && $b['level'] <= 3 ) {
					$headings[] = $b['text'];
				}
				continue;
			}
			$n = count( self::words( $b['text'] ) );
			if ( $n >= 8 ) {
				$p          = array(
					'text'           => $b['text'],
					'pos'            => $before / $total,
					'under_question' => $question,
				);
				$q          = self::quotability( $p, $w, $lang );
				$passages[] = $p + $q;
			}
			$before += $n;
		}

		$site    = $doc['site'] ?? array();
		$fails   = array();
		$ran     = array();
		$details = array();
		$check   = function ( $id, $failed, $detail = '' ) use ( &$fails, &$ran, &$details ) {
			$ran[] = $id;
			if ( $failed ) {
				$fails[] = $id;
				if ( '' !== $detail ) {
					$details[ $id ] = $detail;
				}
			}
		};

		// Answer quality.
		$question_text = trim( (string) ( $doc['question'] ?? '' ) );
		$terms         = self::terms( '' !== $question_text ? $question_text : (string) ( $doc['title'] ?? '' ) );
		$answer_pos    = null;
		if ( $terms && $passages ) {
			$best = null;
			foreach ( $passages as $p ) {
				$m = self::match( $p['text'], $terms );
				if ( null === $best || $m > $best[0] + 0.0001 ) {
					$best = array( $m, $p['pos'] );
				}
			}
			if ( $best[0] > 0 ) {
				$answer_pos = $best[1];
				/* translators: %d: percent of the page. */
				$check( 'answer-not-first-third', $best[1] > 0.33, sprintf( __( 'The best answer starts at %d %% of the page.', 'getreport' ), (int) round( 100 * $best[1] ) ) );
			}
		}
		if ( $passages ) {
			$first3 = array_slice( $passages, 0, 3 );
			$check( 'no-definitional-sentence', 0 === count( array_filter( array_column( array_column( $first3, 'features' ), 'definitional_form' ) ) ) );
		}
		if ( count( $headings ) >= 2 ) {
			$q = count(
				array_filter(
					$headings,
					function ( $h ) use ( $lang ) {
						return self::is_question( $h, $lang );
					}
				)
			);
			/* translators: 1: question headings, 2: all headings. */
			$check( 'no-question-headings', 0 === $q, sprintf( __( '%1$d of %2$d headings are questions.', 'getreport' ), $q, count( $headings ) ) );
		}
		if ( $total >= 80 ) {
			$check( 'low-fact-density', self::facts( $all ) / $total < 0.05 );
		}
		if ( count( $passages ) >= 2 ) {
			$dependent = array_filter(
				$passages,
				function ( $p ) {
					return 0.0 === $p['features']['self_contained'];
				}
			);
			$first     = reset( $dependent );
			$check( 'passages-not-self-contained', count( $dependent ) / count( $passages ) >= 0.25, $first ? '“' . wp_html_excerpt( $first['text'], 40, '…' ) . '”' : '' );
		}
		$grade = self::grade( $all, $lang );
		if ( $total >= 50 ) {
			$check( 'reading-level-too-hard', $grade > (float) ( $w['readability_penalty_above'] ?? 19 ) );
		}
		$slug = (string) ( $doc['slug'] ?? '' );
		if ( '' !== $slug ) {
			$check( 'opaque-url-slug', (bool) preg_match( '/^(\d+|(p|page|post|untitled|auto-draft)-?\d*)$/', $slug ) || ! preg_match( '/\p{L}{3,}/u', urldecode( $slug ) ) );
		}
		$check( 'no-tables-or-lists', preg_match( self::COMPARISON[ $lang ], $all ) && ! preg_match( '/<(table|ul|ol)\b/i', $doc['html'] ) );
		if ( 'product' !== ( $doc['post_type'] ?? '' ) ) {
			/* translators: %d: number of words. */
			$check( 'thin-content', $total < 150, sprintf( __( '%d words.', 'getreport' ), $total ) );
		}

		// Freshness and trust.
		if ( isset( $site['shows_date'] ) ) {
			$check( 'no-visible-date', ! $site['shows_date'] );
		}
		if ( ! empty( $doc['modified'] ) ) {
			$months = (int) floor( ( (int) ( $doc['now'] ?? time() ) - (int) $doc['modified'] ) / ( 30.44 * DAY_IN_SECONDS ) );
			/* translators: %d: months. */
			$check( 'stale-content', $months >= (int) ( $doc['stale_months'] ?? 18 ), sprintf( __( 'Last changed %d months ago.', 'getreport' ), $months ) );
		}
		if ( isset( $site['shows_author'] ) ) {
			$check( 'no-author', ! $site['shows_author'] );
		}
		if ( preg_match_all( '/\p{N}+/u', $all ) >= 3 ) {
			$home = strtolower( (string) ( $site['host'] ?? '' ) );
			$ext  = false;
			if ( preg_match_all( '/<a\b[^>]*href=["\']https?:\/\/([^\/"\':]+)/i', $doc['html'], $links ) ) {
				foreach ( $links[1] as $h ) {
					$h = strtolower( $h );
					if ( ! in_array( $home, array( $h, 'www.' . $h ), true ) && 'www.' . $home !== $h ) {
						$ext = true;
						break;
					}
				}
			}
			$check( 'no-outbound-sources', ! $ext );
		}
		if ( isset( $site['about_contact'] ) ) {
			$check( 'no-about-contact', ! $site['about_contact'] );
		}

		// Access, readable without JavaScript, entity: site-level facts, per page where it matters.
		if ( isset( $site['robots_blocked'] ) ) {
			$check( 'ai-bot-robots-blocked', (bool) $site['robots_blocked'], (string) ( $site['robots_detail'] ?? '' ) );
		}
		if ( isset( $doc['noindex'] ) ) {
			$check( 'noindex-or-x-robots', (bool) $doc['noindex'] );
		}
		if ( isset( $doc['nojs_missing'] ) && null !== $doc['nojs_missing'] ) {
			/* translators: %d: percent of the words. */
			$check( 'nojs-main-content-missing', $doc['nojs_missing'] >= 0.4, sprintf( __( '%d %% of the text is missing from the HTML crawlers get.', 'getreport' ), (int) round( 100 * $doc['nojs_missing'] ) ) );
		}
		if ( isset( $site['same_as'] ) ) {
			$check( 'missing-sameas', ! $site['same_as'] );
		}
		if ( isset( $site['schema'] ) ) {
			$check( 'schema-missing', ! $site['schema'] );
		}
		if ( isset( $site['llms_txt'] ) ) {
			$check( 'llms-txt-missing', ! $site['llms_txt'] );
		}

		$scores = self::score( $ran, $fails );
		$quote  = self::page_quotability( $passages );
		return array(
			'score'       => $scores['score'],
			'subs'        => $scores['subs'],
			'fails'       => $fails,
			'passed'      => array_values( array_diff( $ran, $fails ) ),
			'passages'    => $passages,
			'quotability' => $quote,
			'words'       => $total,
			'grade'       => $grade,
			'answer_pos'  => $answer_pos,
			'details'     => $details,
		);
	}

	/**
	 * A page's quotability: the mean of its three best passages (what an AI would lift).
	 *
	 * @param array $passages Passages with a score.
	 */
	public static function page_quotability( $passages ) {
		if ( ! $passages ) {
			return 0;
		}
		$s = array_column( $passages, 'score' );
		rsort( $s );
		$top = array_slice( $s, 0, 3 );
		return (int) round( array_sum( $top ) / count( $top ) );
	}

	/**
	 * Readiness score and sub-scores from the checks that ran (spec §5.1): each sub-score is the
	 * share of its check weight that passed; the score is the weighted mean of the sub-scores that
	 * had checks; weak-evidence checks take off at most 2 points in total.
	 *
	 * @param string[] $ran Checks that ran.
	 * @param string[] $fails Checks that failed.
	 * @return array{score:int, subs:array<string, int|null>}
	 */
	public static function score( $ran, $fails ) {
		$checks = GetReport_Registry::checks();
		$subs   = array();
		$tot    = array();
		$bad    = array();
		$weak   = 0.0;
		foreach ( $ran as $id ) {
			$c = $checks[ $id ] ?? null;
			if ( ! $c || empty( $c['weight'] ) || 'observed' === $c['subscore'] ) {
				continue;
			}
			if ( 'weak' === $c['evidence'] ) {
				$weak += in_array( $id, $fails, true ) ? (float) $c['weight'] : 0;
				continue;
			}
			$tot[ $c['subscore'] ] = ( $tot[ $c['subscore'] ] ?? 0 ) + $c['weight'];
			if ( in_array( $id, $fails, true ) ) {
				$bad[ $c['subscore'] ] = ( $bad[ $c['subscore'] ] ?? 0 ) + $c['weight'];
			}
		}
		$sum  = 0.0;
		$wsum = 0.0;
		foreach ( GetReport_Registry::subscores() as $key => $s ) {
			if ( empty( $tot[ $key ] ) ) {
				$subs[ $key ] = null;
				continue;
			}
			$subs[ $key ] = (int) round( 100 * ( 1 - ( $bad[ $key ] ?? 0 ) / $tot[ $key ] ) );
			$sum         += $s['weight'] * $subs[ $key ];
			$wsum        += $s['weight'];
		}
		$score = $wsum > 0 ? $sum / $wsum : 0;
		return array(
			'score' => (int) max( 0, round( $score - min( 2.0, $weak ) ) ),
			'subs'  => $subs,
		);
	}
}