@opacedev/astro-ai-content-checker 0.3.1

package/node_modules/@opacedev/ai-content-checker-core/dist/patterns/en-signals-v3-data.js

@@ -0,0 +1,614 @@
+/**
+ * Rule data for the en-signals 2026.08.3 harvest merge.
+ *
+ * Sources merged here (see research/AI-TELLS-MEGA-PACK.md, research/
+ * ai-tells-pack-seed.json `tells-seed:2026.08.1`, and research/
+ * OWNER-DOCS-TELLS.md):
+ * - Tier A tells ship at their stated severity.
+ * - Tier B tells ship at low severity with `corroboration: true` metadata —
+ *   they are weak alone and only meaningful alongside other findings.
+ * - Tier C tells are NOT implemented; they are recorded in EXCLUDED_TELLS so
+ *   coverage is auditable.
+ *
+ * Binding corrections from the research (do not "fix" these without reading
+ * the mega-pack):
+ * - No naive hedging rule: epistemic hedges DECREASED in post-LLM text
+ *   (arXiv:2603.16131, 2505.09662); a hedge-word counter fires backwards.
+ * - Question-heading / uniform-FAQ shapes stay corroboration-weight only:
+ *   SEO/AEO guides teach humans the identical template.
+ * - Stylometric measures never dominate the document score (cap applied in
+ *   en-signals-v2.ts) — the Stanford TOEFL study found >50% of genuine
+ *   non-native essays falsely flagged by stylometric detectors.
+ *
+ * Licensing: regex/list data here is either (a) adapted from Apache-2.0/MIT
+ * sources (sam-paech/antislop-sampler, slop-forensics, SLOP_Detector,
+ * hwajongpark/slop-gate, avectats7/anti-ai-writing, kjmagnan1s/anti-slop,
+ * aplaceforallmystuff/claude-slop-detector), (b) uncopyrightable facts from
+ * academic papers (Liang et al. 2024, Kobak et al. 2025, Juzek & Ward 2025,
+ * Reinhart et al. PNAS 2025, Pew 2026), (c) independently re-expressed from
+ * CC BY-SA Wikipedia guidance, or (d) the owner's own documents. AGPL
+ * (AlpinDale/gptslop) and unlicensed (jalaalrd) lists were NOT copied; where
+ * the same observations appear they were reimplemented from the underlying
+ * facts. See THIRD_PARTY_NOTICES.md.
+ *
+ * Everything remains Tier B evidence in BRIEF.md §21 terms: editorial hints
+ * about style — never proof of authorship. The artefact-forensics category is
+ * the strongest evidence class (near-zero false positives, model-attributing)
+ * but is still reported as evidence, not verdict.
+ */
+export const RULE_ERA = {
+    // ── existing v2 categories ──
+    "tier1": { era: "2023" },
+    "tier1-clarity": { era: "evergreen" },
+    "tier2": { era: "2023" },
+    "tier3": { era: "2023" },
+    "transition": { era: "2023" },
+    "chatbot": { era: "evergreen", attribution: "multi" },
+    "sycophantic": { era: "evergreen", attribution: "multi" },
+    "filler": { era: "2023" },
+    "generic-conclusion": { era: "2023" },
+    "lets-construction": { era: "evergreen" },
+    "reasoning-artifact": { era: "2025-26" },
+    "acknowledgment-loop": { era: "evergreen" },
+    "significance-inflation": { era: "2024-25" },
+    "vague-attribution": { era: "evergreen" },
+    "hollow-intensifier": { era: "evergreen" },
+    "emotional-flatline": { era: "2024-25" },
+    "lingering-attention": { era: "2025-26" },
+    "novelty-inflation": { era: "2025-26" },
+    "cutoff-disclaimer": { era: "2023", attribution: "multi" },
+    "template-phrase": { era: "2023" },
+    "false-concession": { era: "2024-25" },
+    "rhetorical-question": { era: "2024-25" },
+    "confidence-calibration": { era: "evergreen" },
+    // Post-2025 nuance: OpenAI suppressed em dashes in Nov 2025; by mid-2026
+    // only Claude-family output exceeds professional writers, making density a
+    // partial model-attribution hint rather than a generic AI tell.
+    "em-dash-density": { era: "2025-26", attribution: "claude" },
+    "not-just-contrast": { era: "2024-25" },
+    "uniform-sections": { era: "evergreen" },
+    "uniform-list-items": { era: "evergreen" },
+    "sentence-flatline": { era: "2023" },
+    "uniformity": { era: "evergreen" },
+    "formatting": { era: "evergreen", attribution: "multi" },
+    "tier3-phrase": { era: "2023" },
+    "tier3-phrase-cluster": { era: "2023" },
+    "hashtag-stuff": { era: "evergreen" },
+    "bullet-np-list": { era: "evergreen" },
+    "hedge-stack": { era: "evergreen" },
+    "future-narrative": { era: "2024-25" },
+    "real-actual-inflation": { era: "2025-26" },
+    "social-cta-closer": { era: "2025-26" },
+    "formulaic-opener": { era: "2023" },
+    "speculative-opener": { era: "2023" },
+    "title-case-header": { era: "evergreen" },
+    "parenthetical-hedge": { era: "2025-26" },
+    "smart-punct-signature": { era: "evergreen" },
+    "punct-distribution": { era: "evergreen" },
+    "fnword-trigram-entropy": { era: "evergreen" },
+    "cross-para-burstiness": { era: "evergreen" },
+    "normalization-flag": { era: "evergreen" },
+    "low-ttr": { era: "2023" },
+    "ai-placeholder": { era: "evergreen", attribution: "multi" },
+    "ai-citation-markup": { era: "2025-26", attribution: "multi" },
+    "ai-utm-source": { era: "2024-25", attribution: "multi" },
+    // ── v3 artefact forensics ──
+    "ai-citation-token": { era: "2025-26", attribution: "multi" },
+    "reasoning-leak": { era: "2025-26" },
+    "placeholder-token": { era: "2024-25" },
+    "pua-character": { era: "2025-26", attribution: "chatgpt" },
+    "math-alphanumeric": { era: "2025-26", attribution: "chatgpt" },
+    "arrow-decoration": { era: "2025-26", attribution: "chatgpt" },
+    "escaped-markup-literal": { era: "evergreen", attribution: "multi" },
+    // ── v3 tier A phrase/structural ──
+    "neg-parallelism": { era: "2024-25" },
+    "tripled-negation": { era: "2025-26", attribution: "chatgpt" },
+    "despite-challenges-arc": { era: "2023" },
+    "metaphor-cluster": { era: "2023" },
+    "participial-tail": { era: "evergreen" },
+    "focal-density": { era: "2023" },
+    "owner-phrase": { era: "2023" },
+    "power-verb-compound": { era: "2023" },
+    "outcome-tail": { era: "2023" },
+    "conclusion-cta": { era: "2023" },
+    // ── v3 tier B (corroboration-weight) ──
+    "liang-cluster": { era: "2024-25" },
+    "kobak-density": { era: "2023" },
+    "promo-travel": { era: "2023" },
+    "pivotal-role": { era: "2024-25" },
+    "legacy-framing": { era: "2024-25", attribution: "multi" },
+    "notability-canned": { era: "2025-26" },
+    "buzzword-phrase": { era: "2023" },
+    "faux-insight": { era: "2025-26" },
+    "rhetorical-qa": { era: "2024-25", attribution: "chatgpt" },
+    "didactic-note": { era: "2023" },
+    "narrative-cliche": { era: "2024-25" },
+    "valuable-insights": { era: "2024-25" },
+    "copula-avoidance": { era: "evergreen" },
+    "bold-label-bullets": { era: "evergreen", attribution: "multi" },
+    "emoji-decoration": { era: "2024-25", attribution: "chatgpt" },
+    "heading-inflation": { era: "2024-25", attribution: "gemini" },
+    "staccato-fragments": { era: "2025-26" },
+    "tricolon-density": { era: "evergreen" },
+    "transition-stacking": { era: "2023" },
+    "quote-inconsistency": { era: "2025-26", attribution: "chatgpt" },
+    "token-cutoff": { era: "2023", attribution: "chatgpt" },
+    "setup-expansion-cadence": { era: "2025-26" },
+    "passive-ratio": { era: "evergreen" },
+    "low-specificity": { era: "evergreen" },
+    "adjacent-lemma-repeat": { era: "evergreen" },
+    "fiction-claudeism": { era: "2024-25", attribution: "claude" },
+    "fiction-promptonym": { era: "2024-25", attribution: "multi" },
+    "fiction-slop-phrase": { era: "2024-25", attribution: "multi" },
+    "owner-phrase-b": { era: "2023" },
+    "owner-vocab-b": { era: "2023" },
+    "directive-colon-bullets": { era: "2024-25" },
+    "teach-preach-headings": { era: "2023" },
+    "by-ving-template": { era: "2023" },
+    "invalid-isbn": { era: "evergreen" },
+    "proximity-cluster": { era: "evergreen" },
+    // ── 2026.08.6 provider-eval furniture rules ──
+    // deepseek 99.3% / google-25 95.3% / mistral 94.7% carry bold; heaviest in
+    // the 2024+ chat register across vendors → multi attribution.
+    "markdown-bold": { era: "2024-25", attribution: "multi" },
+    "markdown-heading": { era: "2025-26", attribution: "multi" },
+    "markdown-furniture": { era: "2024-25", attribution: "multi" },
+};
+/** Categories whose findings are corroboration-weight only (research Tier B). */
+export const CORROBORATION_CATEGORIES = new Set([
+    "liang-cluster", "kobak-density", "promo-travel", "pivotal-role",
+    "legacy-framing", "notability-canned", "buzzword-phrase", "faux-insight",
+    "rhetorical-qa", "didactic-note", "narrative-cliche", "valuable-insights",
+    "copula-avoidance", "bold-label-bullets", "emoji-decoration",
+    "heading-inflation", "staccato-fragments", "tricolon-density",
+    "transition-stacking", "quote-inconsistency", "token-cutoff",
+    "setup-expansion-cadence", "passive-ratio", "low-specificity",
+    "adjacent-lemma-repeat", "fiction-claudeism", "fiction-promptonym",
+    "fiction-slop-phrase", "owner-phrase-b", "owner-vocab-b",
+    "directive-colon-bullets", "teach-preach-headings", "by-ving-template",
+    "invalid-isbn", "proximity-cluster", "escaped-markup-literal",
+    // 2026.08.6: markdown-furniture rules are corroboration-weight by design —
+    // their absence must never count in favour of a human verdict, because an
+    // editor paste that strips formatting removes the signal entirely
+    // (PROVIDER-EVAL-2026-08.md §1 honest gaps, §4.1 caveats). Their
+    // classification power comes from the zero-FP escalation floors, not from
+    // severity or weight.
+    "markdown-bold", "markdown-heading", "markdown-furniture",
+]);
+// ─── 2026.08.4 escalation-policy category sets ───────────────────────
+// Derived from the real-world evaluation (research/REAL-WORLD-EVAL-2026-08.md
+// §4a): the rule tier produced artefact evidence on 7/7 artefact-bearing AI
+// samples yet classified almost all of them human_like. These sets feed the
+// post-scoring escalation policy in en-signals-v2.ts. Human controls fired
+// none of these categories (0/4), so the policy adds no human FP risk.
+/**
+ * Tier-A artefact-forensics categories that alone justify flooring the
+ * classification at mixed_signals (near-zero FP; none fired on any human
+ * control in the evaluation).
+ */
+export const ARTEFACT_CORE_CATEGORIES = new Set([
+    "ai-citation-markup", "ai-citation-token", "ai-utm-source",
+    "reasoning-leak", "placeholder-token", "ai-placeholder",
+    "pua-character", "math-alphanumeric",
+]);
+/**
+ * Artefact-adjacent categories that count toward the floor only when
+ * co-occurring with other artefact evidence — the evaluation kept
+ * arrow-decoration corroboration-only on its own (arrows are common in
+ * genuine technical notes; eval §4b).
+ */
+export const ARTEFACT_SUPPORT_CATEGORIES = new Set([
+    "arrow-decoration", "escaped-markup-literal",
+]);
+/**
+ * Chat-export formatting furniture: three or more of these together form the
+ * formatting-cluster compound (eval §4a item 4; evidence opace-openai-006).
+ */
+export const FORMATTING_CLUSTER_CATEGORIES = new Set([
+    "bold-label-bullets", "heading-inflation", "emoji-decoration", "arrow-decoration",
+]);
+// ─── 2026.08.6 provider-eval furniture calibration ───────────────────
+// Measured thresholds from services/local-engine/research/provider-eval/
+// PROVIDER-EVAL-2026-08.md §4.1 (1,727 AI + 169 held-out humans; every
+// number below fired on 0/169 humans, 0/10 business-marketing humans).
+// The bullets rate is R5's measured gate; bold/heading fire on ANY
+// occurrence because both occurred in 0/169 human documents.
+export const V6_FURNITURE_THRESHOLDS = {
+    /** R5: bullet lines per 1,000 words above which bullets alone open the gate. */
+    bulletsPer1000: 10.75,
+};
+/**
+ * Stylometric measurement categories. Their combined contribution to the
+ * document score is capped in en-signals-v2.ts so stylometrics can never
+ * dominate a verdict (research correction: non-native-writer false-positive
+ * landmine; Stanford TOEFL study).
+ */
+export const STYLOMETRIC_CATEGORIES = new Set([
+    "punct-distribution", "cross-para-burstiness", "fnword-trigram-entropy",
+    "sentence-flatline", "uniformity", "uniform-sections", "uniform-list-items",
+    "low-ttr", "smart-punct-signature", "em-dash-density",
+    "setup-expansion-cadence", "passive-ratio", "low-specificity",
+    "adjacent-lemma-repeat", "copula-avoidance", "tricolon-density",
+    "staccato-fragments", "transition-stacking", "heading-inflation",
+]);
+export const AI_CITATION_TOKENS = [
+    { pattern: /【\d+†L\d+(?:-L?\d+)?】/g, attribution: "deepseek" },
+    { pattern: /\bgrok_render_citation_card_json\b/gi, attribution: "grok" },
+    { pattern: /\bgrok-card\s+data-id\b/gi, attribution: "grok" },
+    { pattern: /ppl-ai-file-upload/gi, attribution: "perplexity" },
+    { pattern: /\[attached_file:\d+\]/gi, attribution: "perplexity" },
+    { pattern: /\[web:\d+\]/g, attribution: "perplexity" },
+    { pattern: /\[cite:\s*\d+\]/g, attribution: "gemini" },
+    { pattern: /\[span_\d+\]\(start_span\)/g, attribution: "gemini" },
+    { pattern: /\battributableIndex\b/g, attribution: "chatgpt" },
+    { pattern: /:::writing\{/g, attribution: "chatgpt" },
+    { pattern: /\bciteturn\d+(?:search|image|news|navigation)\d+/gi, attribution: "chatgpt" },
+];
+export const REASONING_LEAKS = [
+    /\bthe\s+user\s+(?:wants|is\s+asking|requested|has\s+asked)\b/gi,
+    /\breviewer\s+note\s*[:—-]/gi,
+    /\bas\s+per\s+(?:the\s+)?(?:system\s+)?prompt\b/gi,
+    /\bso\s+i\s+should\s+(?:structure|frame|word)\s+(?:the|this|my)\b/gi,
+];
+export const PLACEHOLDER_TOKENS = [
+    /\bINSERT_[A-Z][A-Z_]{2,40}\b/g,
+    /\bPASTE_[A-Z][A-Z_]{2,40}\b/g,
+    /\baccess-date\s*=\s*\d{4}-XX-XX\b/gi,
+];
+// Pure codepoint checks implemented HERE as pattern rules because
+// packages/core/src/unicode/ is owned by another workstream — the unicode
+// module reports carriers/confusables; these rules report the pattern-layer
+// editorial reading of the same characters. Overlap is deliberate and noted
+// in the workstream report.
+export const PUA_RANGE_RE = /[\uE000-\uF8FF]/g;
+export const MATH_ALPHANUMERIC_RE = /[\u{1D400}-\u{1D7FF}]/gu;
+/** Spaced arrow used as a prose connector ("input → output"). */
+export const ARROW_CONNECTOR_RE = /(?<=\S)\s[\u2192\u2794\u27A1]\s(?=\S)/g;
+export const ESCAPED_MARKUP_LITERALS = [
+    /&nbsp;/g,
+    /(?<!\\)\\n\\n/g,
+];
+// ─── Tier A phrase / structural rules ────────────────────────────────
+// Negative parallelism, non-dash form ("not only X but also Y"). Flagged only
+// at 2+ occurrences per document (Pew 2026: ~3x human rate; JFK used it too).
+export const NEG_PARALLELISM_RE = /\bnot\s+(?:just|only|merely)\s+[^.!?\n]{2,60}?,?\s+but\s+(?:also\s+|rather\s+)?\w+/gi;
+/** "Not X. Not Y. Just Z." tripled negation (ChatGPT-family favourite). */
+export const TRIPLED_NEGATION_RE = /\b(?:Not|No)\s+[^.!?,\n]{2,30}[.,]\s*(?:Not|no)\s+[^.!?,\n]{2,30}[.,]\s*(?:Just|just)\s+\w+/g;
+/** The rigid "Despite challenges … continues to thrive" essay arc. */
+export const DESPITE_CHALLENGES_RE = /\bDespite\s+(?:these|its|numerous|various|several)\s+(?:challenges|setbacks|obstacles)\b[^.!?\n]{0,120}?(?:continues?\s+to\s+(?:thrive|grow|flourish)|faces?\s+(?:several\s+|numerous\s+)?challenges)/gi;
+/** Figurative metaphor cluster — flag 2+ DISTINCT items per document. */
+export const METAPHOR_CLUSTER_RES = [
+    /\b(?:rich\s+tapestry|tapestry\s+of)\b/gi,
+    /\b(?:complex|intricate)\s+interplay\b/gi,
+    /\b(?:ever-)?evolving\s+landscape\b/gi,
+    /\ba\s+testament\s+to\b/gi,
+    /\bbeacon\s+of\b/gi,
+];
+/**
+ * Sentence-final present-participle significance tail (Reinhart et al. PNAS
+ * 2025; WP:SUPERFICIAL). Flagged only at 3+ per document — one -ing tail is
+ * ordinary English.
+ */
+export const PARTICIPIAL_TAIL_RE = /,\s+(?:highlighting|underscoring|reflecting|symboli[sz]ing|showcasing|emphasi[sz]ing|demonstrating|ensuring|fostering|solidifying|cementing|signaling|signalling|contributing\s+to)\b[^.!?\n]{5,80}[.!?]/g;
+/**
+ * Focal-word density (Juzek & Ward COLING 2025; Kobak Science Advances 2025;
+ * Liang ICML 2024; Pew 2026 — the triple-validated core lexicon). A DENSITY
+ * rule: fires once per document when 3+ hits land at 3+ per 1,000 words.
+ */
+export const FOCAL_WORD_RE = /\b(?:delv(?:e|es|ed|ing)|showcas(?:e|es|ing)|boasts?|underscor(?:e|es|ing)|intricac(?:y|ies)|intricate(?:ly)?|surpass(?:es|ing)|garner(?:ed|s)?|emphasi[sz]ing|groundbreaking|meticulous(?:ly)?|commendable|pivotal|elucidat(?:e|es|ing)|advancements)\b/gi;
+// Owner Tier A phrases (OWNER-DOCS-TELLS.md §8 merge plan; provenance kept in
+// one category so the owner pack stays traceable against the web pack).
+export const OWNER_PHRASES = [
+    /\blook\s+no\s+further\b/gi,
+    /\bcomprehensive\s+suite\s+of\b/gi,
+    /\btailored\s+solutions?\b/gi,
+    /\bunlock(?:ing)?\s+(?:its|your|their|the)\s+full\s+potential\b/gi,
+    /\bbefore\s+diving\s+in(?:to)?\b/gi,
+    /\bcommon\s+pitfalls\s+and\s+how\s+to\s+avoid\s+them\b/gi,
+    /\b(?:break(?:ing)?\s+(?:down\s+)?the\s+process\s+(?:down\s+)?into|process\s+can\s+be\s+broken\s+down\s+into)\b/gi,
+    /\bunderstanding\s+your\s+starting\s+point\b/gi,
+    /\bwhile\s+it\s+might\s+seem\s+counterintuitive\b/gi,
+    /\bcommon\s+choices\s+include\b/gi,
+    /\bwhere\s+things\s+get\s+interesting\b/gi,
+    /\bright\?\s*Well,\s+not\s+any\s*more\b/gi,
+    /\bthe\s+future\s+of\s+[\w\s]{1,30}\s+is\s+increasingly\s+(?:shaped|driven|defined)\s+by\b/gi,
+    /\btake\s+action\s+now\b/gi,
+    /\banother\s+(?:important|key|crucial)\s+(?:aspect|factor|thing|element|point)\s+to\s+consider\b/gi,
+    /\bby\s+doing\s+so,?\s+you(?:['’]ll|\s+can|\s+will)\b/gi,
+    /\bit['’]?s\s+(?:crucial|important|essential)\s+to\s+remember\b/gi,
+    /\bhere\s+are\s+(?:\d+|five|six|seven|eight|nine|ten)\s+(?:simple\s+|key\s+|essential\s+)?(?:steps|things|ways|tips|strategies|best\s+practices|key\s+elements)\b/gi,
+    /\bkey\s+(?:elements|aspects|factors|components|considerations)\s+include\b/gi,
+    /\b(?:to\s+)?put\s+it\s+simply\b/gi,
+    /\bescape\s+the\s+stresses\s+of\s+daily\s+life\b/gi,
+    /\bworks?\s+wonders\b/gi,
+];
+/** Power verb within a few tokens of an intangible buzz-adjective (owner §3e). */
+export const POWER_VERB_COMPOUND_RE = /\b(?:ensur(?:e|es|ing)|leverag(?:e|es|ing)|prioriti[sz](?:e|es|ing)|harness(?:es|ing)?|enhanc(?:e|es|ing)|capitali[sz](?:e|es|ing)\s+on|deliver(?:s|ing)?)\s+(?:\w+\s+){0,3}(?:holistic|seamless|robust|sustainable|scalable|exceptional|strategic|transformative)\b/gi;
+/** ", leading to increased/improved X" outcome tail (owner §3f). */
+export const OUTCOME_TAIL_RE = /,\s*leading\s+to\s+(?:increased|improved|enhanced|greater|better)\s+\w+/gi;
+/** Compound marketing conclusion: "by following these steps … boost/improve". */
+export const CONCLUSION_CTA_RE = /\bby\s+following\s+these\s+(?:steps|tips|strategies|best\s+practices)\b[^.!?\n]{0,80}?\b(?:boost|improve|enhance|transform|elevate|significantly)\w*/gi;
+// ─── Tier B rules (low severity, corroboration) ──────────────────────
+// Liang et al. 2024 overused evaluative adjectives/adverbs (facts from the
+// paper; words already in TIER1/TIER2 omitted to avoid double counting).
+// Cluster rule: 3+ DISTINCT items per document.
+export const LIANG_CLUSTER_RE = /\b(?:versatile|ingenious|methodical(?:ly)?|insightful|laudable|admirable|profound|intriguing(?:ly)?|cogent|lucid|noteworthy|thoughtfully|judiciously|elegantly|compellingly|synergistically|succinctly|comprehensively|strategically|aptly|hitherto|herein)\b/gi;
+// Kobak et al. style-lexicon subset (excess_words.csv derivation; the full
+// 407-word CSV is not bundled — see workstream report). Cluster: 4+ distinct.
+export const KOBAK_CLUSTER_RE = /\b(?:notable|noteworthy|unparalleled|invaluable|culminating|thereby|garnered|surpassing|groundbreaking|commendable|advancements|encompass(?:es|ing)?)\b/gi;
+// Promotional/travel register cluster (seed `lex-promo-travel`): 2+ distinct.
+export const PROMO_TRAVEL_RE = /\b(?:in\s+the\s+heart\s+of|breathtaking|rich\s+cultural\s+heritage|treasure\s+trove|diverse\s+array|must-visit|hidden\s+gem)\b/gi;
+export const PIVOTAL_ROLE_RE = /\bplays?\s+a\s+(?:crucial|pivotal|vital|key)\s+role\s+in(?:\s+shaping)?\b/gi;
+// Legacy/significance framing (WP:AILEGACY): flagged only when STACKED (2+
+// distinct phrases) — historians and obituarists use singles legitimately.
+export const LEGACY_FRAMING_RE = /\b(?:pivotal\s+moment|enduring\s+legacy|indelible\s+mark|key\s+turning\s+point|setting\s+the\s+stage\s+for|deeply\s+rooted)\b/gi;
+export const NOTABILITY_CANNED_RE = /\b(?:profiled\s+in\s+multiple\s+outlets|independent\s+coverage|active\s+social\s+media\s+presence|widely-read\s+outlets)\b/gi;
+export const BUZZWORD_PHRASES = [
+    /\bunlock(?:ing)?\s+the\s+(?:potential|power)\s+of\b/gi,
+    /\bharness(?:ing)?\s+the\s+power\s+of\b/gi,
+    /\bembark(?:ing)?\s+on\s+a\s+journey\b/gi,
+    /\bat\s+the\s+forefront\s+of\b/gi,
+    /\bbridg(?:e|ing)\s+the\s+gap\s+between\b/gi,
+    /\bpav(?:e|ing)\s+the\s+way\s+for\b/gi,
+    /\bpush(?:ing)?\s+the\s+boundaries\s+of\b/gi,
+    /\bnavigat(?:e|ing)\s+the\s+complexit(?:y|ies)\b/gi,
+    /\btake\s+(?:it|things)\s+to\s+the\s+next\s+level\b/gi,
+];
+export const FAUX_INSIGHT_RE = /\b(?:here['’]?s\s+what\s+nobody\s+tells\s+you|what\s+most\s+people\s+get\s+wrong|here['’]?s\s+the\s+kicker|plot\s+twist:|the\s+part\s+everyone\s+misses)\b/gi;
+/** "The result? A platform that scales." — flagged at 2+ per document. */
+export const RHETORICAL_QA_RE = /\b[Tt]he\s+(?:result|goal|answer|solution|problem|catch|best\s+part)\?\s+[A-Z]/g;
+export const DIDACTIC_NOTE_RE = /\b(?:it['’]?s\s+(?:important|crucial|essential)\s+to\s+(?:understand|recognise|recognize)|(?:results|experiences|mileage)\s+may\s+vary|it\s+should\s+be\s+noted)\b/gi;
+export const NARRATIVE_CLICHE_RE = /\b(?:faced\s+numerous\s+challenges|newfound\s+sense\s+of\s+purpose|poignant\s+reminder|serves\s+as\s+a\s+(?:powerful|poignant)\s+reminder)\b/gi;
+export const VALUABLE_INSIGHTS_RE = /\b(?:provid(?:es?|ing)\s+valuable\s+insights?\s+into|key\s+takeaways?\b)/gi;
+/** Copula-avoidance alternatives (Geng & Trotta arXiv:2404.08627). */
+export const COPULA_ALTERNATIVE_RE = /\b(?:serves?|stands?|functions?|operates?)\s+as\s+(?:a|an|the)\b/gi;
+export const BOLD_LABEL_BULLET_RE = /^\s*[-*+•]\s*\*\*[^*\n]{2,40}\*\*[:.]?\s/;
+export const EMOJI_DECOR_RE = /[\u{1F680}\u{2728}\u{1F9E0}\u{2705}\u{1F449}\u{1F4A1}\u{1F3AF}\u{1F525}\u{1F4CC}\u{1F4C8}\u{26A1}\u{1F511}]/u;
+export const RITUAL_HEADING_RE = /^(?:#{1,6}[ \t]+\S.*|<h[1-6][^>]*>.*)$/gim;
+/** ≤4-word staccato fragment run threshold is 3 consecutive. */
+export const STACCATO_MAX_WORDS = 4;
+export const TRANSITION_OPENER_RE = /^\s*(?:Additionally|Moreover|Furthermore|Subsequently|In\s+addition|What['’]s\s+more)\b/i;
+/** Fiction-lane packs (reimplemented from public observations; the AGPL
+ * claudeslop.yaml list was NOT copied — see licensing header). */
+export const FICTION_CLAUDEISM_RE = /\b(?:ministrations|audible\s+pop|rivulets\s+of|half-lidded\s+eyes|despite\s+(?:herself|himself)|with\s+reckless\s+abandon|knuckles\s+(?:turning|turned)\s+white|chuckl(?:es|ed)\s+darkly)\b/gi;
+export const FICTION_PROMPTONYM_RE = /\b(?:Elara\s+(?:Voss|Vex)|Aris\s+Thorne|Elias\s+Vance|Whispering\s+(?:Woods|Pines|Hollow)|Eldoria)\b/g;
+// Frequency-ranked fiction phrases (sam-paech/antislop-sampler, Apache-2.0).
+export const FICTION_SLOP_RE = /\b(?:took\s+a\s+deep\s+breath|voice\s+barely\s+above\s+a\s+whisper|couldn['’]?t\s+help\s+but\s+feel|casting\s+long\s+shadows|shivers?\s+(?:ran|run(?:ning)?)\s+down\s+(?:my|her|his|their)\s+spine|heart\s+pounding\s+in\s+(?:her|his|my)\s+chest|the\s+room\s+fell\s+silent|days\s+turned\s+into\s+weeks|maybe,\s+just\s+maybe|little\s+did\s+(?:she|he|they)\s+know|unbeknownst\s+to\s+(?:them|her|him))\b/gi;
+// Owner Tier B phrases (behind corroboration weighting).
+export const OWNER_PHRASES_B = [
+    /\bshed(?:s|ding)?\s+light\s+on\b/gi,
+    /\ba\s+popular\s+choice\s+for\b/gi,
+    /\bconsult(?:ing)?\s+with\s+a\s+(?:professional|specialist|qualified)\b/gi,
+    /\bas\s+a\s+professional\s*,/gi,
+    /\bstay(?:ing)?\s+ahead\s+of\s+the\s+curve\b/gi,
+    /\bI\s+am\s+not\s+a\s+(?:lawyer|doctor|financial\s+advisor|professional)\s*,?\s+but\b/gi,
+    /\bsafety\s+should\s+never\s+be\s+(?:overlooked|compromised)\b/gi,
+    /\bpresent\s+challenges\s+due\s+to\b/gi,
+    /\bextending\s+their\s+lifespan\b/gi,
+    /\beven\s+the\s+best\s+plans\s+can\b/gi,
+];
+/** Owner ban-list vocabulary, tier B (2+ distinct per document). */
+export const OWNER_VOCAB_B_RE = /\b(?:essence|facets?|exhaustive|pesky|folks)\b/gi;
+export const DIRECTIVE_COLON_BULLET_RE = /^\s*(?:[-*+•]|\d+[.)])\s*(?:Plan|Ensure|Optimi[sz]e|Enhance|Leverage|Prioriti[sz]e|Implement|Utili[sz]e|Consider|Embrace)\b[^.:\n]{0,50}:/;
+export const TEACH_PREACH_HEADING_RE = /^#{1,6}\s+(?:why\s+(?:it|this)\s+(?:matters|is\s+important)|how\s+to\s+get\s+started|final\s+thoughts|key\s+takeaways|common\s+pitfalls)\s*\??\s*$/gim;
+export const BY_VING_TEMPLATE_RE = /\bBy\s+\w+ing\s+[^,.\n]{5,60},\s+you\s+can\s+\w+/g;
+/** Passive-voice heuristic: be-form + past participle. */
+export const PASSIVE_RE = /\b(?:is|are|was|were|been|being|be)\s+(?:\w+ly\s+)?\w{3,}(?:ed|en|wn|lt)\b/gi;
+// ─── Weights and metadata ────────────────────────────────────────────
+export const V3_ISSUE_WEIGHTS = {
+    // artefact forensics
+    "ai-citation-token": 15,
+    "reasoning-leak": 12,
+    "placeholder-token": 10,
+    "pua-character": 14,
+    "math-alphanumeric": 12,
+    "arrow-decoration": 4,
+    "escaped-markup-literal": 3,
+    // tier A phrase/structural
+    "neg-parallelism": 5,
+    "tripled-negation": 5,
+    "despite-challenges-arc": 5,
+    "metaphor-cluster": 4,
+    "participial-tail": 5,
+    "focal-density": 5,
+    "owner-phrase": 5,
+    "power-verb-compound": 6,
+    "outcome-tail": 4,
+    "conclusion-cta": 6,
+    // tier B (low weights; corroboration)
+    "liang-cluster": 2,
+    "kobak-density": 2,
+    "promo-travel": 2,
+    "pivotal-role": 2,
+    "legacy-framing": 3,
+    "notability-canned": 2,
+    "buzzword-phrase": 2,
+    "faux-insight": 2,
+    "rhetorical-qa": 2,
+    "didactic-note": 2,
+    "narrative-cliche": 3,
+    "valuable-insights": 2,
+    "copula-avoidance": 3,
+    "bold-label-bullets": 3,
+    "emoji-decoration": 2,
+    "heading-inflation": 3,
+    "staccato-fragments": 3,
+    "tricolon-density": 2,
+    "transition-stacking": 3,
+    "quote-inconsistency": 2,
+    "token-cutoff": 2,
+    "setup-expansion-cadence": 3,
+    "passive-ratio": 3,
+    "low-specificity": 2,
+    "adjacent-lemma-repeat": 3,
+    "fiction-claudeism": 3,
+    "fiction-promptonym": 3,
+    "fiction-slop-phrase": 2,
+    "owner-phrase-b": 2,
+    "owner-vocab-b": 2,
+    "directive-colon-bullets": 3,
+    "teach-preach-headings": 2,
+    "by-ving-template": 3,
+    "invalid-isbn": 3,
+    "proximity-cluster": 2,
+    // 2026.08.6 furniture rules (low weights; the escalation floors carry the
+    // detection, and the weights stay small so furniture cannot fake breadth).
+    "markdown-bold": 3,
+    "markdown-heading": 3,
+    "markdown-furniture": 4,
+};
+/**
+ * The paste caveat, shared by the three chat-formatting rules.
+ *
+ * It is rule-specific honesty, not the generic claim boundary: these three
+ * rules can only see chat formatting that survived the paste, so their SILENCE
+ * carries no information at all and the message has to say so. Kept as one
+ * constant because the three rules make exactly the same point.
+ */
+const PASTE = "This only shows up when chat formatting survives the paste. If an editor stripped the formatting, the check finds nothing, which says nothing either way.";
+/** Writing that is formal, or by someone whose first language is not English, reads this way too. */
+const FORMAL = "Formal writing, and writing by people whose first language is not English, can read this way too.";
+export const V3_CATEGORY_META = {
+    "ai-citation-token": { severity: "high", message: "A chatbot's own citation code is sitting in the text. The code itself says which chatbot it came from.", suggestion: "Delete the code and put a real reference in its place." },
+    "reasoning-leak": { severity: "high", message: "Parts of this text talk about the writing job itself, the kind of notes an AI leaves in its answer, like \"as requested\" or \"let me revise\".", suggestion: "Delete those notes and keep the finished writing." },
+    "placeholder-token": { severity: "high", message: "A machine placeholder is still in the text, something like \"{{name}}\" or \"<insert>\".", suggestion: "Fill it in or delete it before this goes out." },
+    "pua-character": { severity: "high", message: "The text holds characters from a private corner of Unicode that has no agreed meaning. ChatGPT wraps its citation codes in these. Icon fonts are the only other common reason for them.", suggestion: "Delete them, and look for chatbot citation codes next to them." },
+    "math-alphanumeric": { severity: "high", message: "Fake bold or italic letters built from maths symbols (𝗹𝗶𝗸𝗲 𝘁𝗵𝗶𝘀). They come from chatbot copy-paste and social-media text formatters.", suggestion: "Retype them as ordinary letters and use real bold or italic." },
+    "arrow-decoration": { severity: "medium", message: "Arrows stand in for words again and again: \"input → output → result\".", suggestion: "Write the connection out in words." },
+    "escaped-markup-literal": { severity: "low", message: "A stray scrap of code is showing through the text, like \"&nbsp;\" or \"\\n\". It usually comes from pasting out of a chat window.", suggestion: "Delete it." },
+    "neg-parallelism": { severity: "medium", message: "The \"not only X but Y\" shape comes back more than once.", suggestion: "Keep one. Say the rest plainly." },
+    "tripled-negation": { severity: "medium", message: "The \"Not X. Not Y. Just Z.\" shape.", suggestion: "Say what it is, without the three-part build-up." },
+    "despite-challenges-arc": { severity: "medium", message: "The stock \"despite challenges, it continues to thrive\" story shape.", suggestion: "Name the real problem and what was really done about it." },
+    "metaphor-cluster": { severity: "medium", message: "Several worn-out picture words are stacked together: \"tapestry\", \"interplay\", \"evolving landscape\", \"testament to\".", suggestion: "Say the plain facts they are standing in for." },
+    "participial-tail": { severity: "medium", message: "Sentences keep ending with a tacked-on \"-ing\" clause that tells you why it mattered: \", highlighting the need for…\", \", underscoring the importance of…\". That ending turns up several times more often in machine writing than in human writing.", suggestion: "Stop the sentence at the fact. Cut the tail, or turn it into a claim you can source." },
+    "focal-density": { severity: "medium", message: "A lot of words from the AI-favourite list turn up here: \"delve\", \"showcase\", \"pivotal\", \"meticulous\". Any one of them is normal English. It is how many there are that stands out.", suggestion: "Swap most of them for plainer verbs and adjectives." },
+    "owner-phrase": { severity: "medium", message: "A ready-made phrase from the generic-writing phrasebook.", suggestion: "Replace it with something specific to your subject." },
+    "power-verb-compound": { severity: "high", message: "A big verb glued to a vague adjective: \"leverage a robust solution\", \"ensure seamless delivery\". It sounds like value and says nothing.", suggestion: "Name the real action and the thing you can measure." },
+    "outcome-tail": { severity: "medium", message: "The sentence trails off into a vague result: \", leading to increased engagement\".", suggestion: "Say the exact result, or cut the tail." },
+    "conclusion-cta": { severity: "high", message: "The stock marketing sign-off: \"by following these steps you can boost…\".", suggestion: "Close on something specific, not a general promise." },
+    "liang-cluster": { severity: "low", message: "Several judgement words from the AI-overuse lists sit close together, like \"crucial\", \"notable\" and \"significant\".", suggestion: "Keep the judgements you can back up with detail." },
+    "kobak-density": { severity: "low", message: "Several words sit here that turned up far more often in machine writing when a large body of text was counted.", suggestion: "Vary the words, or back the claims with detail." },
+    "promo-travel": { severity: "low", message: "Brochure words are bunched together: \"nestled\", \"breathtaking\", \"hidden gem\". They stand out most when the piece is not a travel brochure.", suggestion: "Describe the place or the product with real detail." },
+    "pivotal-role": { severity: "low", message: "The \"plays a crucial role in shaping\" formula.", suggestion: "Say what it actually does." },
+    "legacy-framing": { severity: "low", message: "Several grand phrases about legacy and importance are stacked up: \"enduring legacy\", \"pivotal moment\".", suggestion: "Let the events speak for themselves." },
+    "notability-canned": { severity: "low", message: "A canned line about how well known something is: \"profiled in multiple outlets\".", suggestion: "Name the outlets, or drop the claim." },
+    "buzzword-phrase": { severity: "low", message: "A stock office phrase: \"harness the power of\", \"at the forefront of\".", suggestion: "Say what it can actually do." },
+    "faux-insight": { severity: "low", message: "A line promising a secret: \"here's what nobody tells you\".", suggestion: "Show the point. Do not announce it." },
+    "rhetorical-qa": { severity: "low", message: "The \"The result? X.\" trick keeps coming back.", suggestion: "Use it once at most." },
+    "didactic-note": { severity: "low", message: "A teacherly disclaimer: \"it's important to understand\", \"results may vary\".", suggestion: "Cut it, or make it specific." },
+    "narrative-cliche": { severity: "low", message: "A worn-out story phrase: \"faced numerous challenges\", \"a poignant reminder\".", suggestion: "Say what actually happened." },
+    "valuable-insights": { severity: "low", message: "A stock academic filler phrase: \"provides valuable insights into\".", suggestion: "State the insight itself." },
+    "copula-avoidance": { severity: "low", message: "The text keeps dodging plain \"is\" and \"has\" in favour of \"serves as\", \"stands as\" and \"functions as\". " + FORMAL, suggestion: "Use \"is\" and \"has\" where they fit." },
+    "bold-label-bullets": { severity: "low", message: "A run of bullets all shaped \"**Label:** description\". Technical documents use this shape too.", suggestion: "Turn it into sentences, or vary the shape of the items." },
+    "emoji-decoration": { severity: "low", message: "Emoji are used on headings or bullets. That is a chat-window habit in a piece of business writing.", suggestion: "Take the decorative emoji out of this kind of writing." },
+    "heading-inflation": { severity: "low", message: "There are a lot of headings for the amount of writing under them. SEO advice produces the same shape.", suggestion: "Merge sections whose body is only a sentence or two." },
+    "staccato-fragments": { severity: "low", message: "A run of very short, punchy fragments. Ad copy does this on purpose too.", suggestion: "Join some of them into full sentences." },
+    "tricolon-density": { severity: "low", message: "Three-part lists such as \"faster, cheaper, simpler\" turn up a lot for a piece this long.", suggestion: "Break the pattern: use two items, or four." },
+    "transition-stacking": { severity: "low", message: "Most paragraphs open on a formal joining word: \"Furthermore\", \"Moreover\", \"Additionally\". " + FORMAL, suggestion: "Let the content do the joining in most paragraphs." },
+    "quote-inconsistency": { severity: "low", message: "Curly and straight quotation marks are mixed together. That usually comes from pasting out of a chat window, though word processors do it too.", suggestion: "Make the quotation marks match, either way round." },
+    "token-cutoff": { severity: "low", message: "The text stops in the middle of a sentence, the shape of an answer that ran out of room, or a paste that went wrong.", suggestion: "Finish the last sentence, or delete it." },
+    "setup-expansion-cadence": { severity: "low", message: "Short sentence, then long one, over and over, or the other way round. " + FORMAL, suggestion: "Keep the pattern only where the short sentence works on its own." },
+    "passive-ratio": { severity: "low", message: "A lot of the sentences hide who did the thing: \"mistakes were made\". That is high for a blog or a marketing piece. Academic writing does it constantly. " + FORMAL, suggestion: "Rewrite most sentences so the person or thing doing it comes first." },
+    "low-specificity": { severity: "low", message: "Almost no numbers, dates or names for a piece this long. Corporate writers produce empty writing too.", suggestion: "Add facts a reader could go and check." },
+    "adjacent-lemma-repeat": { severity: "low", message: "Neighbouring sentences keep reusing the same word. " + FORMAL, suggestion: "Merge the repetitive sentences, or change the wording where it reads naturally." },
+    "fiction-claudeism": { severity: "low", message: "Phrases that turn up a lot in Claude's fiction writing. Romance authors use several of them quite normally.", suggestion: "Put the stock phrases into your own words." },
+    "fiction-promptonym": { severity: "low", message: "A character name that AI writing produces far more often than people do. \"Elara Voss\" is the best-known one.", suggestion: "Pick a less loaded name if you want one." },
+    "fiction-slop-phrase": { severity: "low", message: "Several well-worn fiction lines land in the same piece. All of them were human clichés long before AI.", suggestion: "Cut or rework the stock moments." },
+    "owner-phrase-b": { severity: "low", message: "A phrase from the second generic-writing phrasebook.", suggestion: "Replace it with something specific." },
+    "owner-vocab-b": { severity: "low", message: "Several second-string filler words sit close together: \"essence\", \"facet\", \"pesky\", \"folks\".", suggestion: "Swap them for plainer words where they add nothing." },
+    "directive-colon-bullets": { severity: "low", message: "Several list items open on an order and a colon: \"Ensure X:\", \"Optimise Y:\". Real technical checklists look like this too.", suggestion: "Vary the way the items are built." },
+    "teach-preach-headings": { severity: "low", message: "Stock tutorial headings hold the piece together: \"Why it matters\", \"Final thoughts\", \"Key takeaways\".", suggestion: "Name each section after what is actually in it." },
+    "by-ving-template": { severity: "low", message: "The \"By doing X, you can Y\" shape keeps coming back.", suggestion: "Say the benefit straight out most of the time." },
+    "invalid-isbn": { severity: "low", message: "An ISBN in the text does not add up. Made-up references often fail this check, and so do typos.", suggestion: "Check the reference against the real book." },
+    "proximity-cluster": { severity: "low", message: "The same flagged buzzword comes back within a sentence or two.", suggestion: "Keep one use at most in each passage." },
+    // 2026.08.6 chat-formatting rules. Each one states the paste caveat: the
+    // check can only see formatting that survived the paste, so its ABSENCE says
+    // nothing about who wrote the text.
+    "markdown-bold": { severity: "low", message: "Raw **bold** markdown is showing in the text. None of the 169 human-written documents we checked had it. " + PASTE, suggestion: "Delete the stars, or apply real bold." },
+    "markdown-heading": { severity: "low", message: "A raw markdown heading line, the kind that starts with # signs, is showing in the text. None of the 169 human-written documents we checked had one. " + PASTE, suggestion: "Turn it into a real heading, or delete it." },
+    "markdown-furniture": { severity: "low", message: "Chat-window formatting shapes this text: runs of bold, heading lines, or a wall of bullets. None of the 169 human-written documents we checked had that combination. " + PASTE, suggestion: "Rebuild the formatting properly for wherever this is going." },
+};
+/**
+ * Harvest tells deliberately NOT implemented as user-facing rules. Each entry
+ * names the seed/owner id and the reason (research tier C, correction, gate
+ * the engine cannot provide, or licensing). Kept exported so coverage of the
+ * research harvest is auditable in code.
+ */
+export const EXCLUDED_TELLS = [
+    { id: "lex-magic-adverbs", reason: "Tier C: feature-writing staple (quietly/deeply/fundamentally); too common in human prose to flag." },
+    { id: "lex-clinical-formality", reason: "Tier C: standard register in government/legal/academic prose; ESL writers are taught this vocabulary." },
+    { id: "lex-latinate-shift", reason: "Tier C: formal register and ESL academic training produce the same corpus shift; needs baselines the rule engine lacks." },
+    { id: "lex-concrete-overuse", reason: "Tier C: ordinary word; only notable at density in platform-specific defensive replies." },
+    { id: "lex-crosslingual-34", reason: "Tier C: non-English module; per-language calibration and time decay required before any flag ships." },
+    { id: "lex-multilingual-translationese", reason: "Tier C: overlaps native bureaucratic registers that predate AI; held for a non-English module." },
+    { id: "phr-in-connection-with", reason: "Tier C: standard legal/police-report phrasing; only iterative abundance signals." },
+    { id: "phr-false-ranges", reason: "Tier C: separating fake from real 'from X to Y' ranges needs semantics, not regex." },
+    { id: "phr-colon-reveal", reason: "Tier C: genuinely good device; recurrence-only signal with heavy human-copywriter overlap." },
+    { id: "phr-audience-bracketing", reason: "Tier C in the seed for the generic form; the engine keeps only its pre-existing template-phrase variant." },
+    { id: "phr-analogy-template", reason: "Tier C: explainer journalism uses 'Think of X as the Y of Z' constantly and well; needs semantic judgement." },
+    { id: "phr-paired-adjectives", reason: "Tier C: legal drafting doubles synonyms deliberately; common human filler." },
+    { id: "phr-weak-verbs", reason: "Tier C: 'is designed to' is accurate when describing genuine design intent." },
+    { id: "phr-flip-script", reason: "Tier C: sports and culture journalists genuinely say it." },
+    { id: "phr-refers-to", reason: "Tier C: textbook glossaries legitimately use 'refers to'." },
+    { id: "phr-fake-profound-kicker", reason: "Tier C: personal essayists earn the aphoristic closer; detecting the unearned variant needs semantics." },
+    { id: "phr-conservation-boilerplate", reason: "Tier C: niche domain; conservationists write about status legitimately." },
+    { id: "phr-x-rather-than-y", reason: "Tier C: completely standard English; only meaningful at very high density (Grok attribution hint recorded in docs)." },
+    { id: "phr-ultimate-guide", reason: "Tier C: human SEO writers used 'ultimate guide' titles since ~2010." },
+    { id: "art-stale-access-dates", reason: "Tier C: copied citations, offline drafting and batch merges confound." },
+    { id: "pun-oxford-comma", reason: "Tier C / anti-tell: house style for many publishers; corpus-drift indicator only, never per-document." },
+    { id: "pun-colon-titles", reason: "Tier C: academic titles used colons for decades; anecdotal evidence base." },
+    { id: "pun-human-marker-deficit", reason: "Tier C: absence-of-punctuation features are weak alone; plain-style human writers avoid semicolons deliberately." },
+    { id: "str-nominalisation", reason: "Tier C: needs POS tagging and per-register calibration; bureaucratic/legal prose is naturally nominalisation-heavy." },
+    { id: "str-title-case-headings", reason: "Tier C as a style-mismatch rule (needs house-style config); the engine's pre-existing register-gated title-case rule is retained unchanged." },
+    { id: "str-compound-headings", reason: "Tier C: humans write 'Awards and recognition' sections; only exact canned wordings at corpus scale signal." },
+    { id: "str-fractal-summaries", reason: "Tier C: essay pedagogy teaches 'tell them what you told them'; needs semantic detection." },
+    { id: "str-anaphora", reason: "Tier C: classical rhetoric and speechwriting staple." },
+    { id: "str-question-headings", reason: "Tier C + binding correction: deliberately taught SEO/AEO practice — human-authored AEO content trips this; needs a genre prior. Not implemented." },
+    { id: "str-uniform-faq", reason: "Tier C + binding correction: SEO guides teach humans this exact snippet template; the pre-existing uniform-sections measurement stays corroboration-weight only and is capped with the other stylometrics." },
+    { id: "str-unnecessary-tables", reason: "Tier C: 'trivial content tabulated' needs judgement; data journalists use tables well." },
+    { id: "str-deletion-test", reason: "Tier C: highest-FP category; could ship as a QUALITY metric someday, never as an authenticity flag." },
+    { id: "str-citation-absence", reason: "Tier C: genre-dependent absence signal; short or informal human texts also skip citations." },
+    { id: "str-elegant-variation", reason: "Tier C: needs coreference resolution; old-school journalism taught the same habit." },
+    { id: "str-genre-glitch", reason: "Tier C: needs register classification; feature journalists write deliberate colour." },
+    { id: "str-balanced-noncommitment", reason: "Tier C: diplomatic and academic hedging is human too." },
+    { id: "str-manufactured-personality", reason: "Tier C frontier: fake casual dev-blog voice defeats wordlists; needs semantic/behavioural detection." },
+    { id: "str-invented-labels", reason: "Tier C: academics coin terms legitimately with definition and citation." },
+    { id: "str-false-agency", reason: "Tier C: metonymy is standard English ('the market rallied'); cluster corroborator only." },
+    { id: "str-edit-summary-formulas", reason: "Tier C: platform-specific (wiki edit summaries); future surface for commits/CMS revision notes, not the core engine." },
+    { id: "str-defensive-comments", reason: "Tier C: platform-specific comment repertoire; polite non-native speakers write formally." },
+    { id: "str-hallucinated-conventions", reason: "Tier C: requires a platform knowledge base; new users make similar mistakes by hand." },
+    { id: "str-essay-scaffold", reason: "Tier B in the seed but needs genre awareness the engine lacks; its measurable parts ship via heading-inflation, transition-stacking and the pre-existing uniformity rules." },
+    { id: "str-markdown-leakage-full", reason: "Tier A only INSIDE a surface gate the core engine does not have (markdown-native vs plain-text venue). Bold/heading density rules pre-exist; only the always-safe escaped-literal subset ships (escaped-markup-literal). Full rule awaits the surface concept in the WordPress layer." },
+    { id: "sty-low-perplexity", reason: "Binding exclusion: >50% non-native TOEFL false-positive rate; legal exposure; out of scope for a rule engine." },
+    { id: "sty-ttr-paradox", reason: "Tier C: contradictory directions by register; existing low-ttr rule stays low-weight instead." },
+    { id: "sty-hedging-reversal", reason: "Binding correction: hedges DECREASED in post-LLM text — a naive hedging rule fires backwards and is deliberately not built. The pre-existing hedge-stack rule targets the specific modal+adverb stack, not hedge frequency." },
+    { id: "sty-positivity-skew", reason: "Tier C: PR and brand copy is deliberately positive; lazy human reviewers write the same praise sandwich." },
+    { id: "sty-homogenisation", reason: "Tier C: corpus-level (multi-document) feature; not a per-document rule." },
+    { id: "sty-style-shift", reason: "Tier C: needs per-author history; future multi-document feature." },
+    { id: "sty-machine-cleanliness", reason: "Anti-tell: Grammarly makes human text machine-clean; penalising cleanliness smears good writers. Never scored." },
+    { id: "sty-anti-tells", reason: "The false-positive spec: perfect grammar, 'robotic feel', formality, isolated transitions, Oxford commas and em-dash presence are documented INEFFECTIVE indicators — encoded here as exclusions and exercised by QA fixtures." },
+    { id: "sty-human-whitelist", reason: "Inverse signals (typos, first-person anecdote, pre-Nov-2022 dates) belong to a future negative-weight/protect-list layer, not to positive flags; flagging their absence would cross BRIEF.md §5's claim boundary." },
+    { id: "sty-first-words", reason: "Unverified methodology and unlicensed source repo; not copied, not implemented." },
+    { id: "mod-grok-idiolect", reason: "Tier C: social scientists use the vocabulary natively; attribution hint recorded in docs only." },
+    { id: "mod-vendor-accents", reason: "Tier C: ML-classifier roadmap item (BRIEF §21 Tier C model), not a rule." },
+    { id: "mod-structural-dialects", reason: "Tier C: vendor-blog evidence; used for calibration expectations, not scoring." },
+    { id: "mod-rlhf-root-cause", reason: "Engine-design finding, not a surface tell." },
+    { id: "art-fabricated-refs-network", reason: "Dead-link/DOI checks need network access the offline engine forbids; only the offline ISBN checksum ships (invalid-isbn)." },
+    { id: "owner-journey-bare", reason: "Owner list, high FP: bare 'journey' needs a user-journey/travel guard the rule layer cannot express reliably; only phrase forms ship." },
+    { id: "owner-tier-c-vocab", reason: "Owner tier C (thus, ultimately, nonetheless, promptly, remnant, arena, era, cool, essential, substantially, on the other hand, world of, competitive advantages, break down, stepping into): very high FP ordinary English; documented, not flagged." },
+    { id: "owner-shed-article-fragments", reason: "Topic-specific residue from one AI article (caulk/weatherstripping etc.); not generic tells." },
+    { id: "owner-absence-of-humanity", reason: "Zero-first-person, no-typos, no-emphasis-irregularity: flagging the absence of human traits crosses BRIEF.md §5 from editorial evidence toward authorship inference. Documented only." },
+    { id: "owner-code-comment-tells", reason: "Owner tier C: only relevant if the engine ever scans code identifiers/comments." },
+    { id: "owner-meta-description-template", reason: "Owner tier B but requires the WordPress meta-field surface; not implementable on plain text." },
+    { id: "owner-cross-page-templates", reason: "Owner D2 §3.1 cross-page comparison is site-level; out of scope for the single-document engine." },
+    { id: "lex-kobak-style-407-full", reason: "The full 407-word excess_words.csv is not bundled (licence file unverified in-repo); a representative regex subset ships as kobak-density." },
+    // ── 2026.08.6: provider-eval risk-tiered candidates (PROVIDER-EVAL-2026-08.md
+    // §4.2) — measured NONZERO human false positives on the 169-human corpus.
+    // Deliberately NOT implemented; each needs an explicit owner decision.
+    { id: "pe-tier3-threshold-lowering", reason: "Provider-eval §4.2/§4.3: lowering the tier3 threshold (0.8555/0.85/0.80) buys 33.7-75.7% TPR at 1-7 measured human FPs, dominated by business-marketing prose. ML workstream's recalibration problem, not a rules change; do not ship without owner decision." },
+    { id: "pe-emdash-p99-6.59", reason: "Provider-eval §4.2: em-dash > 6.59/1000 (human p99) flags 2/169 humans (1 business-marketing). Em-dash rules already carry a documented human-FP history; owner decision required." },
+    { id: "pe-fragmentshare-p99", reason: "Provider-eval §4.2: fragmentShare > 0.319 flags 2/169 humans. Business-marketing humans are themselves choppy; owner decision required." },
+    { id: "pe-shortsentshare-p99", reason: "Provider-eval §4.2: shortSentShare > 0.535 flags 2/169 humans (1 business-marketing); owner decision required." },
+    { id: "pe-flatline-escalation", reason: "Provider-eval §4.2: sentence-flatline with >=3 findings => mixed_signals flags 1/169 humans (qa-finance); owner decision required." },
+    { id: "pe-emdash-extreme-10.07", reason: "Provider-eval §4.1 footnote: em-dash > 10.07/1000 (the human maximum) measured zero-FP but yields only openai-25 18.0 / grok 12.0 with all other slices <=10.7, is GPT/Grok-attributing rather than AI-detecting, and sits one essayist beyond the measured human max. Held back with the risk tier pending owner decision; the five headline zero-FP changes shipped instead." },
+];