feat(core,markdown): harden medical/scientific identifiers, arrow reactions, and table base direction

action/dist/index.cjs

@@ -3536,7 +3536,151 @@
   "tailwind",
   "turbopack",
   "vite",
-  "vitest"
+  "vitest",
+  "atp",
+  "adp",
+  "pep",
+  "rbc",
+  "wbc",
+  "dna",
+  "rna",
+  "g6pd",
+  "pklr",
+  "opsi",
+  "cbc",
+  "ldh",
+  "mchc",
+  "ema",
+  "dat",
+  "aiha",
+  "hs",
+  "b12",
+  "igg",
+  "igm",
+  "c3b",
+  "nadph",
+  "pk",
+  "spleen",
+  "splenectomy",
+  "macrophage",
+  "macrophages",
+  "capsule",
+  "encapsulated",
+  "capsulated",
+  "phagocytosis",
+  "opsonin",
+  "opsonized",
+  "opsonization",
+  "bacteremia",
+  "sepsis",
+  "meningitis",
+  "spherocyte",
+  "spherocytes",
+  "reticulocyte",
+  "reticulocytes",
+  "glycolysis",
+  "mitochondria",
+  "mitochondrial",
+  "hemolysis",
+  "thalassemia",
+  "echinocyte",
+  "echinocytes",
+  "deficiency",
+  "cell",
+  "cells",
+  "autosomal",
+  "recessive",
+  "mutation",
+  "hereditary",
+  "chronic",
+  "nonspherocytic",
+  "hemolytic",
+  "anemia",
+  "pyruvate",
+  "kinase",
+  "glucose",
+  "extravascular",
+  "dissociation",
+  "affinity",
+  "asymptomatic",
+  "symptomatic",
+  "severity",
+  "neonatal",
+  "compensated",
+  "jaundice",
+  "splenomegaly",
+  "gallstones",
+  "retic",
+  "destruction",
+  "marrow",
+  "bilirubin",
+  "haptoglobin",
+  "coombs",
+  "normocytic",
+  "macrocytosis",
+  "morphology",
+  "pathognomonic",
+  "oxidative",
+  "oxidant",
+  "phenotype",
+  "parvovirus",
+  "erythropoiesis",
+  "bpg",
+  "enzyme",
+  "enzymes",
+  "tissue",
+  "assay",
+  "ferritin",
+  "aplastic",
+  "crisis",
+  "prevalence",
+  "underdiagnosis",
+  "prototype",
+  "pneumonia",
+  "pneumoniae",
+  "influenzae",
+  "meningitidis",
+  "streptococcus",
+  "strep",
+  "pneumo",
+  "hib",
+  "neisseria",
+  "polysaccharide",
+  "antigen",
+  "antigens",
+  "antibody",
+  "antibodies",
+  "complement",
+  "bacterium",
+  "bacteria",
+  "clearance",
+  "infection",
+  "organism",
+  "organisms",
+  "immune",
+  "nonimmune",
+  "abnormal",
+  "transfusion",
+  "infancy",
+  "adulthood",
+  "membrane",
+  "integrity",
+  "deformability",
+  "slit",
+  "slits",
+  "ion",
+  "pumps",
+  "retrieval",
+  "mnemonic",
+  "decoder",
+  "fingerprint",
+  "defect",
+  "negative",
+  "positive",
+  "hb",
+  "indirect",
+  "direct",
+  "obvious"
 ]);
 var KNOWN_TECHNICAL_TOKENS = new Set(DEFAULT_TECHNICAL_IDENTIFIERS);
 var NUMERIC_VALUE = "[0-9\\u0660-\\u0669\\u06F0-\\u06F9]+(?:[.,\\u066B\\u066C][0-9\\u0660-\\u0669\\u06F0-\\u06F9]+)*";
@@ -3799,6 +3943,23 @@
   }
   return /[0-9_.]/u.test(token) || /[a-z][A-Z]/u.test(token) || !uppercaseProse && token.length <= ACRONYM_MAXIMUM_LENGTH && /^[A-Z]{2,}$/u.test(token);
 }
+function addReactionRanges(text, ranges) {
+  if (!/[→←↔⇒⇐⇌⇄]|->|<-/u.test(text)) return;
+  const lineRegex = /^[^\r\n]*(?:[→←↔⇒⇐⇌⇄]|->|<-)[^\r\n]*$/gmu;
+  const chainRegex = /(?:[A-Za-z0-9\u2070-\u209F_./+−–,-]{1,50}(?:\s+[A-Za-z0-9\u2070-\u209F_./+−–,-]{1,50}){0,6}(?:\s*[↑↓])?\s*(?:[→←↔⇒⇐⇌⇄]|->|<-)\s*)+(?:[A-Za-z0-9\u2070-\u209F_./+−–,-]{1,50}(?:\s+[A-Za-z0-9\u2070-\u209F_./+−–,-]{1,50}){0,6}(?:\s*[↑↓])?)/gu;
+  let lineMatch;
+  while ((lineMatch = lineRegex.exec(text)) !== null) {
+    const line2 = lineMatch[0];
+    const lineOffset = lineMatch.index;
+    chainRegex.lastIndex = 0;
+    let match;
+    while ((match = chainRegex.exec(line2)) !== null) {
+      const start = lineOffset + match.index;
+      const end = start + match[0].length;
+      addRange(ranges, text, start, end, "identifier");
+    }
+  }
+}
 function findTechnicalTokenRanges(text, technicalIdentifiers = []) {
   const ranges = [];
   addCodeRanges(text, ranges);
@@ -3857,13 +4018,31 @@
   addMatches(text, ranges, /\b[A-Z]{1,4}\s+(?:[IVXLCDM]{1,8}|\d{1,3})\b/gu, "identifier");
   addMatches(text, ranges, /\b[A-Z]{1,4}\/[A-Z]{1,4}\b/gu, "identifier");
   addMatches(text, ranges, /\b[A-Z]\b(?=\s*(?:=|:|→|->))/gu, "identifier");
+  addMatches(text, ranges, /\b[A-Z]\.\s+[a-z]{3,}(?:\s+type\s+[a-z0-9]+)?\b/gu, "identifier");
+  addMatches(text, ranges, /\b\d+,\d+-[A-Z0-9]+(?:\s*[↑↓])?\b/gu, "identifier");
+  addMatches(text, ranges, /\b(?:Worked|Thought)\s+for\s+\d+s?\b/giu, "identifier");
+  addReactionRanges(text, ranges);
+  addMatches(
+    text,
+    ranges,
+    /\b[A-Za-z0-9\u2070-\u209F_.-]+\s*[↑↓]|\b(?:[A-Z0-9]{2,}|Coombs|Rh)\s*[+−-](?!\w)|\b[A-Za-z0-9\u2070-\u209F_.-]+[+−](?!\S)/gu,
+    "identifier"
+  );
+  addMatches(
+    text,
+    ranges,
+    /(?<=(?:ضد|پیش|پس|زیر|ابر|نیمه|بی|نا|غیر)[\u200C-])[A-Za-z][A-Za-z0-9_.-]*/gu,
+    "identifier"
+  );
   const words = /\b[A-Za-z][A-Za-z0-9_.-]*\b/gu;
   const customIdentifiers = customTechnicalIdentifiers(technicalIdentifiers);
   const uppercaseProse = usesUppercaseProse(text);
   let match;
   while ((match = words.exec(text)) !== null) {
     const token = match[0];
-    if (isTechnicalIdentifier(token, customIdentifiers, uppercaseProse)) {
+    const tail = text.slice(match.index + token.length, match.index + token.length + 8);
+    const hasAffix = /^[\u200C]?(?:ها|هایی|های|ای|اش|مان|تان|شان|تر|ترین|ات|ت|م|ی|یی|یها)(?![\p{L}\p{N}])/u.test(tail);
+    if (hasAffix || isTechnicalIdentifier(token, customIdentifiers, uppercaseProse)) {
       addRange(ranges, text, match.index, match.index + token.length, "identifier");
     }
   }

corpus/fixtures/fa/README.md

@@ -0,0 +1,66 @@
+# Real-world ChatGPT Persian bidirectional test fixtures
+
+This directory contains 28 schema-validated conformance fixtures derived directly
+from real-world OpenAI ChatGPT conversations in Persian and mixed English/Persian.
+
+## Background and problem statement
+
+In real-world LLM chat interfaces (such as ChatGPT, Claude, and Gemini), mixed-direction
+Persian text breaks down catastrophically when medical, scientific, or technical terminology
+is introduced.
+
+Standard web implementations (such as `<p dir="auto">` or default LTR containers) apply
+the Unicode Bidirectional Algorithm's P2/P3 rule ("first strong character"). When an AI
+generates a paragraph or sentence that begins with an English technical term, `dir="auto"`
+selects LTR. As a result:
+- The entire Persian paragraph is laid out left-to-right.
+- The leading Latin term appears visually isolated on the far left.
+- Persian words flow backwards relative to the paragraph axis.
+- Terminal punctuation (periods, question marks, colons) jumps to the opposite margin.
+- Neutral arrows (`→`, `←`, `↑`, `↓`) reorder chaotically between RTL and LTR runs.
+- Table columns in Markdown invert or misalign.
+
+## Fixture categories
+
+### 1. Leading English terms in Persian paragraphs
+- `fa-chatgpt-capsule-001`: `Capsule مثل یک روکش لیز و ضدچسب دور باکتری است.`
+- `fa-chatgpt-macrophage-002`: `Macrophage برای بلعیدن باکتری خیلی بهتر عمل میکند...`
+- `fa-chatgpt-pyruvate-kinase-003`: `Pyruvate kinase یکی از آن enzymeهایی است...`
+- `fa-chatgpt-splenic-macrophages-004`: `1) Splenic macrophages خون از spleen عبور میکند...`
+- `fa-chatgpt-spherocyte-005`: `spherocyte واضح ندارد`
+- `fa-chatgpt-target-cell-006`: `target-cell pattern تیپیک thalassemia ندارد`
+- `fa-chatgpt-reticulocyte-007`: `Reticulocyteها enzyme activity بیشتری...`
+- `fa-chatgpt-marginal-zone-028`: `Marginal-zone B cells اینها به polysaccharide capsule...`
+
+### 2. Direction indicators and reaction arrows
+- `fa-chatgpt-equation-atp-009`: `PEP + ADP → Pyruvate + ATP`
+- `fa-chatgpt-immunology-arrow-010`: `IgG / IgM + C3b → phagocytosis ↑`
+- `fa-chatgpt-species-arrow-008`: `S. pneumoniae ← مهمترین`
+- `fa-chatgpt-bacteremia-chain-011`: `bacteremia میتواند خیلی سریع بالا برود → sepsis / meningitis / OPSI`
+- `fa-chatgpt-pathway-process-027`: `glucose → glycolysis → mitochondria → lots of ATP`
+- `fa-chatgpt-bpg-pathway-018`: `PK deficiency = anemia، ولی tissue oxygen delivery...`
+
+### 3. Mixed-script diagnostic tables
+- `fa-chatgpt-table-row-pk-019`: `chronic DAT− hemolysis، معمولاً no spherocytes، گاهی echinocytes`
+- `fa-chatgpt-table-row-aiha-020`: `spherocytes ولی DAT+`
+- `fa-chatgpt-table-row-thalassemia-021`: `microcytosis شدید + target cells`
+- `fa-chatgpt-decoder-cell-022`: `2,3-BPG پرتکننده O₂`
+
+### 4. Leading emoji and bullet callouts
+- `fa-chatgpt-bottom-line-emoji-012`: `🔴 Bottom line:`
+- `fa-chatgpt-memory-hook-emoji-013`: `🧠 تصویر ذهنی`
+- `fa-chatgpt-take-home-emoji-014`: `🔑 Take-home:`
+- `fa-chatgpt-reality-emoji-015`: `✅ واقعیت: ATP deficiency باعث کاهش بقای RBC...`
+
+### 5. Persian quotations and dialogue
+- `fa-chatgpt-quote-persian-016`: `«پس قبلش چرا مشکل ایجاد نمیکنند؟»`
+- `fa-chatgpt-quote-slang-017`: `«این آشغالِ بیبرق دیگه رد نمیشه.»`
+
+### 6. Persian affixes on Latin loanwords
+- Suffixes like `ها` (plural), `هایی` (indefinite plural), `های` (ezāfe), `اش` (pronominal):
+  `macrophageها`, `enzymeهایی`, `slitهای`, `RBCهای`, `Reticulocyteها`, `complicationهایی`.
+
+## Validation
+
+All fixtures in this directory validate against `corpus/fixture.schema.json` via
+`scripts/generate-chatgpt-fixtures.ts` and are exercised by `packages/core/src/conformance-chatgpt-fa.test.ts`.

corpus/fixtures/fa/fa-chatgpt-active-recall-024.json

@@ -0,0 +1,69 @@
+{
+  "id": "fa-chatgpt-active-recall-024",
+  "description": "Active recall test question in Persian with clinical lab values and English test names.",
+  "text": "یک کودک دارد: Hb 8.5، retic 12%، indirect bilirubin↑، haptoglobin↓، DAT−، splenomegaly، no spherocytes.",
+  "words": [
+    "یک",
+    "کودک",
+    "دارد",
+    ":",
+    "Hb",
+    "8.5",
+    "،",
+    "retic",
+    "12",
+    "%",
+    "،",
+    "indirect",
+    "bilirubin↑",
+    "،",
+    "haptoglobin↓",
+    "،",
+    "DAT",
+    "−",
+    "،",
+    "splenomegaly",
+    "،",
+    "no",
+    "spherocytes",
+    "."
+  ],
+  "expected": "rtl",
+  "expectedVisualOrderRightToLeft": [
+    1,
+    2,
+    3,
+    4,
+    5,
+    6,
+    7,
+    8,
+    9,
+    10,
+    11,
+    12,
+    13,
+    14,
+    15,
+    16,
+    17,
+    18,
+    19,
+    20,
+    21,
+    22,
+    23,
+    24
+  ],
+  "tags": [
+    "fa",
+    "chatgpt",
+    "medical",
+    "lab-values",
+    "arrows",
+    "question",
+    "real-world"
+  ],
+  "curation": "user-provided",
+  "nativeSpeakerReviewed": false
+}

corpus/fixtures/fa/fa-chatgpt-bacteremia-chain-011.json

@@ -0,0 +1,43 @@
+{
+  "id": "fa-chatgpt-bacteremia-chain-011",
+  "description": "Medical pathophysiology chain with Persian prose and English clinical outcome.",
+  "text": "bacteremia میتواند خیلی سریع بالا برود → sepsis / meningitis / OPSI",
+  "words": [
+    "bacteremia",
+    "میتواند",
+    "خیلی",
+    "سریع",
+    "بالا",
+    "برود",
+    "→",
+    "sepsis",
+    "/",
+    "meningitis",
+    "/",
+    "OPSI"
+  ],
+  "expected": "rtl",
+  "expectedVisualOrderRightToLeft": [
+    1,
+    2,
+    3,
+    4,
+    5,
+    6,
+    7,
+    8,
+    9,
+    10,
+    11,
+    12
+  ],
+  "tags": [
+    "fa",
+    "chatgpt",
+    "medical",
+    "pathway-arrow",
+    "real-world"
+  ],
+  "curation": "user-provided",
+  "nativeSpeakerReviewed": false
+}