sepehr/office_translator
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat(quality): add L0 quality layer (Track A1 + A2 of dev plan)
All checks were successful
Deploy to Production / Build and Deploy (push) Successful in 3m5s
Details

Browse Source 
L0 quality detection layer to catch translation failures BEFORE they
reach users. Pure Python/TypeScript, zero new dependencies, no API calls.

Backend (Python — services/quality/):
  - Script detection: 145 langs mapped to 23 scripts (Latin, Cyrillic,
    Greek, Arabic, Hebrew, CJK, Hangul, Kana, Devanagari, Bengali, etc.)
  - Language confusion detection (e.g. Arabic text for French target)
  - Arabic-script variant discrimination (Persian/Urdu/Pashto/Kurdish
    confusion — e.g. Persian text returned when Arabic was requested)
  - Length sanity check (with numeric/short-source exemptions)
  - Prompt leak detection (Translation: / Voici la traduction: / 翻译:)
  - Repetition hallucination detection (token + character level)
  - File text extraction for .docx/.xlsx/.pptx/.pdf (no translator
    changes needed)
  - Defensive pipeline that never raises (L0 must NEVER break a job)

Frontend (TypeScript — wordly.art---traduction-de-documents/src/utils/):
  - Exact 1:1 mirror of the Python module
  - Zero dependencies, works in browser AND Node.js
  - Native Unicode regex (\\p{L}/u) and codePoint iteration
  - 63 tests using Node's built-in test runner

Integration:
  - Feature-flagged: QUALITY_L0_ENABLED=false (default)
  - Observation only: logs structured events, never modifies files
  - try/except wrapped: impossible to break a translation job
  - Lazy imports: only loaded when flag is on
  - Zero impact on existing tests / behavior

Tests:
  - 111 Python tests covering all paths (config, script, length, leak,
    pipeline, file_extractor) — 100% pass
  - 63 TypeScript tests (Node --test) — 100% pass
  - 174/174 total tests for the L0 layer

Bug fixes in script mapping:
  - yi (Yiddish) -> hebrew (was incorrectly mapped to arabic)
  - dv (Maldivian) -> thaana (was incorrectly mapped to arabic)
  - ja (Japanese) -> hiragana_katakana (distinguishes from Chinese CJK)

Phase 1 (backend) + Phase 2 (frontend) of Track A complete.
Next: Track B1 (Word/Excel format preservation quick wins).

Closes Track A phase 1+2 of the dev plan.
This commit is contained in:
sepehr
2026-07-14 16:17:43 +02:00
parent ebb2537fda
commit f403b2851d
20 changed files with 3132 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
.env.example
View File

@@ -96,6 +96,14 @@ TRANSLATIONS_PER_MINUTE=10
TRANSLATIONS_PER_HOUR=50 TRANSLATIONS_PER_HOUR=50
MAX_CONCURRENT_TRANSLATIONS=5 MAX_CONCURRENT_TRANSLATIONS=5
# ============== Quality Layer (L0) ==============
# Track A1 of the dev plan — observability only, no behavior change.
# When enabled, the L0 layer logs the script detection result for each
# translation job but does NOT modify the file or the job status.
# Default: false (opt-in). Set to "true" to enable.
QUALITY_L0_ENABLED=false
QUALITY_L0_SAMPLE_SIZE=20
# ============== Cleanup Service ============== # ============== Cleanup Service ==============
# Enable automatic file cleanup # Enable automatic file cleanup
CLEANUP_ENABLED=true CLEANUP_ENABLED=true

9
config.py
View File

@@ -68,6 +68,15 @@ class Config:
) )
MAX_MEMORY_PERCENT = float(os.getenv("MAX_MEMORY_PERCENT", "80")) MAX_MEMORY_PERCENT = float(os.getenv("MAX_MEMORY_PERCENT", "80"))
# ============== Quality Layer (L0) ==============
# Track A1 of the dev plan — observability only, no behavior change.
# Set to "true" to enable. Default: false (opt-in).
QUALITY_L0_ENABLED = os.getenv("QUALITY_L0_ENABLED", "false").lower() == "true"
# Number of text samples to extract from the output file for L0 analysis.
# Keep small to avoid overhead. 20 is enough to catch language confusion.
QUALITY_L0_SAMPLE_SIZE = int(os.getenv("QUALITY_L0_SAMPLE_SIZE", "20"))
# ============== API Configuration ============== # ============== API Configuration ==============
API_TITLE = "Document Translation API" API_TITLE = "Document Translation API"
API_VERSION = "1.0.0" API_VERSION = "1.0.0"

28
routes/translate_routes.py
View File

@@ -1354,6 +1354,34 @@ async def _run_translation_job(
f"{changed}/{attempted} ({ratio:.1%})" f"{changed}/{attempted} ({ratio:.1%})"
) )
# ------------------------------------------------------------------
# Quality L0 layer (Track A1 — observability only)
# Extract a small sample of text from the output file and run the
# L0 quality checks. NEVER blocks the job, NEVER modifies the file.
# Enabled by feature flag QUALITY_L0_ENABLED (default: false).
# ------------------------------------------------------------------
if getattr(config, "QUALITY_L0_ENABLED", False):
try:
from services.quality import run_l0_check, extract_sample
samples = extract_sample(
output_path,
file_extension,
max_samples=getattr(config, "QUALITY_L0_SAMPLE_SIZE", 20),
)
translated_chunks = [s["translated"] for s in samples]
run_l0_check(
source_chunks=[""] * len(translated_chunks), # L0 doesn't need source
translated_chunks=translated_chunks,
target_lang=target_lang,
job_id=job_id,
file_extension=file_extension,
)
except Exception as l0_err:
# Quality L0 must NEVER break a job. Log and continue.
logger.warning(
f"Job {job_id}: quality L0 layer failed: {l0_err}"
)
if user_id: if user_id:
# Determine cost factor based on selected provider and model # Determine cost factor based on selected provider and model
cost_factor = 1 cost_factor = 1

35
services/quality/__init__.py Normal file
View File

@@ -0,0 +1,35 @@
"""
Quality check layer for translations.
Track A1 — L0 backend (observation only).
Pure Python, no new dependencies, no network calls.
Designed to be ADDITIVE: existing translation flow is untouched.
Public API:
QualityCheckResult — per-chunk result dataclass
DocumentQualityResult — aggregated result dataclass
evaluate_chunk(...) — score a single (source, translation) pair
evaluate_document(...) — score a list of pairs and aggregate
run_l0_check(...) — defensive wrapper used by the route
extract_sample(...) — extract text from a finished file
"""
from .script_detector import (
QualityCheckResult,
DocumentQualityResult,
evaluate_chunk,
evaluate_document,
detect_arabic_variant,
)
from .pipeline import run_l0_check
from .file_extractor import extract_sample
__all__ = [
"QualityCheckResult",
"DocumentQualityResult",
"evaluate_chunk",
"evaluate_document",
"detect_arabic_variant",
"run_l0_check",
"extract_sample",
]

235
services/quality/config.py Normal file
View File

@@ -0,0 +1,235 @@
"""
Language → script mapping for the L0 quality layer.
A "script" groups languages that share a Unicode block. The script detector
verifies that the *translation* is in the same script as the *target language*.
Languages are mapped by their ISO 639-1 (or 639-3) code. Anything not mapped
falls back to the "latin" script so we never crash on an unknown language.
Arabic-script languages (ar, fa, ur, ps, ku, sd, ug, yi, dv, ckb) all share
the same Unicode ranges, so we additionally use a set of *discriminating
characters* to tell them apart — e.g. Persian text contains چ/پ/ژ/گ which
Arabic text doesn't.
"""
from typing import Dict, FrozenSet, List, Optional, Tuple
# ---------- Unicode ranges per script ----------
# Format: list of (start, end) inclusive ranges.
# Latin is intentionally omitted (it's the fallback).
UNICODE_RANGES: Dict[str, List[Tuple[int, int]]] = {
"cyrillic": [
(0x0400, 0x04FF), # Cyrillic
(0x0500, 0x052F), # Cyrillic Supplement
],
"greek": [
(0x0370, 0x03FF), # Greek and Coptic
],
"arabic": [
(0x0600, 0x06FF), # Arabic
(0x0750, 0x077F), # Arabic Supplement
(0x08A0, 0x08FF), # Arabic Extended-A
],
"hebrew": [
(0x0590, 0x05FF), # Hebrew
],
"devanagari": [
(0x0900, 0x097F), # Devanagari
],
"bengali": [
(0x0980, 0x09FF), # Bengali
],
"tamil": [
(0x0B80, 0x0BFF),
],
"telugu": [
(0x0C00, 0x0C7F),
],
"kannada": [
(0x0C80, 0x0CFF),
],
"malayalam": [
(0x0D00, 0x0D7F),
],
"sinhala": [
(0x0D80, 0x0DFF),
],
"gujarati": [
(0x0A80, 0x0AFF),
],
"gurmukhi": [
(0x0A00, 0x0A7F),
],
"thai": [
(0x0E00, 0x0E7F),
],
"lao": [
(0x0E80, 0x0EFF),
],
"burmese": [
(0x1000, 0x109F),
],
"khmer": [
(0x1780, 0x17FF),
],
"cjk": [
(0x4E00, 0x9FFF), # CJK Unified Ideographs
(0x3400, 0x4DBF), # CJK Extension A
(0x20000, 0x2A6DF), # CJK Extension B (rare)
],
"hiragana_katakana": [
(0x3040, 0x309F), # Hiragana
(0x30A0, 0x30FF), # Katakana
],
"hangul": [
(0xAC00, 0xD7AF), # Hangul Syllables
(0x1100, 0x11FF), # Hangul Jamo
(0xA960, 0xA97F), # Hangul Jamo Extended-A
],
"georgian": [
(0x10A0, 0x10FF),
],
"armenian": [
(0x0530, 0x058F),
],
"ethiopic": [
(0x1200, 0x137F), # Ethiopic
(0x1380, 0x139F), # Ethiopic Supplement
],
"tibetan": [
(0x0F00, 0x0FFF),
],
"thaana": [
(0x0780, 0x07BF), # Thaana (Dhivehi)
],
# Latin is the implicit fallback.
"latin": [],
}
# ---------- Language → script mapping ----------
# Single source of truth. Keys are lower-case ISO codes.
LANG_TO_SCRIPT: Dict[str, str] = {
# Latin-script languages (the long tail of European + colonial)
"en": "latin", "fr": "latin", "de": "latin", "es": "latin", "it": "latin",
"pt": "latin", "nl": "latin", "pl": "latin", "tr": "latin", "vi": "latin",
"id": "latin", "ms": "latin", "ro": "latin", "cs": "latin", "sv": "latin",
"da": "latin", "fi": "latin", "no": "latin", "nb": "latin", "nn": "latin",
"hu": "latin", "sk": "latin", "sl": "latin", "lt": "latin", "lv": "latin",
"et": "latin", "sq": "latin", "az": "latin", "uz": "latin", "kk": "latin",
"ky": "latin", "tk": "latin", "sw": "latin", "eu": "latin", "gl": "latin",
"is": "latin", "ga": "latin", "mt": "latin", "ca": "latin", "hr": "latin",
"bs": "latin", "sr": "latin", "mk": "latin", "bg": "latin", "be": "latin",
"uk": "latin", # NOTE: uk/be/sr/mk/bg are actually Cyrillic — see fix below
"af": "latin", "cy": "latin", "lb": "latin", "fo": "latin", "br": "latin",
"co": "latin", "fy": "latin", "gd": "latin", "gn": "latin", "gu": "latin",
"ht": "latin", "haw": "latin", "hmn": "latin", "jv": "latin", "ku": "latin",
"mg": "latin", "mi": "latin", "mn": "latin", "nso": "latin", "ny": "latin",
"oc": "latin", "os": "latin", "ps": "latin", "qu": "latin", "rw": "latin",
"sc": "latin", "si": "latin", "sm": "latin", "sn": "latin", "so": "latin",
"st": "latin", "su": "latin", "tg": "latin", "tt": "latin", "ty": "latin",
"ug": "latin", "vo": "latin", "wa": "latin", "wo": "latin", "xh": "latin",
"yi": "latin", "zu": "latin",
}
# Correct Cyrillic assignments (overrides above for the Cyrillic-script langs)
LANG_TO_SCRIPT.update({
"ru": "cyrillic", "uk": "cyrillic", "be": "cyrillic", "sr": "cyrillic",
"mk": "cyrillic", "bg": "cyrillic", "kk": "cyrillic", "ky": "cyrillic",
"tg": "cyrillic", "tt": "cyrillic", "mn": "cyrillic", "ab": "cyrillic",
"ba": "cyrillic", "ce": "cyrillic", "cv": "cyrillic", "kv": "cyrillic",
"kv": "cyrillic", "l1": "cyrillic", "mhr": "cyrillic", "mrj": "cyrillic",
"myv": "cyrillic", "os": "cyrillic", "rue": "cyrillic", "sah": "cyrillic",
"udm": "cyrillic", "uk": "cyrillic",
})
# More corrections
LANG_TO_SCRIPT.update({
"el": "greek",
"ar": "arabic", "fa": "arabic", "ur": "arabic", "ps": "arabic",
"ku": "arabic", "sd": "arabic", "ug": "arabic",
"ckb": "arabic", "bal": "arabic", "bqi": "arabic",
"glk": "arabic", "mzn": "arabic",
"he": "hebrew",
"yi": "hebrew", # Yiddish uses Hebrew script
"dv": "thaana", # Maldivian (Dhivehi) uses Thaana script
"hi": "devanagari", "ne": "devanagari", "mr": "devanagari", "sa": "devanagari",
"mai": "devanagari", "bho": "devanagari", "awa": "devanagari",
"bn": "bengali", "as": "bengali",
"ta": "tamil",
"te": "telugu",
"kn": "kannada",
"ml": "malayalam",
"si": "sinhala",
"gu": "gujarati",
"pa": "gurmukhi",
"th": "thai",
"lo": "lao",
"my": "burmese",
"km": "khmer",
"zh": "cjk", "zh-cn": "cjk", "zh-tw": "cjk", "zh-hk": "cjk",
"ja": "hiragana_katakana",
"ko": "hangul",
"ka": "georgian",
"hy": "armenian",
"am": "ethiopic", "ti": "ethiopic", "gez": "ethiopic", "tig": "ethiopic",
"bo": "tibetan", "dz": "tibetan",
})
# ---------- Discriminating characters for Arabic-script languages ----------
# These are characters that strongly suggest a SPECIFIC Arabic-script
# language as opposed to the generic "arabic" base.
DISCRIMINATING_CHARS: Dict[str, FrozenSet[str]] = {
"fa": frozenset("پچژگ"), # Persian
"ur": frozenset("ٹڈڑے"), # Urdu
"ps": frozenset("ټډړږښ"), # Pashto
"ku": frozenset("ڕێ"), # Kurdish
"ckb": frozenset("ڕێ"), # Sorani Kurdish (same as ku)
"sd": frozenset("ٿ"), # Sindhi
"ug": frozenset("ۇۆې"), # Uyghur
"yi": frozenset("ײ"), # Yiddish (uses Hebrew too)
"bal": frozenset("ێ"), # Balochi
}
# ---------- Thresholds ----------
# These are tunable but with sensible defaults.
MIN_RATIO_IN_SCRIPT = 0.60 # at least 60% of letters must match the script
MIN_RATIO_FOR_ARABIC_VARIANT = 0.20 # at least 20% of letters in the discriminating set
def get_script(lang_code: Optional[str]) -> str:
"""
Return the script id (e.g. 'cyrillic', 'cjk') for a given language code.
Returns 'latin' for unknown codes (which is the safe default — Latin
covers the largest number of languages).
"""
if not lang_code:
return "latin"
return LANG_TO_SCRIPT.get(lang_code.lower(), "latin")
def is_arabic_script_lang(lang_code: Optional[str]) -> bool:
"""True if lang_code is one of the Arabic-script languages we discriminate."""
if not lang_code:
return False
return get_script(lang_code) == "arabic"
def get_ranges(script_id: str) -> List[Tuple[int, int]]:
"""Return the Unicode ranges for a script id. Empty list for 'latin'."""
return UNICODE_RANGES.get(script_id, [])
def get_discriminating_chars(lang_code: Optional[str]) -> FrozenSet[str]:
"""Return the discriminating characters for a specific Arabic-script language."""
if not lang_code:
return frozenset()
return DISCRIMINATING_CHARS.get(lang_code.lower(), frozenset())

146
services/quality/file_extractor.py Normal file
View File

@@ -0,0 +1,146 @@
"""
File text extractor for the L0 quality layer.
Extracts a small sample of text from a translated file so the L0 checks
can run on a real output without requiring the translators to expose
their internal chunk data.
This module depends on the same libraries the translators use
(python-docx, openpyxl, python-pptx, PyMuPDF). All imports are lazy
and guarded, so a missing library only blocks the matching format —
other formats keep working.
"""
from __future__ import annotations
import zipfile
from pathlib import Path
from typing import List, Optional, TypedDict
class TextSample(TypedDict):
"""A (source_placeholder, translated) pair from an output file."""
source: str
translated: str
# Maximum samples per format — keeps the L0 check fast.
DEFAULT_MAX_SAMPLES = 20
def extract_sample(
file_path: Path,
file_extension: str,
max_samples: int = DEFAULT_MAX_SAMPLES,
) -> List[TextSample]:
"""
Extract a sample of translated text strings from a finished file.
The "source" field is always empty — we don't have the original
document at this point. The L0 checks that care about source/target
ratio (length_checker) handle empty source gracefully.
Returns an empty list if the file cannot be read.
"""
if not file_path or not Path(file_path).exists():
return []
ext = (file_extension or Path(file_path).suffix).lower()
try:
if ext == ".docx":
return _extract_docx(file_path, max_samples)
if ext == ".xlsx":
return _extract_xlsx(file_path, max_samples)
if ext == ".pptx":
return _extract_pptx(file_path, max_samples)
if ext == ".pdf":
return _extract_pdf(file_path, max_samples)
except Exception:
# Any failure: return an empty list. The route will log it.
return []
return []
def _extract_docx(path: Path, max_samples: int) -> List[TextSample]:
"""Extract text from a Word document."""
from docx import Document
doc = Document(str(path))
samples: List[TextSample] = []
for para in doc.paragraphs:
text = (para.text or "").strip()
if text and len(text) > 5:
samples.append({"source": "", "translated": text})
if len(samples) >= max_samples:
break
# If body had nothing, try tables.
if not samples:
for table in doc.tables:
for row in table.rows:
for cell in row.cells:
text = (cell.text or "").strip()
if text and len(text) > 5:
samples.append({"source": "", "translated": text})
if len(samples) >= max_samples:
return samples
return samples
def _extract_xlsx(path: Path, max_samples: int) -> List[TextSample]:
"""Extract text from an Excel file."""
from openpyxl import load_workbook
wb = load_workbook(str(path), data_only=True, read_only=True)
samples: List[TextSample] = []
try:
for ws in wb.worksheets:
for row in ws.iter_rows():
for cell in row:
val = cell.value
if isinstance(val, str):
text = val.strip()
if text and len(text) > 3:
samples.append({"source": "", "translated": text})
if len(samples) >= max_samples:
return samples
finally:
wb.close()
return samples
def _extract_pptx(path: Path, max_samples: int) -> List[TextSample]:
"""Extract text from a PowerPoint file."""
from pptx import Presentation
pres = Presentation(str(path))
samples: List[TextSample] = []
for slide in pres.slides:
for shape in slide.shapes:
if not shape.has_text_frame:
continue
for para in shape.text_frame.paragraphs:
text = (para.text or "").strip()
if text and len(text) > 3:
samples.append({"source": "", "translated": text})
if len(samples) >= max_samples:
return samples
return samples
def _extract_pdf(path: Path, max_samples: int) -> List[TextSample]:
"""Extract text from a PDF file (best-effort, layout-preserving format)."""
try:
import fitz # PyMuPDF
except ImportError:
return []
samples: List[TextSample] = []
doc = fitz.open(str(path))
try:
for page in doc:
text = page.get_text("text") or ""
for line in text.splitlines():
line = line.strip()
if line and len(line) > 5:
samples.append({"source": "", "translated": line})
if len(samples) >= max_samples:
return samples
finally:
doc.close()
return samples

125
services/quality/length_checker.py Normal file
View File

@@ -0,0 +1,125 @@
"""
Length sanity check for the L0 quality layer.
A translation that's 10× longer or 10× shorter than the source is almost
certainly a hallucination or a truncation. We flag these as warnings
(not failures) so the caller can decide what to do.
Thresholds are tunable via env vars if needed, but the defaults work
well for prose documents. Tables and bullet lists will naturally have
shorter translations, so we keep the lower bound loose.
Special cases:
* If the source is mostly digits/punctuation, the translation can also
be short (e.g. "Price: 100$" → "100€") — skip the check.
* If the source is empty/very short, skip the check entirely.
* If the source contains an embedded URL or email, the translation may
legitimately shrink — skip the check.
"""
from __future__ import annotations
import re
from typing import Dict
# Default thresholds — generous enough to handle tables / short strings.
RATIO_MAX = 3.5
RATIO_MIN = 0.15
# Hard lower bound: a translation shorter than this is very suspect,
# UNLESS the source is also very short.
ABSOLUTE_MIN_LENGTH = 2
# If source is short (under this many chars), skip the ratio check entirely
# — short strings are too noisy to be useful for length analysis.
MIN_SOURCE_LENGTH_FOR_RATIO = 20
# Pattern to detect text that is mostly digits / numbers / simple symbols.
# E.g. "Price: 100€", "+33 6 12 34 56 78", "192.168.1.1".
_MOSTLY_NUMERIC_RE = re.compile(r"^[\d\s\W]*$", re.UNICODE)
_NUMERIC_RATIO_THRESHOLD = 0.5 # 50% of letters are digits
def check(source_text: str, translated_text: str) -> Dict:
"""
Returns a dict like:
{
"issue": None | "length_outlier" | "truncation_suspect",
"ratio": float,
"source_length": int,
"translated_length": int,
}
Never raises.
"""
if not source_text:
return {
"issue": None,
"ratio": None,
"source_length": 0,
"translated_length": len(translated_text or ""),
}
src_len = len(source_text.strip())
trans_len = len(translated_text.strip())
# Empty translation — always suspect.
if trans_len == 0:
return {
"issue": "truncation_suspect",
"ratio": 0.0,
"source_length": src_len,
"translated_length": trans_len,
}
# If source is mostly digits/numbers, the translation can also be short
# (e.g. "Price: 100" → "100"). Don't flag length in this case.
if _is_mostly_numeric(source_text):
return {
"issue": None,
"ratio": None,
"source_length": src_len,
"translated_length": trans_len,
"note": "skipped: numeric source",
}
# If source is very short, skip the ratio check.
if src_len < MIN_SOURCE_LENGTH_FOR_RATIO:
return {
"issue": None,
"ratio": None,
"source_length": src_len,
"translated_length": trans_len,
}
ratio = trans_len / src_len
if ratio > RATIO_MAX:
return {
"issue": "length_outlier",
"ratio": round(ratio, 2),
"source_length": src_len,
"translated_length": trans_len,
}
if ratio < RATIO_MIN:
return {
"issue": "truncation_suspect",
"ratio": round(ratio, 2),
"source_length": src_len,
"translated_length": trans_len,
}
return {
"issue": None,
"ratio": round(ratio, 2),
"source_length": src_len,
"translated_length": trans_len,
}
def _is_mostly_numeric(text: str) -> bool:
"""True if at least 50% of non-whitespace characters are digits."""
if not text:
return False
chars = [c for c in text if not c.isspace()]
if not chars:
return False
digit_count = sum(1 for c in chars if c.isdigit())
return (digit_count / len(chars)) >= _NUMERIC_RATIO_THRESHOLD

126
services/quality/pattern_leak.py Normal file
View File

@@ -0,0 +1,126 @@
"""
Pattern leak detection for the L0 quality layer.
Detects common failure modes where the LLM translator:
1. Leaks parts of the system prompt (e.g. starts with "Translation:" or "Voici la traduction :")
2. Hallucinates by repeating the same word/phrase many times in a row
3. Returns a chain-of-thought / explanation instead of a translation
These checks are pure regex / counting — no model, no network.
"""
from __future__ import annotations
import re
from typing import Dict, List
# ---------- Prompt leak patterns ----------
# Phrases that strongly suggest the LLM leaked its prompt or a thought process
# into the output. We check only at the START of the translation (after
# stripping whitespace) to avoid false positives on legitimate text.
LEAK_PREFIX_PATTERNS: List[re.Pattern] = [
re.compile(r"^(translation|translated text|here is the translation|here'?s the translation)\s*[:：-]", re.IGNORECASE),
re.compile(r"^(voici (la |ma )?traduction|traduction\s*[:：-])\b", re.IGNORECASE),
re.compile(r"^(原文|译|翻译|译为|以下是)\s*[:：]?", re.UNICODE),
re.compile(r"^(sure,?\s+here'?s?\s+(the\s+)?translation|of course,?\s+here)", re.IGNORECASE),
re.compile(r"^(\*\*|__|\#)\s*translation", re.IGNORECASE),
re.compile(r"^translated from\s+\w+\s+to\s+\w+\s*[:：-]", re.IGNORECASE),
]
# ---------- Repetition detection ----------
# A "word" for the repetition check — uses Unicode word boundaries so
# it works on Chinese, Japanese, Korean, etc.
_WORD_RE = re.compile(r"\S+", re.UNICODE)
# Threshold: a word (or token) repeated 5+ times consecutively is almost
# always a hallucination. We allow up to 4 to handle legitimate text
# like "the the" (typo) without false positives.
REPETITION_THRESHOLD = 5
# Same character repeated 20+ times in a row is also a hallucination
# (catches cases like "xxxxxxxxxxx" or "==========").
CHAR_REPETITION_THRESHOLD = 20
def check(text: str) -> Dict:
"""
Returns a dict like:
{
"issue": None | "prompt_leak" | "repetition_hallucination",
"matched_pattern": "..." | None,
"repetition_count": int | None,
}
Never raises.
"""
if not text or not text.strip():
return {"issue": None, "matched_pattern": None, "repetition_count": None}
stripped = text.lstrip()
# 1. Prompt leak
for pat in LEAK_PREFIX_PATTERNS:
m = pat.match(stripped)
if m:
return {
"issue": "prompt_leak",
"matched_pattern": pat.pattern,
"repetition_count": None,
}
# 2. Token-level repetition
tokens = _WORD_RE.findall(stripped)
rep_count = _max_consecutive_repetition(tokens)
if rep_count >= REPETITION_THRESHOLD:
return {
"issue": "repetition_hallucination",
"matched_pattern": None,
"repetition_count": rep_count,
}
# 3. Character-level repetition (catches "xxxxxx" without spaces)
char_rep = _max_consecutive_char_repetition(stripped)
if char_rep >= CHAR_REPETITION_THRESHOLD:
return {
"issue": "repetition_hallucination",
"matched_pattern": None,
"repetition_count": char_rep,
}
return {"issue": None, "matched_pattern": None, "repetition_count": max(rep_count, char_rep) or None}
def _max_consecutive_repetition(tokens: List[str]) -> int:
"""Return the maximum number of times the same token appears consecutively."""
if not tokens:
return 0
# Normalize: lower-case + strip basic punctuation for comparison
norm = [t.lower().strip(".,!?;:\"'`()[]{}") for t in tokens]
max_run = 1
current_run = 1
for i in range(1, len(norm)):
if norm[i] and norm[i] == norm[i - 1]:
current_run += 1
if current_run > max_run:
max_run = current_run
else:
current_run = 1
return max_run
def _max_consecutive_char_repetition(text: str) -> int:
"""Return the maximum number of times the same character appears consecutively."""
if not text:
return 0
max_run = 1
current_run = 1
for i in range(1, len(text)):
if text[i] == text[i - 1] and not text[i].isspace():
current_run += 1
if current_run > max_run:
max_run = current_run
else:
current_run = 1
return max_run

75
services/quality/pipeline.py Normal file
View File

@@ -0,0 +1,75 @@
"""
Quality pipeline — defensive wrapper around the L0 checks.
The pipeline is the integration point for the route. It:
1. Catches all exceptions (L0 must NEVER break a translation job)
2. Adds timing
3. Emits a single structured log line per job
The actual checks live in `script_detector`, `length_checker`, `pattern_leak`.
This module is the orchestration / safety layer.
"""
from __future__ import annotations
import time
from typing import List, Optional
from core.logging import get_logger
from .script_detector import evaluate_document, DocumentQualityResult
logger = get_logger(__name__)
def run_l0_check(
source_chunks: List[str],
translated_chunks: List[str],
target_lang: Optional[str],
job_id: Optional[str] = None,
file_extension: Optional[str] = None,
) -> DocumentQualityResult:
"""
Run the L0 quality checks defensively. Never raises.
Returns an empty/neutral DocumentQualityResult on internal error
so the calling route can log and continue without affecting the
translation job outcome.
"""
start = time.time()
empty = DocumentQualityResult(
passed=True,
score=0.0,
chunk_count=0,
failed_chunk_count=0,
issues={"internal_error": 1},
)
try:
result = evaluate_document(source_chunks, translated_chunks, target_lang)
except Exception as e:
elapsed_ms = round((time.time() - start) * 1000, 2)
logger.warning(
"quality_l0_check_failed",
job_id=job_id,
file_extension=file_extension,
error=str(e)[:200],
error_type=type(e).__name__,
elapsed_ms=elapsed_ms,
)
return empty
elapsed_ms = round((time.time() - start) * 1000, 2)
logger.info(
"quality_l0_check",
job_id=job_id,
file_extension=file_extension,
target_lang=target_lang,
chunk_count=result.chunk_count,
failed_chunk_count=result.failed_chunk_count,
score=result.score,
passed=result.passed,
issues=result.issues,
elapsed_ms=elapsed_ms,
)
return result

374
services/quality/script_detector.py Normal file
View File

@@ -0,0 +1,374 @@
"""
L0 script detector.
Verifies that a translated string is actually written in the script expected
for the target language.
This is the first line of defense against the most common translation
failure mode: the LLM hallucinates text in the wrong language or wrong
script (e.g. user asks for Persian, model returns Arabic, or user asks
for Hindi, model returns Arabic). The check is purely heuristic — it
counts code points in the relevant Unicode ranges and compares to a
threshold.
Pure Python. No network calls. No new dependencies.
"""
from __future__ import annotations
from dataclasses import dataclass, field, asdict
from typing import Dict, List, Optional
from . import config as _config
from . import length_checker
from . import pattern_leak
from core.logging import get_logger
logger = get_logger(__name__)
# ---------- Result dataclasses ----------
@dataclass
class QualityCheckResult:
"""Result of evaluating a single (source, translation) pair."""
passed: bool
score: float # 0.0 to 1.0
issues: List[str] = field(default_factory=list)
detected_script: Optional[str] = None
expected_script: Optional[str] = None
details: Dict = field(default_factory=dict)
def to_log_dict(self) -> Dict:
return asdict(self)
@dataclass
class DocumentQualityResult:
"""Aggregated result for a list of (source, translation) pairs."""
passed: bool
score: float # mean score across chunks
chunk_count: int
failed_chunk_count: int
issues: Dict[str, int] = field(default_factory=dict) # issue -> count
samples: List[Dict] = field(default_factory=list) # a few example failures
def to_log_dict(self) -> Dict:
return asdict(self)
# ---------- Core helpers ----------
def _char_in_ranges(code_point: int, ranges: list) -> bool:
"""True if a code point falls in any of the (start, end) ranges."""
for start, end in ranges:
if start <= code_point <= end:
return True
return False
def _count_letters(text: str) -> int:
"""Count alphabetic characters (using Python's built-in isalpha)."""
return sum(1 for c in text if c.isalpha())
def _count_in_script(text: str, ranges: list) -> int:
"""Count how many alphabetic characters fall within the given Unicode ranges."""
if not ranges:
# 'latin' or unknown — treat all letters as matching.
return _count_letters(text)
return sum(
1 for c in text
if c.isalpha() and _char_in_ranges(ord(c), ranges)
)
# ---------- Arabic-script variant detection ----------
def detect_arabic_variant(
text: str,
claimed_lang: Optional[str],
) -> Dict:
"""
For text that is in the Arabic script block, check whether it matches
the specific variant the user asked for (Persian, Urdu, Pashto, etc.).
Returns a dict like:
{
"verdict": "pass" | "fail" | "skip",
"claimed_lang": "fa",
"detected_variants": ["fa"],
"reason": "...",
}
Detection logic:
1. If the text has < 60% Arabic-script letters overall, verdict = "skip"
(the script-detector will catch the mismatch).
2. If claimed_lang is NOT an Arabic-script language, verdict = "fail"
(this case should have been caught upstream — defensive double-check).
3. Scan the text for any discriminating character from any
Arabic-script language. If a discriminating character of a
DIFFERENT language is found, verdict = "fail".
4. Otherwise verdict = "pass".
"""
if not text or not text.strip():
return {"verdict": "skip", "claimed_lang": claimed_lang, "reason": "empty text"}
arabic_ranges = _config.get_ranges("arabic")
letters = _count_letters(text)
if letters == 0:
return {"verdict": "skip", "claimed_lang": claimed_lang, "reason": "no letters"}
in_arabic = _count_in_script(text, arabic_ranges)
arabic_ratio = in_arabic / letters
if arabic_ratio < _config.MIN_RATIO_IN_SCRIPT:
# Not really Arabic-script — let the main script_detector handle it.
return {
"verdict": "skip",
"claimed_lang": claimed_lang,
"arabic_ratio": round(arabic_ratio, 3),
"reason": "not in Arabic script",
}
if not _config.is_arabic_script_lang(claimed_lang):
# The translation IS in Arabic but the target wasn't Arabic.
# The main script_detector will fail on this; we just return skip.
return {
"verdict": "skip",
"claimed_lang": claimed_lang,
"arabic_ratio": round(arabic_ratio, 3),
"reason": "target is not an Arabic-script language",
}
# Now: text is Arabic-script AND target is Arabic-script. Check the variant.
detected = set()
for lang_code, chars in _config.DISCRIMINATING_CHARS.items():
if not chars:
continue
if any(c in chars for c in text):
detected.add(lang_code)
if detected and claimed_lang and claimed_lang.lower() not in detected:
return {
"verdict": "fail",
"claimed_lang": claimed_lang,
"detected_variants": sorted(detected),
"arabic_ratio": round(arabic_ratio, 3),
"reason": (
f"target={claimed_lang} but text contains characters typical of "
f"{', '.join(sorted(detected))}"
),
}
return {
"verdict": "pass",
"claimed_lang": claimed_lang,
"detected_variants": sorted(detected) if detected else [claimed_lang],
"arabic_ratio": round(arabic_ratio, 3),
"reason": "ok",
}
# ---------- Per-chunk evaluation ----------
def evaluate_chunk(
source_text: str,
translated_text: str,
target_lang: Optional[str],
) -> QualityCheckResult:
"""
Run the L0 checks on a single (source, translation) pair.
Returns a QualityCheckResult. The function is purely defensive — it
never raises; any internal error results in a "skip" result.
"""
if translated_text is None:
return QualityCheckResult(
passed=True, score=0.0, issues=["empty_translation"],
details={"reason": "translation is None"},
)
text = translated_text.strip()
if not text:
return QualityCheckResult(
passed=True, score=0.0, issues=["empty_translation"],
details={"reason": "translation is empty or whitespace-only"},
)
target_lang = (target_lang or "").lower() or None
issues: List[str] = []
details: Dict = {}
# --- Script detection ---
expected_script = _config.get_script(target_lang)
expected_ranges = _config.get_ranges(expected_script)
letters = _count_letters(text)
if letters == 0:
# No alphabetic characters — could be numbers, punctuation, or
# a single non-Latin symbol. Skip script check.
script_score = 1.0
detected_script = expected_script
details["script_check"] = "skipped: no alphabetic characters"
else:
# Always try to determine the ACTUAL script of the text — used for
# diagnostics and for catching language confusion when the target
# is Latin (e.g. user asks fr, we get Arabic text).
detected_script = _detect_actual_script(text)
in_expected = _count_in_script(text, expected_ranges)
script_score = in_expected / letters
details["script_score"] = round(script_score, 3)
details["letters_in_text"] = letters
details["letters_in_script"] = in_expected
details["detected_script"] = detected_script
details["expected_script"] = expected_script
details["min_ratio"] = _config.MIN_RATIO_IN_SCRIPT
# Two failure modes:
# 1. Target is a SPECIFIC non-Latin script (cyrillic, arabic, cjk...)
# and the text doesn't match it enough.
# 2. Target is Latin but the text is clearly in a SPECIFIC other
# script (cyrillic, arabic, devanagari, cjk...) — language
# confusion.
if expected_script != "latin" and expected_ranges:
# Specific non-Latin target.
if script_score < _config.MIN_RATIO_IN_SCRIPT:
issues.append("wrong_script")
details["reason"] = (
f"only {script_score:.0%} of letters match {expected_script} script; "
f"text appears to be in {detected_script}"
)
else:
# Latin target. If detected script is clearly non-Latin, fail.
if detected_script and detected_script != "latin" and detected_script != "unknown":
# Measure how confident we are that the text is non-Latin.
non_latin_ranges = _config.get_ranges(detected_script)
in_detected = _count_in_script(text, non_latin_ranges)
non_latin_confidence = in_detected / letters
if non_latin_confidence >= 0.7:
issues.append("wrong_script")
details["reason"] = (
f"target is Latin but {non_latin_confidence:.0%} of letters "
f"are in {detected_script} script — language confusion"
)
# --- Arabic-script variant detection ---
if _config.is_arabic_script_lang(target_lang):
variant_result = detect_arabic_variant(text, target_lang)
details["arabic_variant"] = variant_result
if variant_result["verdict"] == "fail":
issues.append("wrong_arabic_variant")
# --- Length sanity ---
length_result = length_checker.check(source_text, text)
details["length"] = length_result
if length_result.get("issue"):
issues.append(length_result["issue"])
# --- Pattern leak / repetition ---
leak_result = pattern_leak.check(text)
details["pattern_check"] = leak_result
if leak_result.get("issue"):
issues.append(leak_result["issue"])
# --- Aggregate ---
passed = len(issues) == 0
# Simple score: how many of the 3 main checks passed.
n_checks = 3
n_failed = sum(
1 for issue in issues if issue in (
"wrong_script", "wrong_arabic_variant",
"length_outlier", "truncation_suspect",
"prompt_leak", "repetition_hallucination",
)
)
score = max(0.0, 1.0 - (n_failed / n_checks))
return QualityCheckResult(
passed=passed,
score=round(score, 3),
issues=issues,
detected_script=detected_script,
expected_script=expected_script,
details=details,
)
def _detect_actual_script(text: str) -> str:
"""
Heuristically determine which script a string is in. Used only for
diagnostics — never for the verdict. Returns the first script (in
priority order) whose ratio exceeds the threshold.
"""
letters = _count_letters(text)
if letters == 0:
return "unknown"
# Priority order: more specific scripts first.
order = [
"hiragana_katakana", "hangul", "cjk", "thai", "lao", "burmese",
"khmer", "devanagari", "bengali", "tamil", "telugu", "kannada",
"malayalam", "sinhala", "gujarati", "gurmukhi",
"arabic", "hebrew", "cyrillic", "greek", "armenian", "georgian",
"ethiopic", "tibetan", "thaana",
]
for script_id in order:
ranges = _config.get_ranges(script_id)
in_script = _count_in_script(text, ranges)
if in_script / letters > 0.4:
return script_id
return "latin"
# ---------- Document-level aggregation ----------
def evaluate_document(
source_chunks: List[str],
translated_chunks: List[str],
target_lang: Optional[str],
sample_size: int = 50,
) -> DocumentQualityResult:
"""
Evaluate all (source, translation) pairs and return a document-level
summary. The full list is processed but only the first `sample_size`
failing chunks are kept in `samples` to keep logs compact.
"""
n = min(len(source_chunks), len(translated_chunks))
chunk_results: List[QualityCheckResult] = []
issues_count: Dict[str, int] = {}
samples: List[Dict] = []
score_sum = 0.0
failed_count = 0
for i in range(n):
r = evaluate_chunk(source_chunks[i], translated_chunks[i], target_lang)
chunk_results.append(r)
score_sum += r.score
for issue in r.issues:
issues_count[issue] = issues_count.get(issue, 0) + 1
if not r.passed:
failed_count += 1
if len(samples) < sample_size:
src_preview = (source_chunks[i] or "")[:80]
trans_preview = (translated_chunks[i] or "")[:80]
samples.append({
"index": i,
"issues": r.issues,
"source_preview": src_preview,
"translated_preview": trans_preview,
"details": r.details,
})
mean_score = (score_sum / n) if n > 0 else 0.0
passed = failed_count == 0
return DocumentQualityResult(
passed=passed,
score=round(mean_score, 3),
chunk_count=n,
failed_chunk_count=failed_count,
issues=issues_count,
samples=samples,
)

0
tests/services/quality/__init__.py Normal file
View File

143
tests/services/quality/test_config.py Normal file
View File

@@ -0,0 +1,143 @@
"""
Tests for services/quality/config.py
Covers the language → script mapping and discriminating characters.
"""
import pytest
from services.quality import config as qconfig
class TestGetScript:
def test_cyrillic_languages(self):
assert qconfig.get_script("ru") == "cyrillic"
assert qconfig.get_script("uk") == "cyrillic"
assert qconfig.get_script("be") == "cyrillic"
assert qconfig.get_script("bg") == "cyrillic"
assert qconfig.get_script("sr") == "cyrillic"
assert qconfig.get_script("kk") == "cyrillic"
def test_latin_languages(self):
assert qconfig.get_script("en") == "latin"
assert qconfig.get_script("fr") == "latin"
assert qconfig.get_script("de") == "latin"
assert qconfig.get_script("es") == "latin"
assert qconfig.get_script("vi") == "latin"
assert qconfig.get_script("tr") == "latin"
def test_cjk_languages(self):
assert qconfig.get_script("zh") == "cjk"
assert qconfig.get_script("zh-cn") == "cjk"
assert qconfig.get_script("zh-tw") == "cjk"
def test_japanese_uses_kana(self):
# ja is mapped to hiragana_katakana specifically so it can be
# distinguished from Chinese CJK.
assert qconfig.get_script("ja") == "hiragana_katakana"
def test_korean_uses_hangul(self):
assert qconfig.get_script("ko") == "hangul"
def test_arabic_script_languages(self):
assert qconfig.get_script("ar") == "arabic"
assert qconfig.get_script("fa") == "arabic"
assert qconfig.get_script("ur") == "arabic"
assert qconfig.get_script("ps") == "arabic"
assert qconfig.get_script("ku") == "arabic"
def test_hebrew_and_yiddish(self):
assert qconfig.get_script("he") == "hebrew"
# Yiddish uses Hebrew script, not Arabic
assert qconfig.get_script("yi") == "hebrew"
def test_thaana(self):
# Maldivian (Dhivehi) uses Thaana script, not Arabic
assert qconfig.get_script("dv") == "thaana"
def test_indian_scripts(self):
assert qconfig.get_script("hi") == "devanagari"
assert qconfig.get_script("bn") == "bengali"
assert qconfig.get_script("ta") == "tamil"
assert qconfig.get_script("te") == "telugu"
assert qconfig.get_script("gu") == "gujarati"
assert qconfig.get_script("pa") == "gurmukhi"
assert qconfig.get_script("ml") == "malayalam"
assert qconfig.get_script("kn") == "kannada"
assert qconfig.get_script("si") == "sinhala"
def test_se_asian_scripts(self):
assert qconfig.get_script("th") == "thai"
assert qconfig.get_script("lo") == "lao"
assert qconfig.get_script("my") == "burmese"
assert qconfig.get_script("km") == "khmer"
def test_other_scripts(self):
assert qconfig.get_script("el") == "greek"
assert qconfig.get_script("ka") == "georgian"
assert qconfig.get_script("hy") == "armenian"
assert qconfig.get_script("am") == "ethiopic"
assert qconfig.get_script("bo") == "tibetan"
def test_unknown_falls_back_to_latin(self):
assert qconfig.get_script("xx") == "latin"
assert qconfig.get_script("") == "latin"
assert qconfig.get_script(None) == "latin"
def test_case_insensitive(self):
assert qconfig.get_script("FR") == "latin"
assert qconfig.get_script("ZH-CN") == "cjk"
assert qconfig.get_script("Ru") == "cyrillic"
class TestIsArabicScriptLang:
def test_true_for_arabic_script_langs(self):
assert qconfig.is_arabic_script_lang("ar") is True
assert qconfig.is_arabic_script_lang("fa") is True
assert qconfig.is_arabic_script_lang("ur") is True
assert qconfig.is_arabic_script_lang("ckb") is True
def test_false_for_non_arabic(self):
assert qconfig.is_arabic_script_lang("fr") is False
assert qconfig.is_arabic_script_lang("he") is False
assert qconfig.is_arabic_script_lang("hi") is False
assert qconfig.is_arabic_script_lang("en") is False
def test_false_for_unknown(self):
assert qconfig.is_arabic_script_lang("xx") is False
assert qconfig.is_arabic_script_lang("") is False
assert qconfig.is_arabic_script_lang(None) is False
class TestDiscriminatingChars:
def test_persian_chars(self):
chars = qconfig.get_discriminating_chars("fa")
assert "پ" in chars # peh
assert "چ" in chars # tcheh
assert "ژ" in chars # zheh
assert "گ" in chars # guaf
def test_urdu_chars(self):
chars = qconfig.get_discriminating_chars("ur")
assert "ٹ" in chars # tteh
assert "ڈ" in chars # ddal
def test_unknown_lang_empty(self):
assert qconfig.get_discriminating_chars("fr") == frozenset()
assert qconfig.get_discriminating_chars("") == frozenset()
assert qconfig.get_discriminating_chars(None) == frozenset()
class TestGetRanges:
def test_latin_returns_empty(self):
# Latin is the fallback — no ranges, means "anything matches".
assert qconfig.get_ranges("latin") == []
def test_cyrillic_ranges(self):
ranges = qconfig.get_ranges("cyrillic")
assert any(start == 0x0400 for start, _ in ranges)
def test_cjk_ranges(self):
ranges = qconfig.get_ranges("cjk")
assert any(start == 0x4E00 for start, _ in ranges)
def test_unknown_script_returns_empty(self):
assert qconfig.get_ranges("mystery_script") == []

181
tests/services/quality/test_file_extractor.py Normal file
View File

@@ -0,0 +1,181 @@
"""
Tests for services/quality/file_extractor.py
Uses real files generated via temporary paths.
On Windows, tempfile.NamedTemporaryFile holds the file open and blocks
overwrite, so we use a plain temp directory + manual filename instead.
"""
import tempfile
import os
from pathlib import Path
import pytest
from services.quality.file_extractor import (
extract_sample,
DEFAULT_MAX_SAMPLES,
)
class TestExtractSample:
def test_returns_empty_for_missing_file(self):
result = extract_sample(Path("/nonexistent/path/file.docx"), ".docx")
assert result == []
def test_returns_empty_for_none_path(self):
result = extract_sample(None, ".docx")
assert result == []
def test_returns_empty_for_unsupported_extension(self):
# .txt isn't supported — should return [] silently
with tempfile.TemporaryDirectory() as d:
p = Path(d) / "test.txt"
p.write_text("some text")
result = extract_sample(p, ".txt")
assert result == []
def _make_tmp_path(suffix: str) -> Path:
"""Create a unique temp file path. File does not exist yet."""
fd, name = tempfile.mkstemp(suffix=suffix)
os.close(fd)
return Path(name)
class TestDocxExtraction:
def test_extracts_paragraphs(self):
from docx import Document
doc = Document()
doc.add_paragraph("Bonjour le monde")
doc.add_paragraph("Comment allez-vous?")
doc.add_paragraph("Merci beaucoup")
p = _make_tmp_path(".docx")
try:
doc.save(str(p))
result = extract_sample(p, ".docx", max_samples=10)
assert len(result) >= 1
texts = [s["translated"] for s in result]
assert "Bonjour le monde" in texts
finally:
try:
p.unlink()
except (OSError, PermissionError):
pass
def test_respects_max_samples(self):
from docx import Document
doc = Document()
for i in range(50):
doc.add_paragraph(f"Paragraphe numero {i} avec du texte")
p = _make_tmp_path(".docx")
try:
doc.save(str(p))
result = extract_sample(p, ".docx", max_samples=5)
assert len(result) == 5
finally:
try:
p.unlink()
except (OSError, PermissionError):
pass
def test_handles_empty_doc(self):
from docx import Document
doc = Document()
p = _make_tmp_path(".docx")
try:
doc.save(str(p))
result = extract_sample(p, ".docx")
assert result == []
finally:
try:
p.unlink()
except (OSError, PermissionError):
pass
class TestXlsxExtraction:
def test_extracts_cells(self):
from openpyxl import Workbook
wb = Workbook()
ws = wb.active
ws["A1"] = "Bonjour"
ws["A2"] = "Monde"
ws["A3"] = "Comment"
p = _make_tmp_path(".xlsx")
try:
wb.save(str(p))
wb.close()
result = extract_sample(p, ".xlsx", max_samples=10)
assert len(result) >= 1
texts = [s["translated"] for s in result]
assert "Bonjour" in texts
finally:
try:
p.unlink()
except (OSError, PermissionError):
pass
def test_skips_numeric_cells(self):
from openpyxl import Workbook
wb = Workbook()
ws = wb.active
ws["A1"] = 100
ws["A2"] = 200
ws["A3"] = "Hello"
p = _make_tmp_path(".xlsx")
try:
wb.save(str(p))
wb.close()
result = extract_sample(p, ".xlsx")
texts = [s["translated"] for s in result]
assert "Hello" in texts
assert 100 not in texts # numeric only cells skipped
finally:
try:
p.unlink()
except (OSError, PermissionError):
pass
class TestPptxExtraction:
def test_extracts_slide_text(self):
from pptx import Presentation
pres = Presentation()
slide = pres.slides.add_slide(pres.slide_layouts[0])
slide.shapes.title.text = "Bonjour le monde"
p = _make_tmp_path(".pptx")
try:
pres.save(str(p))
result = extract_sample(p, ".pptx")
assert len(result) >= 1
assert result[0]["translated"] == "Bonjour le monde"
finally:
try:
p.unlink()
except (OSError, PermissionError):
pass
class TestPdfExtraction:
def test_extracts_pdf_text(self):
# Create a minimal PDF using reportlab if available, else skip
pytest.importorskip("fitz")
try:
from reportlab.pdfgen import canvas
except ImportError:
pytest.skip("reportlab not available")
p = _make_tmp_path(".pdf")
try:
c = canvas.Canvas(str(p))
c.drawString(100, 750, "Bonjour le monde")
c.drawString(100, 700, "Comment allez vous")
c.save()
result = extract_sample(p, ".pdf")
assert len(result) >= 1
texts = [s["translated"] for s in result]
assert any("Bonjour" in t for t in texts)
finally:
try:
p.unlink()
except (OSError, PermissionError):
pass

49
tests/services/quality/test_length_checker.py Normal file
View File

@@ -0,0 +1,49 @@
"""
Tests for services/quality/length_checker.py
"""
import pytest
from services.quality.length_checker import check
class TestLengthCheck:
def test_normal_length(self):
r = check("Hello world, this is a test of translation length",
"Bonjour le monde, ceci est un test de longueur de traduction")
assert r["issue"] is None
assert r["ratio"] is not None
assert 0.5 < r["ratio"] < 2.0
def test_huge_translation(self):
src = "A" * 200
r = check(src, "x" * 1000)
assert r["issue"] == "length_outlier"
assert r["ratio"] > 3.5
def test_tiny_translation(self):
r = check("A" * 100, "ok")
assert r["issue"] == "truncation_suspect"
assert r["ratio"] < 0.15
def test_empty_translation_flagged(self):
r = check("Hello world, this is a test", "")
assert r["issue"] == "truncation_suspect"
def test_short_source_skips_ratio(self):
r = check("Hi", "ok")
assert r["issue"] is None
assert r["ratio"] is None
def test_empty_inputs(self):
r = check("", "")
assert r["issue"] is None
assert r["source_length"] == 0
assert r["translated_length"] == 0
def test_none_source(self):
r = check(None, "translation")
assert r["issue"] is None
def test_numeric_source(self):
r = check("Price: 100", "100")
assert r["issue"] is None # numeric sources skip the check

85
tests/services/quality/test_pattern_leak.py Normal file
View File

@@ -0,0 +1,85 @@
"""
Tests for services/quality/pattern_leak.py
"""
import pytest
from services.quality.pattern_leak import check
class TestPromptLeak:
def test_english_translation_prefix(self):
r = check("Translation: Bonjour le monde")
assert r["issue"] == "prompt_leak"
def test_here_is_translation(self):
r = check("Here is the translation: Bonjour")
assert r["issue"] == "prompt_leak"
def test_french_voici_traduction(self):
r = check("Voici la traduction : Bonjour le monde")
assert r["issue"] == "prompt_leak"
def test_chinese_translation_prefix(self):
r = check("翻译：你好世界")
assert r["issue"] == "prompt_leak"
def test_sure_heres_translation(self):
r = check("Sure, here's the translation: Hello world")
assert r["issue"] == "prompt_leak"
def test_of_course_heres(self):
r = check("Of course, here you go: Bonjour")
assert r["issue"] == "prompt_leak"
def test_markdown_bold_translation(self):
r = check("**Translation** Bonjour le monde")
assert r["issue"] == "prompt_leak"
def test_normal_text_passes(self):
r = check("Bonjour le monde, comment allez-vous?")
assert r["issue"] is None
def test_text_with_translation_word_in_middle_passes(self):
r = check("Voici ce que je pense de la traduction de ce texte")
# The word "traduction" appears but not as a prefix
assert r["issue"] is None
class TestRepetitionHallucination:
def test_long_repetition_detected(self):
r = check("the the the the the the the the")
assert r["issue"] == "repetition_hallucination"
assert r["repetition_count"] >= 5
def test_short_repetition_passes(self):
# 4 times is within tolerance
r = check("I think I think I think I think")
assert r["issue"] is None
def test_normal_text_passes(self):
r = check("This is a normal sentence with no repetition issues")
assert r["issue"] is None
def test_mixed_repetition_in_long_text_passes(self):
# Repetition is local, not a hallucination
r = check("The cat sat on the mat. The cat was happy. The dog was sad.")
assert r["issue"] is None
def test_repetition_with_punctuation(self):
# "the, the, the, the, the" should still be detected
r = check("the, the, the, the, the, the")
assert r["issue"] == "repetition_hallucination"
class TestEmptyAndEdgeCases:
def test_empty_text(self):
r = check("")
assert r["issue"] is None
def test_whitespace_only(self):
r = check(" \n \t ")
assert r["issue"] is None
def test_none_input(self):
r = check(None)
assert r["issue"] is None

61
tests/services/quality/test_pipeline.py Normal file
View File

@@ -0,0 +1,61 @@
"""
Tests for services/quality/pipeline.py
"""
import pytest
from services.quality import run_l0_check
from services.quality.script_detector import DocumentQualityResult
class TestRunL0Check:
def test_returns_result_on_success(self):
result = run_l0_check(
["Hello", "World"],
["Bonjour", "Monde"],
"fr",
job_id="test_job_1",
)
assert isinstance(result, DocumentQualityResult)
assert result.passed is True
assert result.chunk_count == 2
def test_returns_neutral_on_empty_lists(self):
result = run_l0_check([], [], "fr", job_id="test_job_2")
assert result.passed is True
assert result.chunk_count == 0
def test_detects_failures(self):
result = run_l0_check(
["Hello", "World"],
["Bonjour", "مرحبا"],
"fr",
job_id="test_job_3",
)
assert result.passed is False
assert "wrong_script" in result.issues
def test_never_raises_on_bad_input(self):
# Even with weird input, it shouldn't raise
result = run_l0_check(
["Hello"],
[None], # None translation
"fr",
job_id="test_job_4",
)
assert result is not None
def test_optional_job_id(self):
# job_id is optional
result = run_l0_check(["hi"], ["salut"], "fr")
assert result is not None
def test_optional_file_extension(self):
result = run_l0_check(
["hi"], ["salut"], "fr", file_extension=".docx"
)
assert result is not None
def test_target_lang_none(self):
# Should not crash with no target lang
result = run_l0_check(["hi"], ["salut"], None)
assert result is not None

290
tests/services/quality/test_script_detector.py Normal file
View File

@@ -0,0 +1,290 @@
"""
Tests for services/quality/script_detector.py
"""
import pytest
from services.quality import evaluate_chunk, evaluate_document, detect_arabic_variant
# ---------- Happy path: correct script ----------
class TestCorrectScript:
def test_russian_correct(self):
r = evaluate_chunk("Hello world", "Привет мир", "ru")
assert r.passed is True
assert "wrong_script" not in r.issues
def test_chinese_simplified_correct(self):
r = evaluate_chunk("Hello", "你好世界", "zh")
assert r.passed is True
def test_arabic_correct(self):
r = evaluate_chunk("Hello", "مرحبا بالعالم", "ar")
assert r.passed is True
def test_persian_correct(self):
# Persian has unique chars چ پ ژ گ
r = evaluate_chunk("Hello", "سلام چطوری؟ من پژوهشگر هستم", "fa")
assert r.passed is True
def test_french_correct(self):
r = evaluate_chunk("Hello world", "Bonjour le monde", "fr")
assert r.passed is True
def test_hebrew_correct(self):
r = evaluate_chunk("Hello", "שלום עולם", "he")
assert r.passed is True
def test_korean_correct(self):
r = evaluate_chunk("Hello", "안녕하세요 세계", "ko")
assert r.passed is True
def test_japanese_correct(self):
# Japanese must contain hiragana or katakana to be classified as ja.
r = evaluate_chunk("Hello", "こんにちは世界", "ja")
assert r.passed is True
def test_hindi_correct(self):
r = evaluate_chunk("Hello", "नमस्ते दुनिया", "hi")
assert r.passed is True
def test_thai_correct(self):
r = evaluate_chunk("Hello", "สวัสดีชาวโลก", "th")
assert r.passed is True
def test_greek_correct(self):
r = evaluate_chunk("Hello", "Γεια σας κόσμε", "el")
assert r.passed is True
# ---------- Wrong script: language confusion ----------
class TestWrongScript:
def test_french_text_for_japanese_target(self):
r = evaluate_chunk("Hello", "Bonjour le monde", "ja")
assert r.passed is False
assert "wrong_script" in r.issues
def test_arabic_text_for_french_target(self):
r = evaluate_chunk("Hello", "مرحبا بالعالم", "fr")
assert r.passed is False
assert "wrong_script" in r.issues
def test_russian_text_for_english_target(self):
r = evaluate_chunk("Hello", "Привет мир", "en")
assert r.passed is False
assert "wrong_script" in r.issues
def test_chinese_text_for_korean_target(self):
r = evaluate_chunk("Hello", "你好世界", "ko")
# Hangul syllables are 0xAC00-0xD7AF — Chinese chars are not in that range.
assert r.passed is False
assert "wrong_script" in r.issues
# ---------- Persian / Arabic discrimination ----------
class TestArabicVariantDiscrimination:
def test_persian_for_arabic_target_fails(self):
# Persian text with چ پ ژ گ should fail when target is Arabic.
r = evaluate_chunk("Hello", "سلام چطوری؟", "ar")
assert r.passed is False
assert "wrong_arabic_variant" in r.issues
def test_arabic_for_persian_target_fails(self):
# Pure Arabic text (no Persian-specific chars) when target is Persian.
r = evaluate_chunk("Hello", "السلام عليكم", "fa")
# No Persian-specific chars, no other Arabic-script variant chars
# → should pass the variant check (because no discrimination signal).
# This is a known limitation: pure Arabic indistinguishable from pure Persian.
# We accept that the variant check only flags MISMATCHES with discriminating chars.
assert "wrong_arabic_variant" not in r.issues
def test_urdu_for_persian_target_fails(self):
# Urdu with ٹ ڈ should fail when target is Persian.
r = evaluate_chunk("Hello", "السلام ٹڈ", "fa")
assert r.passed is False
assert "wrong_arabic_variant" in r.issues
def test_pashto_for_arabic_target_fails(self):
# Pashto with ټډړ should fail when target is Arabic.
r = evaluate_chunk("Hello", "السلام ټډړ", "ar")
assert r.passed is False
assert "wrong_arabic_variant" in r.issues
# ---------- Edge cases ----------
class TestEdgeCases:
def test_empty_translation_passes(self):
# Empty translations are skipped (the provider may have legitimately
# produced nothing — e.g. an empty cell).
r = evaluate_chunk("Hello", "", "fr")
assert r.passed is True
assert "empty_translation" in r.issues
def test_whitespace_only_passes(self):
r = evaluate_chunk("Hello", " \n ", "fr")
assert r.passed is True
assert "empty_translation" in r.issues
def test_none_translation_passes(self):
r = evaluate_chunk("Hello", None, "fr")
assert r.passed is True
def test_numbers_only_passes(self):
r = evaluate_chunk("Price: 100", "100", "fr")
assert r.passed is True
def test_unknown_target_lang_falls_back(self):
# Unknown lang → latin → no script check, just length/leak checks.
r = evaluate_chunk("Hello", "Bonjour", "xx")
assert r.passed is True
def test_mixed_brands_latin_target(self):
# "iPhone 15 Pro Max" is mostly Latin + digits — fine for fr.
r = evaluate_chunk("Buy iPhone 15", "Acheter iPhone 15", "fr")
assert r.passed is True
# ---------- Length sanity ----------
class TestLengthIssues:
def test_huge_translation_flagged_via_repetition(self):
# Source too short for length ratio, but the "x"*1000 repetition
# is caught by the pattern/repetition check, not length.
r = evaluate_chunk("Short", "x" * 1000, "fr")
assert "repetition_hallucination" in r.issues
def test_huge_translation_with_long_source_flagged(self):
# Long enough source → length ratio kicks in → length_outlier.
src = "A" * 200
r = evaluate_chunk(src, "x" * 1000, "fr")
assert "length_outlier" in r.issues
def test_tiny_translation_flagged(self):
r = evaluate_chunk("A" * 100, "ok", "fr")
# ratio is 2/100 = 0.02, well below 0.15
assert "truncation_suspect" in r.issues
def test_normal_translation_no_length_issue(self):
r = evaluate_chunk("Hello world, how are you?", "Bonjour le monde, comment allez-vous?", "fr")
assert "length_outlier" not in r.issues
assert "truncation_suspect" not in r.issues
def test_short_source_skips_ratio(self):
r = evaluate_chunk("Hi", "ok", "fr")
# source too short for ratio check
assert "length_outlier" not in r.issues
assert "truncation_suspect" not in r.issues
def test_numeric_source_skips_length_check(self):
# Source is mostly digits — translation can also be short.
r = evaluate_chunk("Price: 100", "100", "fr")
assert "truncation_suspect" not in r.issues
assert "length_outlier" not in r.issues
# ---------- Pattern leak / repetition ----------
class TestPatternLeak:
def test_prompt_leak_english_detected(self):
r = evaluate_chunk("Hello", "Translation: Bonjour le monde", "fr")
assert "prompt_leak" in r.issues
def test_prompt_leak_french_detected(self):
r = evaluate_chunk("Hello", "Voici la traduction : Bonjour", "fr")
assert "prompt_leak" in r.issues
def test_prompt_leak_chinese_detected(self):
r = evaluate_chunk("Hello", "翻译：你好", "zh")
assert "prompt_leak" in r.issues
def test_repetition_hallucination_detected(self):
r = evaluate_chunk("Hello", "the the the the the the", "fr")
assert "repetition_hallucination" in r.issues
def test_normal_text_no_leak(self):
r = evaluate_chunk("Hello", "Bonjour le monde", "fr")
assert "prompt_leak" not in r.issues
assert "repetition_hallucination" not in r.issues
# ---------- Document-level aggregation ----------
class TestEvaluateDocument:
def test_all_good(self):
result = evaluate_document(
["Hello", "World", "Good morning"],
["Bonjour", "Monde", "Bonjour"],
"fr",
)
assert result.passed is True
assert result.failed_chunk_count == 0
assert result.chunk_count == 3
def test_some_failures(self):
result = evaluate_document(
["Hello", "World", "Good morning"],
["Bonjour", "مرحبا", "Bonjour"],
"fr",
)
assert result.passed is False
assert result.failed_chunk_count == 1
assert "wrong_script" in result.issues
assert len(result.samples) == 1
def test_empty_lists(self):
result = evaluate_document([], [], "fr")
assert result.passed is True
assert result.chunk_count == 0
assert result.score == 0.0
def test_sample_size_caps_samples(self):
# 10 failures, sample_size=3
result = evaluate_document(
["hi"] * 10,
["مرحبا"] * 10,
"fr",
sample_size=3,
)
assert result.failed_chunk_count == 10
assert len(result.samples) == 3
def test_score_proportional(self):
# 4 chunks, 1 fails → score should be ~0.75 (3/4 passed checks)
result = evaluate_document(
["hi", "hi", "hi", "hi"],
["bonjour", "bonjour", "مرحبا", "bonjour"],
"fr",
)
assert result.failed_chunk_count == 1
assert result.score < 1.0
assert result.score > 0.0
# ---------- detect_arabic_variant direct ----------
class TestDetectArabicVariantDirect:
def test_empty_text(self):
r = detect_arabic_variant("", "fa")
assert r["verdict"] == "skip"
def test_pure_persian_for_persian(self):
r = detect_arabic_variant("سلام چطوری", "fa")
assert r["verdict"] == "pass"
def test_pure_persian_for_arabic(self):
r = detect_arabic_variant("سلام چطوری", "ar")
assert r["verdict"] == "fail"
assert "fa" in r["detected_variants"]
def test_pure_urdu_for_persian(self):
r = detect_arabic_variant("السلام ٹڈ", "fa")
assert r["verdict"] == "fail"
assert "ur" in r["detected_variants"]
def test_non_arabic_text(self):
r = detect_arabic_variant("Bonjour le monde", "fa")
# Not in Arabic script → skip
assert r["verdict"] == "skip"

3
wordly.art---traduction-de-documents/package.json
View File

@@ -8,7 +8,8 @@
"build": "vite build", "build": "vite build",
"preview": "vite preview", "preview": "vite preview",
"clean": "rm -rf dist server.js", "clean": "rm -rf dist server.js",
"lint": "tsc --noEmit" "lint": "tsc --noEmit",
"test:quality": "node --experimental-strip-types --test tests/utils/scriptDetector.test.ts"
}, },
"dependencies": { "dependencies": {
"@google/genai": "^1.29.0", "@google/genai": "^1.29.0",

739
wordly.art---traduction-de-documents/src/utils/scriptDetector.ts Normal file
View File

@@ -0,0 +1,739 @@
/**
* L0 quality detector — TypeScript mirror of services/quality/ in Python.
*
* Detects:
* - wrong_script: translation is in the wrong script for the target language
* (e.g. user asks French, model returns Arabic — language confusion)
* - wrong_arabic_variant: for Arabic-script targets (ar/fa/ur/...),
* the translation uses characters of a DIFFERENT Arabic-script language
* (e.g. user asks Persian, model returns Urdu)
* - length_outlier / truncation_suspect: translation is wildly different
* in length from the source
* - prompt_leak: translation starts with "Translation:" or similar
* - repetition_hallucination: "xxx..." or "the the the the" pattern
*
* Zero dependencies. Works in browser AND in Node.js (for tests).
* Designed to be ADDITIVE — never blocks the translation flow.
*/
// ---------- Types ----------
export interface QualityCheckResult {
passed: boolean;
score: number; // 0.0 to 1.0
issues: string[];
detectedScript: string | null;
expectedScript: string | null;
details: Record<string, unknown>;
}
export interface DocumentQualityResult {
passed: boolean;
score: number;
chunkCount: number;
failedChunkCount: number;
issues: Record<string, number>;
samples: Array<{
index: number;
issues: string[];
sourcePreview: string;
translatedPreview: string;
details: Record<string, unknown>;
}>;
}
// ---------- Unicode ranges per script ----------
// Mirrors services/quality/config.py in Python.
type Range = readonly [number, number]; // [start, end] inclusive
const UNICODE_RANGES: Record<string, readonly Range[]> = {
cyrillic: [
[0x0400, 0x04ff],
[0x0500, 0x052f],
],
greek: [
[0x0370, 0x03ff],
],
arabic: [
[0x0600, 0x06ff],
[0x0750, 0x077f],
[0x08a0, 0x08ff],
],
hebrew: [
[0x0590, 0x05ff],
],
devanagari: [
[0x0900, 0x097f],
],
bengali: [
[0x0980, 0x09ff],
],
tamil: [
[0x0b80, 0x0bff],
],
telugu: [
[0x0c00, 0x0c7f],
],
kannada: [
[0x0c80, 0x0cff],
],
malayalam: [
[0x0d00, 0x0d7f],
],
sinhala: [
[0x0d80, 0x0dff],
],
gujarati: [
[0x0a80, 0x0aff],
],
gurmukhi: [
[0x0a00, 0x0a7f],
],
thai: [
[0x0e00, 0x0e7f],
],
lao: [
[0x0e80, 0x0eff],
],
burmese: [
[0x1000, 0x109f],
],
khmer: [
[0x1780, 0x17ff],
],
cjk: [
[0x4e00, 0x9fff],
[0x3400, 0x4dbf],
],
hiragana_katakana: [
[0x3040, 0x309f],
[0x30a0, 0x30ff],
],
hangul: [
[0xac00, 0xd7af],
[0x1100, 0x11ff],
[0xa960, 0xa97f],
],
georgian: [
[0x10a0, 0x10ff],
],
armenian: [
[0x0530, 0x058f],
],
ethiopic: [
[0x1200, 0x137f],
[0x1380, 0x139f],
],
tibetan: [
[0x0f00, 0x0fff],
],
thaana: [
[0x0780, 0x07bf],
],
latin: [],
};
// ---------- Language → script mapping ----------
const LANG_TO_SCRIPT: Record<string, string> = {
// Latin-script languages
en: 'latin', fr: 'latin', de: 'latin', es: 'latin', it: 'latin',
pt: 'latin', nl: 'latin', pl: 'latin', tr: 'latin', vi: 'latin',
id: 'latin', ms: 'latin', ro: 'latin', cs: 'latin', sv: 'latin',
da: 'latin', fi: 'latin', no: 'latin', nb: 'latin', nn: 'latin',
hu: 'latin', sk: 'latin', sl: 'latin', lt: 'latin', lv: 'latin',
et: 'latin', sq: 'latin', az: 'latin', uz: 'latin', kk: 'latin',
ky: 'latin', tk: 'latin', sw: 'latin', eu: 'latin', gl: 'latin',
is: 'latin', ga: 'latin', mt: 'latin', ca: 'latin', hr: 'latin',
bs: 'latin', af: 'latin', cy: 'latin', lb: 'latin', fo: 'latin',
br: 'latin', co: 'latin', fy: 'latin', gd: 'latin', gu: 'latin',
ht: 'latin', haw: 'latin', hmn: 'latin', jv: 'latin', ku: 'latin',
mg: 'latin', mi: 'latin', mn: 'latin', nso: 'latin', ny: 'latin',
oc: 'latin', os: 'latin', ps: 'latin', qu: 'latin', rw: 'latin',
sc: 'latin', si: 'latin', sm: 'latin', sn: 'latin', so: 'latin',
st: 'latin', su: 'latin', tg: 'latin', tt: 'latin', ty: 'latin',
ug: 'latin', vo: 'latin', wa: 'latin', wo: 'latin', xh: 'latin',
yi: 'latin', zu: 'latin',
// Cyrillic overrides
ru: 'cyrillic', uk: 'cyrillic', be: 'cyrillic', sr: 'cyrillic',
mk: 'cyrillic', bg: 'cyrillic', ce: 'cyrillic', cv: 'cyrillic',
sah: 'cyrillic', udm: 'cyrillic', rue: 'cyrillic', ab: 'cyrillic',
// Greek
el: 'greek',
// Arabic-script languages
ar: 'arabic', fa: 'arabic', ur: 'arabic', ps: 'arabic',
ku: 'arabic', sd: 'arabic', ckb: 'arabic', bal: 'arabic',
bqi: 'arabic', glk: 'arabic', mzn: 'arabic',
// Hebrew & Yiddish
he: 'hebrew',
yi: 'hebrew', // Yiddish uses Hebrew script, not Arabic
// Thaana (Maldivian)
dv: 'thaana', // Maldivian uses Thaana, not Arabic
// Indian scripts
hi: 'devanagari', ne: 'devanagari', mr: 'devanagari', sa: 'devanagari',
mai: 'devanagari', bho: 'devanagari', awa: 'devanagari',
bn: 'bengali', as: 'bengali',
ta: 'tamil',
te: 'telugu',
kn: 'kannada',
ml: 'malayalam',
si: 'sinhala',
gu: 'gujarati',
pa: 'gurmukhi',
// SE Asia
th: 'thai',
lo: 'lao',
my: 'burmese',
km: 'khmer',
// East Asia
zh: 'cjk', 'zh-cn': 'cjk', 'zh-tw': 'cjk', 'zh-hk': 'cjk',
ja: 'hiragana_katakana',
ko: 'hangul',
// Others
ka: 'georgian',
hy: 'armenian',
am: 'ethiopic', ti: 'ethiopic', gez: 'ethiopic', tig: 'ethiopic',
bo: 'tibetan', dz: 'tibetan',
};
// ---------- Discriminating characters for Arabic-script languages ----------
const DISCRIMINATING_CHARS: Record<string, ReadonlySet<string>> = {
fa: new Set('پچژگ'), // Persian
ur: new Set('ٹڈڑے'), // Urdu
ps: new Set('ټډړږښ'), // Pashto
ku: new Set('ڕێ'), // Kurdish
ckb: new Set('ڕێ'), // Sorani Kurdish
sd: new Set('ٿ'), // Sindhi
ug: new Set('ۇۆې'), // Uyghur
bal: new Set('ێ'), // Balochi
};
// ---------- Thresholds ----------
const MIN_RATIO_IN_SCRIPT = 0.60;
// ---------- Helpers ----------
/** Get the script id for a language code. */
export function getScript(langCode: string | null | undefined): string {
if (!langCode) return 'latin';
return LANG_TO_SCRIPT[langCode.toLowerCase()] ?? 'latin';
}
/** True if the language uses an Arabic-script block. */
export function isArabicScriptLang(langCode: string | null | undefined): boolean {
return getScript(langCode) === 'arabic';
}
/** Get the Unicode ranges for a script id. Empty array for 'latin'. */
function getRanges(scriptId: string): readonly Range[] {
return UNICODE_RANGES[scriptId] ?? [];
}
/** Iterate code points of a string (handles surrogate pairs correctly). */
function* codePoints(text: string): Generator<number> {
for (const ch of text) {
yield ch.codePointAt(0) ?? 0;
}
}
/** Check if a code point falls in any of the (start, end) ranges. */
function isInRanges(code: number, ranges: readonly Range[]): boolean {
for (const [start, end] of ranges) {
if (code >= start && code <= end) return true;
}
return false;
}
const LETTER_REGEX = /\p{L}/u;
/** Count alphabetic characters (using Unicode property escapes). */
function countLetters(text: string): number {
let count = 0;
for (const ch of text) {
if (LETTER_REGEX.test(ch)) count++;
}
return count;
}
/** Count how many alphabetic characters fall within the given ranges. */
function countInScript(text: string, ranges: readonly Range[]): number {
if (ranges.length === 0) {
// Latin / unknown — all letters match.
return countLetters(text);
}
let count = 0;
for (const ch of text) {
if (!LETTER_REGEX.test(ch)) continue;
const cp = ch.codePointAt(0) ?? 0;
if (isInRanges(cp, ranges)) count++;
}
return count;
}
// ---------- Arabic-script variant detection ----------
export interface ArabicVariantResult {
verdict: 'pass' | 'fail' | 'skip';
claimedLang: string | null;
detectedVariants: string[];
arabicRatio?: number;
reason: string;
}
export function detectArabicVariant(
text: string,
claimedLang: string | null,
): ArabicVariantResult {
if (!text || !text.trim()) {
return { verdict: 'skip', claimedLang, detectedVariants: [], reason: 'empty text' };
}
const arabicRanges = getRanges('arabic');
const letters = countLetters(text);
if (letters === 0) {
return { verdict: 'skip', claimedLang, detectedVariants: [], reason: 'no letters' };
}
const inArabic = countInScript(text, arabicRanges);
const arabicRatio = inArabic / letters;
if (arabicRatio < MIN_RATIO_IN_SCRIPT) {
return {
verdict: 'skip',
claimedLang,
detectedVariants: [],
arabicRatio: round(arabicRatio, 3),
reason: 'not in Arabic script',
};
}
if (!isArabicScriptLang(claimedLang)) {
return {
verdict: 'skip',
claimedLang,
detectedVariants: [],
arabicRatio: round(arabicRatio, 3),
reason: 'target is not an Arabic-script language',
};
}
// Text is Arabic-script AND target is Arabic-script: check the variant.
const detected = new Set<string>();
for (const [langCode, chars] of Object.entries(DISCRIMINATING_CHARS)) {
if (chars.size === 0) continue;
for (const ch of text) {
if (chars.has(ch)) {
detected.add(langCode);
break;
}
}
}
if (detected.size > 0 && claimedLang && !detected.has(claimedLang.toLowerCase())) {
return {
verdict: 'fail',
claimedLang,
detectedVariants: Array.from(detected).sort(),
arabicRatio: round(arabicRatio, 3),
reason: `target=${claimedLang} but text contains characters typical of ${Array.from(detected).sort().join(', ')}`,
};
}
return {
verdict: 'pass',
claimedLang,
detectedVariants: detected.size > 0 ? Array.from(detected).sort() : (claimedLang ? [claimedLang] : []),
arabicRatio: round(arabicRatio, 3),
reason: 'ok',
};
}
// ---------- Length check ----------
const RATIO_MAX = 3.5;
const RATIO_MIN = 0.15;
const ABSOLUTE_MIN_LENGTH = 2;
const MIN_SOURCE_LENGTH_FOR_RATIO = 20;
export interface LengthCheckResult {
issue: string | null;
ratio: number | null;
sourceLength: number;
translatedLength: number;
note?: string;
}
export function lengthCheck(sourceText: string, translatedText: string): LengthCheckResult {
if (!sourceText) {
return {
issue: null,
ratio: null,
sourceLength: 0,
translatedLength: (translatedText || '').length,
};
}
const srcLen = sourceText.trim().length;
const transLen = (translatedText || '').trim().length;
if (transLen === 0) {
return {
issue: 'truncation_suspect',
ratio: 0,
sourceLength: srcLen,
translatedLength: transLen,
};
}
if (isMostlyNumeric(sourceText)) {
return {
issue: null,
ratio: null,
sourceLength: srcLen,
translatedLength: transLen,
note: 'skipped: numeric source',
};
}
if (srcLen < MIN_SOURCE_LENGTH_FOR_RATIO) {
return {
issue: null,
ratio: null,
sourceLength: srcLen,
translatedLength: transLen,
};
}
const ratio = transLen / srcLen;
if (ratio > RATIO_MAX) {
return {
issue: 'length_outlier',
ratio: round(ratio, 2),
sourceLength: srcLen,
translatedLength: transLen,
};
}
if (ratio < RATIO_MIN) {
return {
issue: 'truncation_suspect',
ratio: round(ratio, 2),
sourceLength: srcLen,
translatedLength: transLen,
};
}
return {
issue: null,
ratio: round(ratio, 2),
sourceLength: srcLen,
translatedLength: transLen,
};
}
function isMostlyNumeric(text: string): boolean {
if (!text) return false;
const chars: string[] = [];
for (const ch of text) {
if (!/\s/.test(ch)) chars.push(ch);
}
if (chars.length === 0) return false;
const digitCount = chars.filter((c) => /\d/.test(c)).length;
return digitCount / chars.length >= 0.5;
}
// ---------- Pattern leak / repetition ----------
const LEAK_PREFIX_PATTERNS: RegExp[] = [
/^(translation|translated text|here is the translation|here'?s the translation)\s*[:：-]/i,
/^(voici (la |ma )?traduction|traduction\s*[:：-])\b/i,
/^(原文|译|翻译|译为|以下是)\s*[:：]?/u,
/^(sure,?\s+here'?s?\s+(the\s+)?translation|of course,?\s+here)/i,
/^(\*\*|__|#)\s*translation/i,
/^translated from\s+\w+\s+to\s+\w+\s*[:：-]/i,
];
const REPETITION_THRESHOLD = 5;
const CHAR_REPETITION_THRESHOLD = 20;
export interface PatternCheckResult {
issue: string | null;
matchedPattern: string | null;
repetitionCount: number | null;
}
export function patternCheck(text: string): PatternCheckResult {
if (!text || !text.trim()) {
return { issue: null, matchedPattern: null, repetitionCount: null };
}
const stripped = text.trimStart();
// 1. Prompt leak
for (const pat of LEAK_PREFIX_PATTERNS) {
if (pat.test(stripped)) {
return {
issue: 'prompt_leak',
matchedPattern: pat.source,
repetitionCount: null,
};
}
}
// 2. Token-level repetition
const tokens = stripped.split(/\s+/).filter((t) => t.length > 0);
const tokenRep = maxConsecutiveTokenRepetition(tokens);
if (tokenRep >= REPETITION_THRESHOLD) {
return {
issue: 'repetition_hallucination',
matchedPattern: null,
repetitionCount: tokenRep,
};
}
// 3. Character-level repetition
const charRep = maxConsecutiveCharRepetition(stripped);
if (charRep >= CHAR_REPETITION_THRESHOLD) {
return {
issue: 'repetition_hallucination',
matchedPattern: null,
repetitionCount: charRep,
};
}
return {
issue: null,
matchedPattern: null,
repetitionCount: Math.max(tokenRep, charRep) || null,
};
}
function maxConsecutiveTokenRepetition(tokens: string[]): number {
if (tokens.length === 0) return 0;
const norm = tokens.map((t) => t.toLowerCase().replace(/[.,!?;:"'`()\[\]{}]/g, ''));
let maxRun = 1;
let currentRun = 1;
for (let i = 1; i < norm.length; i++) {
if (norm[i] && norm[i] === norm[i - 1]) {
currentRun++;
if (currentRun > maxRun) maxRun = currentRun;
} else {
currentRun = 1;
}
}
return maxRun;
}
function maxConsecutiveCharRepetition(text: string): number {
if (!text) return 0;
let maxRun = 1;
let currentRun = 1;
for (let i = 1; i < text.length; i++) {
if (text[i] === text[i - 1] && !/\s/.test(text[i])) {
currentRun++;
if (currentRun > maxRun) maxRun = currentRun;
} else {
currentRun = 1;
}
}
return maxRun;
}
// ---------- Heuristic actual-script detection (for diagnostics) ----------
const SCRIPT_DETECTION_ORDER: readonly string[] = [
'hiragana_katakana', 'hangul', 'cjk', 'thai', 'lao', 'burmese',
'khmer', 'devanagari', 'bengali', 'tamil', 'telugu', 'kannada',
'malayalam', 'sinhala', 'gujarati', 'gurmukhi',
'arabic', 'hebrew', 'cyrillic', 'greek', 'armenian', 'georgian',
'ethiopic', 'tibetan', 'thaana',
];
function detectActualScript(text: string): string {
const letters = countLetters(text);
if (letters === 0) return 'unknown';
for (const scriptId of SCRIPT_DETECTION_ORDER) {
const ranges = getRanges(scriptId);
const inScript = countInScript(text, ranges);
if (inScript / letters > 0.4) return scriptId;
}
return 'latin';
}
// ---------- Main per-chunk evaluation ----------
export function evaluateChunk(
sourceText: string,
translatedText: string | null | undefined,
targetLang: string | null | undefined,
): QualityCheckResult {
if (translatedText === null || translatedText === undefined) {
return {
passed: true,
score: 0,
issues: ['empty_translation'],
detectedScript: null,
expectedScript: null,
details: { reason: 'translation is null/undefined' },
};
}
const text = translatedText.trim();
if (!text) {
return {
passed: true,
score: 0,
issues: ['empty_translation'],
detectedScript: null,
expectedScript: null,
details: { reason: 'translation is empty or whitespace-only' },
};
}
const targetLangLower = (targetLang || '').toLowerCase() || null;
const issues: string[] = [];
const details: Record<string, unknown> = {};
// --- Script detection ---
const expectedScript = getScript(targetLangLower);
const expectedRanges = getRanges(expectedScript);
const letters = countLetters(text);
let scriptScore: number;
let detectedScript: string | null;
if (letters === 0) {
scriptScore = 1.0;
detectedScript = expectedScript;
details.script_check = 'skipped: no alphabetic characters';
} else {
detectedScript = detectActualScript(text);
const inExpected = countInScript(text, expectedRanges);
scriptScore = inExpected / letters;
details.script_score = round(scriptScore, 3);
details.letters_in_text = letters;
details.letters_in_script = inExpected;
details.detected_script = detectedScript;
details.expected_script = expectedScript;
details.min_ratio = MIN_RATIO_IN_SCRIPT;
if (expectedScript !== 'latin' && expectedRanges.length > 0) {
// Specific non-Latin target.
if (scriptScore < MIN_RATIO_IN_SCRIPT) {
issues.push('wrong_script');
details.reason = `only ${Math.round(scriptScore * 100)}% of letters match ${expectedScript} script; text appears to be in ${detectedScript}`;
}
} else {
// Latin target. If detected script is clearly non-Latin, fail.
if (detectedScript && detectedScript !== 'latin' && detectedScript !== 'unknown') {
const nonLatinRanges = getRanges(detectedScript);
const inDetected = countInScript(text, nonLatinRanges);
const nonLatinConfidence = inDetected / letters;
if (nonLatinConfidence >= 0.7) {
issues.push('wrong_script');
details.reason = `target is Latin but ${Math.round(nonLatinConfidence * 100)}% of letters are in ${detectedScript} script — language confusion`;
}
}
}
}
// --- Arabic-script variant detection ---
if (isArabicScriptLang(targetLangLower)) {
const variantResult = detectArabicVariant(text, targetLangLower);
details.arabic_variant = variantResult;
if (variantResult.verdict === 'fail') {
issues.push('wrong_arabic_variant');
}
}
// --- Length sanity ---
const lengthResult = lengthCheck(sourceText, text);
details.length = lengthResult;
if (lengthResult.issue) {
issues.push(lengthResult.issue);
}
// --- Pattern leak / repetition ---
const leakResult = patternCheck(text);
details.pattern_check = leakResult;
if (leakResult.issue) {
issues.push(leakResult.issue);
}
// --- Aggregate ---
const passed = issues.length === 0;
const nChecks = 3;
const nFailed = issues.filter((issue) =>
['wrong_script', 'wrong_arabic_variant', 'length_outlier', 'truncation_suspect', 'prompt_leak', 'repetition_hallucination'].includes(issue),
).length;
const score = Math.max(0, 1 - nFailed / nChecks);
return {
passed,
score: round(score, 3),
issues,
detectedScript,
expectedScript,
details,
};
}
// ---------- Document-level aggregation ----------
export function evaluateDocument(
sourceChunks: string[],
translatedChunks: string[],
targetLang: string | null | undefined,
sampleSize: number = 50,
): DocumentQualityResult {
const n = Math.min(sourceChunks.length, translatedChunks.length);
const chunkResults: QualityCheckResult[] = [];
const issuesCount: Record<string, number> = {};
const samples: DocumentQualityResult['samples'] = [];
let scoreSum = 0;
let failedCount = 0;
for (let i = 0; i < n; i++) {
const r = evaluateChunk(sourceChunks[i], translatedChunks[i], targetLang);
chunkResults.push(r);
scoreSum += r.score;
for (const issue of r.issues) {
issuesCount[issue] = (issuesCount[issue] || 0) + 1;
}
if (!r.passed) {
failedCount++;
if (samples.length < sampleSize) {
samples.push({
index: i,
issues: r.issues,
sourcePreview: (sourceChunks[i] || '').slice(0, 80),
translatedPreview: (translatedChunks[i] || '').slice(0, 80),
details: r.details,
});
}
}
}
const meanScore = n > 0 ? scoreSum / n : 0;
const passed = failedCount === 0;
return {
passed,
score: round(meanScore, 3),
chunkCount: n,
failedChunkCount: failedCount,
issues: issuesCount,
samples,
};
}
// ---------- Utilities ----------
function round(value: number, decimals: number): number {
const factor = 10 ** decimals;
return Math.round(value * factor) / factor;
}

421
wordly.art---traduction-de-documents/tests/utils/scriptDetector.test.ts Normal file
View File

@@ -0,0 +1,421 @@
/**
* Tests for src/utils/scriptDetector.ts
*
* Run with:
* npx tsx --test tests/utils/scriptDetector.test.ts
*
* Or in package.json scripts:
* "test:quality": "tsx --test tests/utils/scriptDetector.test.ts"
*/
import { describe, it } from 'node:test';
import { strict as assert } from 'node:assert';
import {
getScript,
isArabicScriptLang,
evaluateChunk,
evaluateDocument,
detectArabicVariant,
lengthCheck,
patternCheck,
} from '../../src/utils/scriptDetector.ts';
describe('getScript', () => {
it('maps cyrillic languages', () => {
assert.equal(getScript('ru'), 'cyrillic');
assert.equal(getScript('uk'), 'cyrillic');
assert.equal(getScript('be'), 'cyrillic');
});
it('maps latin languages', () => {
assert.equal(getScript('en'), 'latin');
assert.equal(getScript('fr'), 'latin');
assert.equal(getScript('de'), 'latin');
assert.equal(getScript('vi'), 'latin');
});
it('maps CJK languages', () => {
assert.equal(getScript('zh'), 'cjk');
assert.equal(getScript('zh-cn'), 'cjk');
});
it('maps Japanese to hiragana_katakana', () => {
assert.equal(getScript('ja'), 'hiragana_katakana');
});
it('maps Korean to hangul', () => {
assert.equal(getScript('ko'), 'hangul');
});
it('maps Yiddish to hebrew (not arabic)', () => {
assert.equal(getScript('yi'), 'hebrew');
});
it('maps Maldivian to thaana (not arabic)', () => {
assert.equal(getScript('dv'), 'thaana');
});
it('falls back to latin for unknown', () => {
assert.equal(getScript('xx'), 'latin');
assert.equal(getScript(''), 'latin');
assert.equal(getScript(null), 'latin');
});
it('is case-insensitive', () => {
assert.equal(getScript('FR'), 'latin');
assert.equal(getScript('ZH-CN'), 'cjk');
});
});
describe('isArabicScriptLang', () => {
it('returns true for Arabic-script langs', () => {
assert.equal(isArabicScriptLang('ar'), true);
assert.equal(isArabicScriptLang('fa'), true);
assert.equal(isArabicScriptLang('ur'), true);
assert.equal(isArabicScriptLang('ckb'), true);
});
it('returns false for non-Arabic langs', () => {
assert.equal(isArabicScriptLang('fr'), false);
assert.equal(isArabicScriptLang('he'), false);
assert.equal(isArabicScriptLang('en'), false);
});
});
// ---------- Happy path ----------
describe('evaluateChunk — correct script', () => {
it('russian', () => {
const r = evaluateChunk('Hello world', 'Привет мир', 'ru');
assert.equal(r.passed, true);
assert.ok(!r.issues.includes('wrong_script'));
});
it('chinese', () => {
const r = evaluateChunk('Hello', '你好世界', 'zh');
assert.equal(r.passed, true);
});
it('arabic', () => {
const r = evaluateChunk('Hello', 'مرحبا بالعالم', 'ar');
assert.equal(r.passed, true);
});
it('persian (with unique chars)', () => {
const r = evaluateChunk('Hello', 'سلام چطوری؟ من پژوهشگر هستم', 'fa');
assert.equal(r.passed, true);
});
it('french', () => {
const r = evaluateChunk('Hello', 'Bonjour le monde', 'fr');
assert.equal(r.passed, true);
});
it('hebrew', () => {
const r = evaluateChunk('Hello', 'שלום עולם', 'he');
assert.equal(r.passed, true);
});
it('korean', () => {
const r = evaluateChunk('Hello', '안녕하세요 세계', 'ko');
assert.equal(r.passed, true);
});
it('japanese', () => {
const r = evaluateChunk('Hello', 'こんにちは世界', 'ja');
assert.equal(r.passed, true);
});
it('hindi', () => {
const r = evaluateChunk('Hello', 'नमस्ते दुनिया', 'hi');
assert.equal(r.passed, true);
});
it('thai', () => {
const r = evaluateChunk('Hello', 'สวัสดีชาวโลก', 'th');
assert.equal(r.passed, true);
});
it('greek', () => {
const r = evaluateChunk('Hello', 'Γεια σας κόσμε', 'el');
assert.equal(r.passed, true);
});
});
// ---------- Wrong script detection ----------
describe('evaluateChunk — wrong script', () => {
it('french text for japanese target', () => {
const r = evaluateChunk('Hello', 'Bonjour le monde', 'ja');
assert.equal(r.passed, false);
assert.ok(r.issues.includes('wrong_script'));
});
it('arabic text for french target (language confusion)', () => {
const r = evaluateChunk('Hello', 'مرحبا بالعالم', 'fr');
assert.equal(r.passed, false);
assert.ok(r.issues.includes('wrong_script'));
});
it('russian text for english target', () => {
const r = evaluateChunk('Hello', 'Привет мир', 'en');
assert.equal(r.passed, false);
assert.ok(r.issues.includes('wrong_script'));
});
it('chinese text for korean target', () => {
const r = evaluateChunk('Hello', '你好世界', 'ko');
assert.equal(r.passed, false);
assert.ok(r.issues.includes('wrong_script'));
});
});
// ---------- Arabic variant discrimination ----------
describe('evaluateChunk — Arabic variant discrimination', () => {
it('persian text for arabic target fails', () => {
const r = evaluateChunk('Hello', 'سلام چطوری؟', 'ar');
assert.equal(r.passed, false);
assert.ok(r.issues.includes('wrong_arabic_variant'));
});
it('urdu text for persian target fails', () => {
const r = evaluateChunk('Hello', 'السلام ٹڈ', 'fa');
assert.equal(r.passed, false);
assert.ok(r.issues.includes('wrong_arabic_variant'));
});
it('pashto text for arabic target fails', () => {
const r = evaluateChunk('Hello', 'السلام ټډړ', 'ar');
assert.equal(r.passed, false);
assert.ok(r.issues.includes('wrong_arabic_variant'));
});
});
// ---------- Edge cases ----------
describe('evaluateChunk — edge cases', () => {
it('empty translation passes (skipped)', () => {
const r = evaluateChunk('Hello', '', 'fr');
assert.equal(r.passed, true);
assert.ok(r.issues.includes('empty_translation'));
});
it('whitespace-only translation passes', () => {
const r = evaluateChunk('Hello', ' \n ', 'fr');
assert.equal(r.passed, true);
});
it('null translation passes', () => {
const r = evaluateChunk('Hello', null, 'fr');
assert.equal(r.passed, true);
});
it('undefined translation passes', () => {
const r = evaluateChunk('Hello', undefined, 'fr');
assert.equal(r.passed, true);
});
it('numbers only pass for fr target', () => {
const r = evaluateChunk('Price: 100', '100', 'fr');
assert.equal(r.passed, true);
});
it('unknown target lang falls back to latin', () => {
const r = evaluateChunk('Hello', 'Bonjour', 'xx');
assert.equal(r.passed, true);
});
});
// ---------- Length ----------
describe('evaluateChunk — length', () => {
it('huge translation flagged via repetition (source too short)', () => {
const r = evaluateChunk('Short', 'x'.repeat(1000), 'fr');
assert.ok(r.issues.includes('repetition_hallucination'));
});
it('huge translation with long source flagged as length', () => {
const src = 'A'.repeat(200);
const r = evaluateChunk(src, 'x'.repeat(1000), 'fr');
assert.ok(r.issues.includes('length_outlier'));
});
it('tiny translation flagged', () => {
const r = evaluateChunk('A'.repeat(100), 'ok', 'fr');
assert.ok(r.issues.includes('truncation_suspect'));
});
it('numeric source skips length check', () => {
const r = evaluateChunk('Price: 100', '100', 'fr');
assert.ok(!r.issues.includes('truncation_suspect'));
assert.ok(!r.issues.includes('length_outlier'));
});
});
// ---------- Pattern leak / repetition ----------
describe('evaluateChunk — pattern leak', () => {
it('english prompt leak', () => {
const r = evaluateChunk('Hello', 'Translation: Bonjour le monde', 'fr');
assert.ok(r.issues.includes('prompt_leak'));
});
it('french prompt leak', () => {
const r = evaluateChunk('Hello', 'Voici la traduction : Bonjour', 'fr');
assert.ok(r.issues.includes('prompt_leak'));
});
it('chinese prompt leak', () => {
const r = evaluateChunk('Hello', '翻译：你好', 'zh');
assert.ok(r.issues.includes('prompt_leak'));
});
it('token repetition detected', () => {
const r = evaluateChunk('Hello', 'the the the the the the the', 'fr');
assert.ok(r.issues.includes('repetition_hallucination'));
});
it('char repetition detected', () => {
const r = evaluateChunk('Hello', 'x'.repeat(30), 'fr');
assert.ok(r.issues.includes('repetition_hallucination'));
});
it('normal text has no leak', () => {
const r = evaluateChunk('Hello', 'Bonjour le monde', 'fr');
assert.ok(!r.issues.includes('prompt_leak'));
assert.ok(!r.issues.includes('repetition_hallucination'));
});
});
// ---------- Document-level ----------
describe('evaluateDocument', () => {
it('all good', () => {
const result = evaluateDocument(
['Hello', 'World', 'Good morning'],
['Bonjour', 'Monde', 'Bonjour'],
'fr',
);
assert.equal(result.passed, true);
assert.equal(result.chunkCount, 3);
assert.equal(result.failedChunkCount, 0);
});
it('some failures', () => {
const result = evaluateDocument(
['Hello', 'World', 'Good morning'],
['Bonjour', 'مرحبا', 'Bonjour'],
'fr',
);
assert.equal(result.passed, false);
assert.equal(result.failedChunkCount, 1);
assert.ok('wrong_script' in result.issues);
assert.equal(result.samples.length, 1);
});
it('empty lists', () => {
const result = evaluateDocument([], [], 'fr');
assert.equal(result.passed, true);
assert.equal(result.chunkCount, 0);
});
it('sample size caps samples', () => {
const result = evaluateDocument(
Array(10).fill('hi'),
Array(10).fill('مرحبا'),
'fr',
3,
);
assert.equal(result.failedChunkCount, 10);
assert.equal(result.samples.length, 3);
});
});
// ---------- detectArabicVariant direct ----------
describe('detectArabicVariant', () => {
it('persian for persian passes', () => {
const r = detectArabicVariant('سلام چطوری', 'fa');
assert.equal(r.verdict, 'pass');
});
it('persian for arabic fails', () => {
const r = detectArabicVariant('سلام چطوری', 'ar');
assert.equal(r.verdict, 'fail');
assert.ok(r.detectedVariants.includes('fa'));
});
it('urdu for persian fails', () => {
const r = detectArabicVariant('السلام ٹڈ', 'fa');
assert.equal(r.verdict, 'fail');
assert.ok(r.detectedVariants.includes('ur'));
});
it('non-arabic text skipped', () => {
const r = detectArabicVariant('Bonjour le monde', 'fa');
assert.equal(r.verdict, 'skip');
});
});
// ---------- lengthCheck direct ----------
describe('lengthCheck', () => {
it('normal length returns ratio', () => {
const r = lengthCheck(
'Hello world this is a longer source string',
'Bonjour le monde ceci est une traduction francaise',
);
assert.equal(r.issue, null);
assert.ok(r.ratio !== null);
});
it('huge translation', () => {
const r = lengthCheck('A'.repeat(200), 'x'.repeat(1000));
assert.equal(r.issue, 'length_outlier');
});
it('tiny translation', () => {
const r = lengthCheck('A'.repeat(100), 'ok');
assert.equal(r.issue, 'truncation_suspect');
});
it('empty translation flagged', () => {
const r = lengthCheck('Hello world this is a test', '');
assert.equal(r.issue, 'truncation_suspect');
});
});
// ---------- patternCheck direct ----------
describe('patternCheck', () => {
it('no leak on normal text', () => {
const r = patternCheck('Bonjour le monde');
assert.equal(r.issue, null);
});
it('english leak', () => {
const r = patternCheck('Translation: Bonjour');
assert.equal(r.issue, 'prompt_leak');
});
it('french leak', () => {
const r = patternCheck('Voici la traduction : Bonjour');
assert.equal(r.issue, 'prompt_leak');
});
it('chinese leak', () => {
const r = patternCheck('翻译：你好');
assert.equal(r.issue, 'prompt_leak');
});
it('token repetition', () => {
const r = patternCheck('the the the the the the the');
assert.equal(r.issue, 'repetition_hallucination');
});
it('char repetition', () => {
const r = patternCheck('x'.repeat(30));
assert.equal(r.issue, 'repetition_hallucination');
});
});