Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Number Lock: Translation Figure Check, English & Arabic

    1

    A verdict of PASS, REVIEW or FAIL, with exit codes for scripts and CI.

    $12

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Number Lock: Translation Figure Check, English & Arabic

    Number Lock: Translation Figure Check, English & Arabic

    Example session with this skill installed

    Check that the Arabic translation kept every figure.
    Source (EN): Revenue rose 15% to $2.4 billion in 2025. / The fee is SAR 1,250 per year. / Profit margin improved by 2.5 percentage points. / The board meets on 12/04/2026.
    Target (AR): ارتفعت الإيرادات بنسبة ٥١٪ لتصل إلى ٢٫٤ مليون دولار في عام ٢٠٢٥. / الرسوم ١٬٢٥٠ دولارًا سنويًا. / تحسّن هامش الربح بمقدار ٢٫٥٪. / يجتمع المجلس في ٢١/٠٤/٢٠٢٦.

    • Read your context and instructions
    • Compiled the number lock

    Number Lock report

    Verdict: FAIL (5 fail, 2 review, 2 style, 0 note)

    • Direction: latin-script -> arabic | alignment: lines | segments: 13
    • Figures read: 20 in source, 19 in target; 16 matched
    • Source/Target cells show figures as read (Arabic-Indic digits as 0-9, diacritics removed)
    #SevCheckSegmentSourceTargetWhySuggest
    1FAILNUMBER_CHANGEDline 115%51%15% became 51%١٥٪
    2FAILSCALE_CHANGEDline 1$2.4 billion2.4 مليون دولارdollar 2,400,000,000 became dollar 2,400,000 (off by a factor of 1000)٢٫٤ مليار دولار
    3FAILCURRENCY_CHANGEDline 6SAR 1,2501,250 دولاراcurrency SAR in source, dollar in target١٬٢٥٠ ريال سعودي
    4FAILPERCENT_VS_POINTSline 72.5 percentage points2.5%percentage points vs percent: different quantities٢٫٥ نقطة مئوية
    5FAILDATE_CHANGEDline 1112/04/202621/04/20262026-12-04 or 2026-04-12 in source, 2026-04-21 in targetuse the source date: ١٢/٠٤/٢٠٢٦
    6REVIEWNEGATION_DROPPEDline 41 negation word(s)nonesource has a negation (not/no/never, la/lam/lan/laysa/mish) and the target has no negation at all; heuristic-
    7REVIEWDUAL_OR_SINGULARline 52-Arabic can express 1 or 2 with a singular or dual noun (e.g. two days = yawmayn); confirm the count survived-
    8STYLEMIXED_DIGITSline 8--target mixes Western (0-9) and Arabic-Indic digits in one segment-
    9STYLEMIXED_DIGITSline 12--target mixes Western (0-9) and Arabic-Indic digits in one segment-

    Connects securely to your tools. The creator never sees your data.

    About this skill

    A translation can read perfectly and still say 51% where the source said 15%, or 2.4 million where it said 2.4 billion. Machine translation and LLMs make figure errors often enough that researchers now study numerical translation as its own error type (e.g. arXiv 2501.04927), and a reviewer reading for flow rarely re-counts every figure. In English-Arabic work it is harder still: different digits (٠-٩), different separators (٫ ٬ ٪), number words with gender and case, dual nouns that hide the number 2, and three sets of month names.

    What you get. A short table for every segment where a figure changed: the source value, the target value, why it matters, and the value to put back, written in the target's digit style (١٥٪, ٢٫٤ مليار دولار, ١٬٢٥٠ ريال سعودي). A verdict of PASS, REVIEW or FAIL, with exit codes for scripts and CI.

    How it works. A standard-library Python script reads both sides the same way and matches figures across scripts:

    • 14 = ١٤ = fourteen = أربعة عشر; 2.5 million = مليونين ونصف; 3.5 million = ٣ ملايين و٥٠٠ ألف
    • scales (thousand/million/billion, مليار, bn/mn, 10K), percent vs percentage points vs basis points
    • money: $, €, £, ISO codes, ر.س، د.إ، ج.م، د.ك, currency words with nationality (ريال سعودي, جنيه مصري)
    • dates in English, Egyptian/Gulf, Levantine (آذار، تشرين الأول) and Maghrebi (شتنبر، جانفي) month names, ordinal days (الثالث من يناير), numeric dates, Hijri months and years
    • times: 9:30 a.m., 5 p.m., ٥ مساءً, الساعة الخامسة والنصف, noon/midnight
    • Arabic duals and singulars flagged for review, never failed
    • a negation heuristic for a dropped "not" (لا، لم، لن، ليس، ألا، مش)

    It reads two line-aligned text files (and realigns them by their figures when a line was split, merged or dropped), TSV/CSV tables, and XLIFF 1.2/2.0 from CAT tools, with segment ids in every finding.

    Limits. Figures only: it does not judge meaning, tone or terminology. Units and calendars are flagged, not converted. Ordinals and minus signs are not compared. Two figures that swap places inside one sentence are not detected.

    Figures only; it does not judge meaning, tone or terminology. Unit conversions (5 miles vs 8 km) and Hijri/Gregorian conversions are flagged, not converted. Minus signs and ordinals are not compared. Two figures swapping places inside one sentence are not detected. A count carried only by an Arabic dual noun (أسبوعين) is not read as a number. Number words are read for English and Arabic only; other languages are checked on digits. Text files, tables and XLIFF only: export DOCX/PDF to text first.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 1 day ago

    Requires

    Python 3.8+
    Terminal access
    Python

    Permissions

    File Scopes

    reads
    the
    source/target
    or
    pairs
    files
    you
    name
    writes
    only
    --out
    /
    --json
    report
    (refuses
    to
    overwrite
    without
    --force)

    Runs fully offline. No network calls, no subprocesses. Never modifies your translation files.

    Listed1 day ago

    What's inside

    Frequently Asked Questions

    No. It only reads the source and the translation and reports figures that do not match, with the value to put back. Your files are never modified; reports are written only where you ask with --out/--json.

    It is built for English <-> Arabic: Arabic-Indic digits, Arabic separators, Arabic and English number words, Arabic month names in Egyptian/Gulf, Levantine and Maghrebi forms, Hijri dates. For other language pairs it checks every figure written in digits.

    Yes. Pass an XLIFF 1.2/2.0 file or a bilingual TSV/CSV with --pairs. Each finding carries the unit or segment id so you can jump to it in Trados, memoQ, Phrase or similar tools.

    It realigns the lines by their figures and compares split, merged or dropped lines together, and says so in the report. You can also use --whole to compare the files as one document.

    It is tuned to stay quiet on correct translations. In our pre-release tests, 3 of 289 correct English-Arabic pairs were flagged (an idiom, Egyptian ما as a negation, and halalas vs riyals). Arabic duals, calendar differences and the negation check are REVIEW, never FAIL, so a human decides.

    New in Translation & Localization