multilingual-score

What this skill does

# /digital-marketing-pro:multilingual-score

## Purpose

Score translated or localized content across multiple quality dimensions to determine whether it is ready for publishing, needs native speaker review, or requires re-translation. Combines technical translation scoring (length ratios, formatting preservation, placeholder integrity, key term consistency) with content quality evaluation, brand voice consistency checking, and market-specific compliance verification into a single composite multilingual quality score.

Use this command after any translation or localization workflow to validate quality before content goes live. It replaces subjective "looks good" assessments with a structured, repeatable scoring methodology that catches issues automated translation often introduces — brand voice drift, formatting damage, missing do-not-translate terms, compliance gaps in the target market, and length distortion that signals missing or added content. The composite score provides a clear publish/review/re-translate classification so the team knows exactly what action to take.

## Input Required

The user must provide (or will be prompted for):

- **Original content**: The source text that was translated — provided as inline text, a file path, or a URL to the source content. This serves as the reference for translation accuracy scoring. Required for technical translation scoring; if omitted, only content quality, brand voice, and compliance dimensions are scored
- **Translated content**: The translated or localized text to score — provided as inline text, a file path, or a URL. This is the primary content being evaluated. Required
- **Source language**: The language code of the original content (e.g., `en-US`, `en-GB`, `de-DE`). Defaults to the brand's primary language from the language configuration if not specified
- **Target language**: The language code of the translated content (e.g., `de-DE`, `fr-FR`, `hi-IN`, `ja-JP`). Required — determines which compliance rules apply and which translation service benchmarks to reference
- **Do-not-translate terms** (optional): Specific terms that must appear unchanged in the translation. Defaults to the brand profile's `language.do_not_translate` list. Additional terms can be provided to supplement the brand list for this specific scoring run
- **Content type** (optional): The type of content being scored — `blog`, `email`, `ad`, `landing_page`, `social`, `product_description`, `legal`, `technical`. Affects content quality scoring weights and brand voice expectations. Defaults to auto-detection based on content structure

## Process

1. **Load brand context**: Read `~/.claude-marketing/brands/_active-brand.json` for the active slug, then load `~/.claude-marketing/brands/{slug}/profile.json`. Apply brand voice dimensions, compliance rules for target markets (`skills/context-engine/compliance-rules.md`), and industry context. Load the language configuration — specifically the do-not-translate terms from `language.do_not_translate` and any translation quality baselines from past scoring runs. Also check for guidelines at `~/.claude-marketing/brands/{slug}/guidelines/_manifest.json` — if present, load restrictions and voice-and-tone rules that apply across all languages. Check for agency SOPs at `~/.claude-marketing/sops/`. If no brand exists, ask: "Set up a brand first (/digital-marketing-pro:brand-setup)?" — or proceed with defaults.
2. **Run technical translation scoring**: Execute `language-router.py --action score` with the original content, translated content, source language, target language, and do-not-translate terms. This produces four sub-scores: **length ratio** (translated content length versus expected length for the language pair — e.g., German typically runs 20-30% longer than English, Japanese typically runs 10-20% shorter; deviations beyond expected ranges indicate missing or added content), **formatting preservation** (markdown structure, HTML tags, merge tags like {{first_name}}, UTM parameters, and link structures survived translation intact), **key term consistency** (every do-not-translate term from the brand profile and any additional specified terms appears exactly as specified in the translation), and **placeholder integrity** (all variables, template tokens, and dynamic content markers are present and correctly positioned in the translated version).
3. **Run content quality evaluation**: Execute `eval-runner.py --action run-quick` on the translated content alone, scoring it as standalone content in the target language. This assesses structural quality, readability for the target audience, completeness, and coherence — catching cases where a translation is technically accurate but reads poorly as native content. The eval-runner scores content against the plugin's standard quality dimensions regardless of whether it was translated or originally authored.
4. **Run brand voice consistency check**: Execute `brand-voice-scorer.py --brand {slug} --text "{translated_content}"` to score how well the translated content matches the brand's voice profile. Brand voice should survive translation — the brand should sound recognizably like itself in every language, adapted for local expectations but maintaining its core personality dimensions (formality, energy, humor, authority). Score the translated content against the same voice dimensions as the original to detect voice drift introduced during translation.
5. **Check compliance for target market**: Based on the target language-region code, identify applicable regulatory requirements from `skills/context-engine/compliance-rules.md`. For EU languages: GDPR consent language, cookie consent, right-to-erasure references. For hi-IN and other Indian languages: DPDPA compliance. For pt-BR: LGPD. For ko-KR: PIPA. For ja-JP: APPI. For en-US: CAN-SPAM, CCPA/CPRA where applicable. Verify that required compliance elements are present and correctly localized in the translated content — not just copied in English. Score as compliant, partially compliant (elements present but not fully localized), or non-compliant (required elements missing).
6. **Compute multilingual composite score**: Calculate the weighted composite score across all four dimensions — translation technical score (40% weight, reflecting the core accuracy of the translation), content quality score (25% weight, assessing readability and structural quality in the target language), brand voice score (20% weight, measuring voice consistency across languages), and compliance score (15% weight, verifying regulatory requirements for the target market). Each dimension is scored 0-100, and the composite is the weighted average. If original content is not provided (skipping technical translation scoring), redistribute weights: content quality 40%, brand voice 35%, compliance 25%.
7. **Classify the result**: Based on the composite score, assign a clear action classification — **85 and above: Publish Ready** (content meets quality standards for the target market, no blocking issues, can proceed to publishing workflow), **70-84: Native Speaker Review Recommended** (content is functional but has quality gaps that a native speaker should review and correct before publishing — list the specific issues requiring review), **Below 70: Re-translate** (content has significant quality issues that spot corrections cannot fix — recommend re-translation with specific guidance on what went wrong and which dimensions need the most improvement).
8. **Generate improvement suggestions**: For any dimension scoring below its threshold (technical < 85, content quality < 80, brand voice < 75, compliance < 100), generate specific, actionable improvement suggestions. For technical issues: cite the exact problem (e.g., "Do-not-translate term 'BrandName Pro' was translated to 'BrandName Profi' on line 3 — must appear as 'BrandName Pro'"). For voice issues: cite the dimension and specific text (e.g., "Formality is at 8 in the translati