lunarunes-semantic-group-classifier

Purpose

Classify text into exactly one LunaRunes semantic group while preserving a full nine-way distribution and explicit dispute metadata.

This Skill is intended as a trial layer before the classifier is promoted into shared LOC runtime modules.

Governance scope — highest priority

This Skill operates inside the LunaRunes cultural/system context. LunaRunes Canon governs LunaRunes; it is not a universal standard for judging other cultures, language systems, author systems, or cultural frameworks.

Governance is culture-scoped:

1. Identify the culture/system being analyzed before applying governance rules.
2. Use that culture/system's own established governance when it exists.
3. Do not use LunaRunes Canon to declare another culture correct, incorrect, valid, invalid, superior, or inferior.
4. When comparing cultures/systems, describe differences, context, provenance, history, and disputed interpretations without converting one culture's rules into another culture's authority.
5. If the target culture/system has no established governance, analyze descriptively and preserve uncertainty; do not invent governance on its behalf.
6. A Skill executes applicable governance rules; a Skill does not create governance authority by itself.

For this classifier, the requested output is specifically a LunaRunes semantic projection of supplied text. Therefore the LunaRunes Canon may determine the resulting LunaRunes group classification, but that classification must not be presented as a judgment on the source culture itself.

Non-negotiable constraints

1. No external API calls.

   • Do not call external LLM APIs.
   • Do not call embedding APIs.
   • Do not require an API key.
   • Analyze the supplied text directly with the current LunaRunes semantic rules and available local/project context.

2. Semantic classification, not literal rune matching.

   • Do not classify a segment merely because a rune character or keyword appears.
   • Rune names, keywords and reverse keywords are evidence only.
   • Determine what the sentence or paragraph is actually about before assigning a group.

3. Exactly one final group.
   The final group must be one of:

   • 靈魂
   • 連結
   • 生命
   • 自然
   • 礦物
   • 元素
   • 秩序
   • 無序
   • 特殊

4. 特殊 is the default value.

   • Start conservatively from 特殊.
   • A general group must earn classification through sufficient semantic evidence.
   • Do not force text into one of the eight general groups just to increase coverage.

5. Distribution is required.

   • Estimate a relative nine-way distribution for every semantic unit.
   • The distribution preserves uncertainty; it does not create multiple labels.
   • Normalize the distribution to approximately 1.0.

6. Close scores create a dispute record.

   • If the top groups are semantically close, set disputed: true.
   • Still output one current group; keep 特殊 when evidence is insufficient.
   • Preserve alternative candidates and explain the conflict.
   • Do not discard disputed samples; they are governance data for later calibration.

7. Thresholds are provisional.

   • Do not claim a permanent Canon threshold.
   • Use qualitative judgment and, when useful, a temporary score-gap heuristic.
   • Mark uncertain cases rather than pretending false precision.

8. Current LunaRunes Canon overrides historical LunaRunes meanings.
   Apply current semantic governance within the LunaRunes scope, including:

   • 水 = Water, not Flow.
   • 流動 belongs primarily to 氣, not 水.
   • 氣 = Air.
   • 暗 = Shadow.
   • 空 = Space.
   • 無 = Blank / all possibilities.
   • 虛 = Void.
   • 玄 = Chaos, never Mystery.
   • 誤 = Error.
   • 辰 focuses on period; 時 on time; 緣 may carry timing/opportunity.
   • Do not map ordinary disorder automatically to 玄.

Classification method

For each semantic unit:

1. Read the unit as a whole and identify its source culture/system when that context is available.
2. Identify its main subject and semantic role.
3. Distinguish literal wording, metaphor, modifier and actual semantic focus.
4. Compare the unit with the nine LunaRunes group domains as a LunaRunes semantic projection.
5. Use the current LunaRunes Canon, Spec, keyword relations, reverse-keyword relations and exclusion rules as evidence.
6. Build a nine-way distribution.
7. Apply disambiguation and exclusions.
8. Select exactly one final group.
9. Mark disputes and retain alternative candidates.
10. Explain why the winning LunaRunes group was chosen and why nearby groups were not, without turning the result into a judgment of the source culture.

Group subjectivity guide

These are first-pass LunaRunes group domains, not keyword lists. Always defer to current rune-level Canon when a LunaRunes boundary is unclear.

• 靈魂: inner self, spirit, memory, boundary, personal domain, reflection, core.
• 連結: direction and relationship operations such as connecting, maintaining, severing, separating, initiating, understanding and error in relation/context.
• 生命: living-process experience, body/mind state, emotion, love, language and rhythm as lived experience.
• 自然: growth and plant/natural-life structures such as root, seed, tree, flower, leaf, grass, fruit and branch.
• 礦物: material/mineral structure, geology, hardness, pressure, rarity, crystal/mineral/solid-material properties.
• 元素: elemental properties and actions of light, shadow, water, fire, wind, earth, thunder and air; obey rune-specific exclusions.
• 秩序: observable or locatable structure involving sun, moon, star, period, clarity, time, space and cause.
• 無序: uncertainty/non-linearity involving fortune, misfortune, blank possibility, dream, illusion, opportunity, void and result.
• 特殊: safe default and special layer. Do not automatically equate 特殊 with 玄 or 命.

Input

Accept either:

Plain text

A paragraph, article, lyric, note or other text.

JSON

If JSON is supplied, identify text-bearing fields and classify their textual content. Preserve source identifiers where available.

Example invocation concept:

Use lunarunes-semantic-group-classifier on article.json.

Output

Return structured JSON for each semantic unit and an aggregate summary.

Minimum per-unit shape:

{
  "source_id": "optional-source-id",
  "text": "...",
  "group": "連結",
  "distribution": {
    "靈魂": 0.10,
    "連結": 0.41,
    "生命": 0.05,
    "自然": 0.02,
    "礦物": 0.02,
    "元素": 0.04,
    "秩序": 0.29,
    "無序": 0.03,
    "特殊": 0.04
  },
  "disputed": true,
  "candidates": [
    {"group": "連結", "score": 0.41},
    {"group": "秩序", "score": 0.29}
  ],
  "reason": "核心主體是關係脈絡，但同時具有明顯因果結構。",
  "evidence": ["relationship semantics", "context maintenance"],
  "exclusions": ["不是單純時間描述，因此不以秩序為主分類"],
  "ruleset_version": "trial-0.1",
  "api_used": false
}

Aggregate summary:

{
  "analysis_mode": "lunarunes_semantic_group_classifier",
  "api_used": false,
  "ruleset_version": "trial-0.1",
  "unit_count": 120,
  "group_counts": {
    "靈魂": 0,
    "連結": 0,
    "生命": 0,
    "自然": 0,
    "礦物": 0,
    "元素": 0,
    "秩序": 0,
    "無序": 0,
    "特殊": 0
  },
  "disputed_count": 0,
  "disputed_units": []
}

Governance behavior

• Apply LunaRunes governance only to the LunaRunes projection/classification produced by this Skill.
• Never use a LunaRunes classification result to judge the correctness or legitimacy of the source culture/system.
• When a case is ambiguous, preserve the ambiguity instead of silently forcing a result.
• When a classification depends on a weak or provisional LunaRunes rule, say so.
• When a new recurring dispute pattern appears, recommend reviewing the relevant culture/system governance or LunaRunes semantic rules at the correct scope rather than patching individual outputs.
• Never rewrite Base66 meanings from classifier output alone.
• Treat disputed LunaRunes records as future calibration material.

Relationship to LOC

This Skill is a modular first-stage LunaRunes classifier:

Text
  ↓
Semantic units
  ↓
LunaRunes group classification
  ↓
Rune candidates
  ↓
Search / RAG / Graph / Culture

The primary benefit is to reduce the search space before rune-level analysis while keeping the result explainable and governable. The projection belongs to LunaRunes analysis; it does not transfer LunaRunes governance authority to the source culture.

Authoritative principle document

Use the current LunaRunes governance/Canon documents as the governing specification for LunaRunes classification. If a referenced historical document conflicts with current Canon or an explicit later governance decision, current governance takes precedence.