amcs-validator

AMCS Validator

Evaluates composed artifacts (lyrics, style, producer notes) against the blueprint's scoring rubric to determine if the composition meets quality thresholds. If scores fall below the threshold (min_total < 0.85), the workflow transitions to the FIX loop.

When to Use

Invoke this skill after COMPOSE completes. This is the quality gate before RENDER.

Input Contract

inputs:
  - name: lyrics
    type: string
    required: true
    description: Complete lyrics with section markers
  - name: style
    type: amcs://schemas/style-1.0.json
    required: true
    description: Validated style specification
  - name: producer_notes
    type: amcs://schemas/producer-notes-1.0.json
    required: true
    description: Production arrangement and mix guidance
  - name: blueprint
    type: amcs://schemas/blueprint-1.0.json
    required: true
    description: Genre-specific rules and scoring rubric
  - name: composed_prompt
    type: amcs://schemas/composed-prompt-0.2.json
    required: false
    description: Optional composed prompt for additional validation
  - name: seed
    type: integer
    required: true
    description: Determinism seed (use seed+5 for this node)

Output Contract

outputs:
  - name: scores
    type: object
    description: |
      Scoring breakdown:
      - total: Weighted average score (0-1)
      - hook_density: Hook repetition score (0-1, target ≥ 0.7)
      - singability: Syllable/meter consistency (0-1, target ≥ 0.8)
      - rhyme_tightness: Rhyme scheme adherence (0-1, target ≥ 0.75)
      - section_completeness: Required sections present (0-1, target 1.0)
      - profanity_score: Policy compliance (0-1, target 1.0 for clean)
  - name: issues
    type: array[string]
    description: List of specific failures (e.g., "Low hook density: 0.5 (target 0.7)")
  - name: pass
    type: boolean
    description: True if scores meet rubric thresholds (total ≥ min_total)

Determinism Requirements

• Seed: run_seed + 5 (for any probabilistic scoring, if needed)
• Temperature: N/A (rule-based scoring, no LLM generation)
• Top-p: N/A
• Retrieval: None
• Hashing: Hash scores object for provenance

Constraints & Policies

• min_total threshold: Default 0.85 (from blueprint.eval_rubric.thresholds.min_total)
• Hook density target: ≥ 0.7 (chorus hooks repeated, memorable)
• Singability target: ≥ 0.8 (consistent syllable counts, natural phrasing)
• Rhyme tightness target: ≥ 0.75 (rhyme scheme followed)
• Section completeness target: 1.0 (all required sections present)
• Profanity target: 1.0 for clean songs (0.9 if explicit allowed)
• Rubric weights: Must sum to 1.0, used for weighted total

Implementation Guidance

Step 1: Load Rubric from Blueprint

Extract rubric configuration:

rubric = blueprint["eval_rubric"]
weights = rubric["weights"]
thresholds = rubric["thresholds"]
min_total = thresholds["min_total"]  # Default: 0.85

Step 2: Evaluate Hook Density

Definition: Percentage of lines that contain hook phrases from the chorus.

Algorithm:

1. Extract chorus section from lyrics
2. Identify hook phrases (repeated phrases ≥ 3 words)
3. Count total lines across all sections
4. Count lines containing hook phrases
5. Score = (hook_lines / total_lines)

Target: ≥ 0.7

Example:

Chorus:
Family time is what we need  <-- hook
Love and joy in every deed   <-- hook

Verse:
Family time is what we need  <-- hook repeated
...

Score: 3 hook lines / 16 total lines = 0.1875 → LOW

Issues:

• If score < 0.7: "Low hook density: {score:.2f} (target 0.7)"

Step 3: Evaluate Singability

Definition: Consistency of syllable counts and natural phrasing.

Algorithm:

1. For each section, extract lines
2. Count syllables per line (use pyphen or simple vowel counting)
3. Compute standard deviation of syllable counts within section
4. Score = 1.0 - (stddev / mean_syllables)
5. Average across all sections

Target: ≥ 0.8

Example:

Verse:
Gathering 'round on Christmas Eve (9 syllables)
The kids decorate, we all believe (9 syllables)
Family time is what we need (8 syllables)

Stddev = 0.47, Mean = 8.67
Score = 1.0 - (0.47 / 8.67) = 0.95 → PASS

Issues:

• If score < 0.8: "Weak singability: {score:.2f} (target 0.8) - inconsistent syllable counts"

Step 4: Evaluate Rhyme Tightness

Definition: Adherence to the intended rhyme scheme (ABAB, AABB, etc.)

Algorithm:

1. Parse lyrics.constraints.rhyme_scheme from SDS (e.g., "ABAB")
2. Extract end words from each line in verse/chorus
3. Check phonetic similarity (use pronouncing library or simple suffix matching)
4. Score = (matching_rhymes / expected_rhymes)

Target: ≥ 0.75

Example:

Rhyme scheme: ABAB
Verse:
Gathering 'round on Christmas Eve (A)
The kids decorate, we all believe (B) ✓ rhymes with A
Family time is what we need (A) ✓ rhymes with A
Love and joy in every deed (B) ✓ rhymes with B

Score = 4/4 = 1.0 → PASS

Issues:

• If score < 0.75: "Weak rhyme tightness: {score:.2f} (target 0.75) - rhyme scheme not followed"

Step 5: Evaluate Section Completeness

Definition: All required sections from blueprint are present in lyrics.

Algorithm:

1. Get required_sections from blueprint (e.g., ["Verse", "Chorus", "Bridge"])
2. Extract section markers from lyrics (e.g., [Verse], [Chorus])
3. Check if all required sections present
4. Score = (present_sections / required_sections)

Target: 1.0

Example:

Required: ["Verse", "Chorus", "Bridge"]
Present: ["Intro", "Verse", "Chorus", "Verse", "Chorus"]
Missing: ["Bridge"]

Score = 2/3 = 0.67 → FAIL

Issues:

• If score < 1.0: "Missing required sections: {missing_sections}"

Step 6: Evaluate Profanity Score

Definition: Compliance with profanity policy based on explicit flag.

Algorithm:

1. Get banned_terms from blueprint.rules
2. Check constraints.explicit from SDS
3. Scan lyrics for banned terms (case-insensitive)
4. If explicit=false and banned terms found: score = 0.0
5. If explicit=true: score = 0.9 (allowed but noted)
6. If clean: score = 1.0

Target: 1.0 for clean, 0.9 for explicit allowed

Example:

Explicit: false
Banned terms: ["damn", "hell"]
Lyrics: "What the hell is going on?"

Score = 0.0 → FAIL
Issue: "Profanity detected (explicit=false): hell"

Issues:

• If explicit=false and banned terms found: "Profanity detected (explicit=false): {terms}"

Step 7: Compute Weighted Total Score

Algorithm:

1. Multiply each score by its weight from rubric
2. Sum weighted scores
3. Total = (hook_density * w1) + (singability * w2) + (rhyme * w3) + (section * w4) + (profanity * w5)

Example:

weights = {
  "hook_density": 0.25,
  "singability": 0.25,
  "rhyme_tightness": 0.20,
  "section_completeness": 0.20,
  "profanity_score": 0.10
}

scores = {
  "hook_density": 0.65,
  "singability": 0.90,
  "rhyme_tightness": 0.80,
  "section_completeness": 0.67,
  "profanity_score": 1.0
}

total = (0.65 * 0.25) + (0.90 * 0.25) + (0.80 * 0.20) + (0.67 * 0.20) + (1.0 * 0.10)
      = 0.1625 + 0.225 + 0.16 + 0.134 + 0.10
      = 0.7815 → FAIL (< 0.85)

Step 8: Determine Pass/Fail

if total_score >= min_total:
    pass_validation = True
    issues = []  # No critical issues
else:
    pass_validation = False
    # Collect all failing criteria

Step 9: Build Issues List

Collect all failing criteria with specific thresholds:

issues = []
if hook_density < 0.7:
    issues.append(f"Low hook density: {hook_density:.2f} (target 0.7)")
if singability < 0.8:
    issues.append(f"Weak singability: {singability:.2f} (target 0.8)")
if rhyme_tightness < 0.75:
    issues.append(f"Weak rhyme tightness: {rhyme_tightness:.2f} (target 0.75)")
if section_completeness < 1.0:
    issues.append(f"Missing required sections: {missing_sections}")
if profanity_score < 1.0:
    issues.append(f"Profanity detected: {banned_terms_found}")

Step 10: Return Validation Results

{
  "scores": {
    "total": 0.7815,
    "hook_density": 0.65,
    "singability": 0.90,
    "rhyme_tightness": 0.80,
    "section_completeness": 0.67,
    "profanity_score": 1.0
  },
  "issues": [
    "Low hook density: 0.65 (target 0.7)",
    "Missing required sections: ['Bridge']"
  ],
  "pass": false,
  "_hash": "abc123..."
}

Examples

Example 1: Passing Validation

Input:

{
  "lyrics": "[Verse]\nFamily time is what we need\n...\n[Chorus]\nFamily time is what we need\nLove and joy in every deed\n...\n[Bridge]\nTogether we can share the light\n...",
  "style": {...},
  "producer_notes": {...},
  "blueprint": {
    "rules": {"required_sections": ["Verse", "Chorus", "Bridge"]},
    "eval_rubric": {
      "weights": {"hook_density": 0.25, "singability": 0.25, "rhyme_tightness": 0.20, "section_completeness": 0.20, "profanity_score": 0.10},
      "thresholds": {"min_total": 0.85}
    }
  }
}

Output:

{
  "scores": {
    "total": 0.92,
    "hook_density": 0.85,
    "singability": 0.95,
    "rhyme_tightness": 0.90,
    "section_completeness": 1.0,
    "profanity_score": 1.0
  },
  "issues": [],
  "pass": true
}

Example 2: Failing Validation (Low Hook Density)

Input:

{
  "lyrics": "[Verse]\nWalking through the snowy night\n...\n[Chorus]\nChristmas time is here again\n...",
  "blueprint": {...}
}

Output:

{
  "scores": {
    "total": 0.78,
    "hook_density": 0.45,
    "singability": 0.88,
    "rhyme_tightness": 0.82,
    "section_completeness": 1.0,
    "profanity_score": 1.0
  },
  "issues": [
    "Low hook density: 0.45 (target 0.7)"
  ],
  "pass": false
}

Common Pitfalls

1. Weights Not Summing to 1.0: Validate weights before computing total score
2. Missing Sections Not Detected: Ensure section parsing handles variations ([Chorus], [CHORUS], [Chorus 1])
3. **False Positives on Profani

---

Content truncated.