MasteryMade · Intelligence PRD

PRD 5: Competitor Intelligence + Marketing Rubric

5.1 Purpose

Two lenses sharing data, different outputs. Competitor Intelligence: what others are doing, where the gaps are. Marketing Rubric: what works, what doesn't, what will work next. Both consume Gate 4 (competitors) + Gates 2/3 (expert's own content for comparison). Neither produces clone training data — they inform marketing and positioning only.

5.2 competitor-intel-analyzer Skill

Input

{ "entity_id": "UUID of expert", "competitor_ids": ["UUID array — Gate 4"], "analysis_depth": "quick | standard | deep" }

Output: Competitive Landscape Report

{
  "positioning_map": {
    "axes": { "x": "price_point", "y": "depth_of_engagement" },
    "placements": [{ "entity":"", "x":7, "y":4, "positioning_statement":"" }],
    "white_space": ["underserved positioning areas"]
  },
  "gap_analysis": {
    "topics_no_one_covers": [], "formats_no_one_uses": [],
    "audiences_no_one_targets": [], "objections_no_one_addresses": []
  },
  "ad_hook_taxonomy": {
    "hook_categories": [{ "category":"pain_agitation", "examples":[], "frequency":0.65, "engagement":"high" }],
    "winning_patterns": [], "saturated_patterns": [], "untested_patterns": []
  },
  "content_cadence": { "per_competitor": [{ "name":"", "posts_per_week":{}, "primary_format":"", "engagement_rate":0.0 }] },
  "offer_comparison": [{ "competitor":"", "offers":[{"name":"","price":"","format":""}], "value_ladder":"" }],
  "strategic_recommendations": { "positioning_opportunity":"", "content_gap":"", "offer_differentiation":"", "hooks_to_test":[] }
}

Ad Hook Taxonomy Classification

Categories: pain_agitation, curiosity_gap, social_proof, authority, before_after, contrarian, question_lead, story_open. Each ad classified by hook type + CTA type (click, sign up, buy, book, download) + creative format (video, image, carousel). Timeline analysis tracks launch date, run duration, seasonal patterns.

5.3 Marketing Rubric Scoring Engine

Schema: content_scores

CREATE TABLE content_scores (
  id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
  content_id UUID NOT NULL REFERENCES content(id),
  entity_id UUID REFERENCES entities(id),
  hook_score FLOAT, hook_analysis TEXT,
  body_score FLOAT, body_analysis TEXT,
  cta_score FLOAT, cta_analysis TEXT,
  format_score FLOAT, format_analysis TEXT,
  overall_score FLOAT,
  actual_engagement JSONB,
  predicted_vs_actual FLOAT,
  rubric_version INT NOT NULL DEFAULT 1,
  scored_at TIMESTAMPTZ DEFAULT now()
);

Scoring Weights

Hook: 30% (disproportionately important — if they don't stop scrolling nothing else matters). Body: 25% (value density, story structure, framework usage, proof). CTA: 25% (clarity, motivation alignment, friction, urgency). Format: 20% (platform optimization, production quality, length, trend alignment).

Hook criteria:

Pattern interrupt, curiosity gap, relevance signal (self-selects audience), emotional trigger, specificity.

Winner/Loser Pattern Detection

Batch score all content for entity + competitors. Top 20% = winners, bottom 20% = losers. Extract differentiating patterns. Output: specific, actionable differences.

Rubric Refinement Loop

Performance data re-ingests (PRD 8 feedback) → compare predicted score vs actual engagement → if systematic bias → adjust weights → increment rubric_version → re-score recent content → log to registry_changelog.

5.4 Synthetic Audience Testing

ICP Avatar Construction

Built from Module 7 (pattern recognition) + entity metadata:

{
  "avatar_name": "Stressed Professional Sarah",
  "demographics": { "age_range":"35-45", "income":"$80K-150K", "role":"mid-management" },
  "psychographics": { "pain_points":[], "desires":[], "beliefs":[], "objections":[] },
  "media_consumption": { "platforms":[], "content_preferences":[], "scroll_behavior":"" }
}

Process: Generate 3-5 ICP avatars per expert. Per content piece: Claude role-plays each avatar reading content. Evaluate: stop scrolling? Read full? Click CTA? Share? Aggregate into synthetic engagement prediction. Compare to rubric score. When actual performance arrives, validate both against reality.

5.5 Cross-Portfolio Aggregation

Patterns discovered for one expert inform all others. Group content scores by hook type across ALL experts/niches. If variance < 0.2 → tag as universal (works everywhere). If high variance → tag as niche-specific. Same for format, CTA, body patterns. Output: cross-portfolio insight report stored as Gate 1 content.

5.6 Acceptance Criteria

• competitor-intel-analyzer running end-to-end for at least one expert
• Positioning map with accurate competitor placements
• Gap analysis identifying 3+ actionable gaps
• Ad hook taxonomy classifying hooks into categories
• content_scores table populated for test content
• Scoring algorithm producing consistent, defensible scores
• Winner/loser pattern detection identifying differentiating patterns
• At least one ICP avatar constructed from extraction data
• Synthetic audience testing producing predictions
• Rubric refinement loop triggered by simulated performance data
• Cross-portfolio aggregation running across 2+ experts