Winning Pattern Synthesis

Takes the outputs of the other three analysis skills and turns them into a ranked set of creative directions worth actually building — the step between "here's what we found" and "here's what to brief."

When to use this skill

• "What should we test next?"
• "Where's the whitespace vs competitors?"
• Any time competitor intel + own-creative diagnosis have both run and the user wants a single synthesized direction rather than two separate reports
• As the step before ad-brief-generator

Inputs

This skill doesn't fetch new data — it reasons over what's already been produced:

• Competitor patterns from competitor-ad-intelligence (angles, hooks, formats, scale signals)
• Own performance patterns from own-creative-diagnosis (what showed strong first-party results in the stated analysis window)
• Psychological grading from creative-psychology-hooks (mechanism, valence/intensity, micro-moment)

If one or more inputs are missing, say explicitly which are missing and whether to proceed with a partial synthesis or run the missing skill(s) first. Competitor-only or own-creative-only synthesis is possible but weaker — flag it as partial.

Step 1: Build the pattern matrix

Lay out every distinct pattern (angle × hook mechanism × format) found across both sources in one table, tagging each occurrence:

Pattern	Seen in competitors?	Seen in own strong performers?	Scale/strength signal

Step 2: Classify each pattern

• Cross-source overlap — the pattern has strong first-party results and competitor duration/spend/variation proxy evidence. This is the strongest observed overlap, but competitor performance remains unverified; recommend a controlled variation test rather than calling it proven.
• Competitor-backed test gap — competitors are running it with meaningful proxy signals, but the user hasn't tested it. This is an own-account test gap, not market whitespace or a guaranteed winner.
• Own-only market whitespace — has strong first-party results in the analysis window but was not observed in the competitor sample. It could be a differentiator, a sample-coverage gap, or an account-specific result — note all applicable interpretations.
• Neither — untested by anyone visible. Highest risk/highest potential differentiation; only surface if the user wants aggressive experimentation, not as a default recommendation.

Step 3: Rank by confidence and apply psychological lens

For each pattern surfaced, pull in the Baader mechanism / valence-intensity classification from creative-psychology-hooks if hooks for that pattern were graded. Present it as a testable hypothesis for why the pattern may attract attention, not a causal explanation of performance.

Rank output: cross-source overlap first, then competitor-backed test gaps, then own-only market whitespace. Cap at the top 5–7 directions unless the user asks for the full list — a synthesis that recommends everything recommends nothing.

Step 4: Flag conflicts

If a pattern is a competitor favorite but has performed poorly in the user's own account (or vice versa), don't average it out — call out the conflict directly and suggest a hypothesis for the divergence (audience difference, execution quality, offer mismatch).

Output Format

### Winning Pattern Synthesis: [Product/Category]

Inputs used: [competitor intel / own diagnosis / psych grading — note any missing]

Top directions:
1. [Pattern] — CROSS-SOURCE OVERLAP
   - Competitor signal: [brands, scale]
   - Own signal: [ad, metrics]
   - Mechanism: [Baader tag, valence/intensity]
   - Recommendation: [controlled variation test / new variant angle]

2. [Pattern] — COMPETITOR-BACKED TEST GAP
   - ...

Conflicts worth noting:
- [Pattern]: strong for competitors, weak in our account — possible reason: [hypothesis]

Not recommended right now:
- [pattern seen but low-confidence, and why]

Guardrails

• Never present competitor proxy evidence as verified performance or competitor-backed patterns as proven for the user's account.
• Don't let this skill turn into a restatement of the two input reports — its entire value is the cross-reference and ranking. If there's no real overlap or conflict to surface, say that plainly rather than padding the output.
• Hand off the final ranked list directly to ad-brief-generator — don't re-derive angles from scratch there.