feat: add metadata-ai-readiness skill and enrich mart model YAML

[aboyalejandro]

Summary

• Enriched all 46 column descriptions across campaign_performance and daily_summary with [Business Purpose] and [Known Issues / Caveats] sections
• Applied dbt Agent Skills writing-documentation standards — no description restates its column name
• Documented COALESCE masking behavior, synthetic data uniformity, date gaps, and misleading zero values discovered via database profiling

AI Readiness Audit Report

campaign_performance

dbt Schema

Check	Status	Detail
Model description exists	PASS	Has grain, key metrics, business context
All 26 SQL columns in YAML	PASS	Full coverage
Descriptions pass writing-documentation	PASS (was FAIL)	All 26 columns now enriched with [Business Purpose] and caveats
Grain tests (composite: campaign_id + date)	FAIL	Both have not_null but no composite unique test — grain is unverified

Query Guidance

Check	Finding
Grain holds	PASS — 400 rows, 400 distinct (campaign_id, date) pairs
Date range	2025-12-01 to 2025-12-21 (21 days expected, 20 present)
Date gap	2025-12-20 missing — absent from source stg_campaigns_daily
Sessions constant	total_sessions = 110 for ALL 400 rows — likely synthetic/uniform data
Zero conversions	41/400 rows (10.3%) have 0 conversions → avg_order_value COALESCE'd to 0 (misleading: 0 ≠ "no data")
COALESCE masking	Session and conversion columns LEFT JOIN'd with COALESCE(x, 0) — zeros hide NULLs from missing joins
ROAS spread	0.066 to 6.20 — wide range, no outlier flagging
Channels	8 channels, 20 campaigns (2 paused: tiktok, linkedin)

---

daily_summary

dbt Schema

Check	Status	Detail
Model description exists	PASS	Has grain, key metrics, business context
All 20 SQL columns in YAML	PASS	Full coverage
Descriptions pass writing-documentation	PASS (was FAIL)	All 20 columns now enriched
Grain tests (date)	PASS	not_null + unique present

Query Guidance

Check	Finding
Grain holds	PASS — 20 rows, 20 distinct dates
Date gap	2025-12-20 missing — same source gap
Sessions constant	total_sessions = 2,200 every day (20 campaigns × 110)
Conversions constant	total_conversions = 150 every day — suspiciously uniform
NULLIF columns	budget_utilization, overall_conversion_rate, overall_roas, overall_cpa — 0 NULLs currently (no divide-by-zero days exist, but would produce NULLs if they did)
Spend range	$33,878 – $50,386/day

---

Summary

PASS: 9/10 | Auto-fixed: 3 (descriptions enriched) | Manual: 1 (composite unique test on campaign_performance)

Remaining manual action

campaign_performance needs a composite uniqueness test on its grain. Add to _marts.yml:

    data_tests:
      - dbt_utils.unique_combination_of_columns:
          combination_of_columns:
            - campaign_id
            - date

Test plan

• Verify _marts.yml parses without errors: dbt parse --profiles-dir . --project-dir .
• Confirm enriched descriptions render correctly in dbt docs
• Review flagged caveats against business knowledge (especially uniform session/conversion data)

🤖 Generated with Claude Code