DATA PROVENANCE — Chilly's Bottles Meta Ads Audit
==========================================

Account:        Chilly's Bottles (act_864659590254750)
Currency:       GBP
Timezone:       Europe/London

3-LAYER ANALYSIS:
  Layer 1 (Scoring):  2025-12-02 to 2026-03-02 (3 months)
  Layer 2 (Trends):   2025-09-02 to 2026-03-02 (6 months, monthly)
  Layer 3 (YoY):      2024-09-02 to 2026-03-02 (18 months, monthly)

API:            Meta Marketing API v21.0
Pull method:    Insights-based entity discovery (spend > 0 filter)
                — 23 API calls in 54 seconds
Pull timestamp: 2026-03-03 ~16:55 UTC+1

LAYER 1 DATA (Scoring — 3 months):
  Account-level insights (default + 7d_click attribution)
  15 campaigns with spend > 0 (batch metadata via ?ids=)
  37 ad sets with spend > 0 (batch metadata with targeting)
  199 ads with spend > 0 (batch metadata)
  181 unique creatives
  91 daily insight rows
  14 placement breakdown rows
  2 pixels

LAYER 2 DATA (Trends — 6 months monthly):
  7 account-level monthly rows
  70 campaign-level monthly rows

LAYER 3 DATA (YoY — 18 months monthly):
  19 account-level monthly rows

ROAS CALCULATION:
  Source field:   action_values where action_type = "purchase"
  Formula:        purchase_value / spend
  NEVER used:     total action_values (includes non-purchase actions)
  Cross-check:    purchase_roas field from API (match: 0.0000 diff)

ATTRIBUTION:
  Default:        7d_click + 1d_view (Meta default)
  Click-only:     action_attribution_windows: ["7d_click"]
  View derived:   Default total - Click only
  Post-view:      41.8% of purchases are view-through (HIGH)

SANITY CHECKS (all passed):
  1. ROAS field validation:    PASS (API 7.5742x = Manual 7.5742x)
  2. AOV check:                PASS (GBP 35.66, expected GBP 25-55)
  3. Spend reconciliation:     PASS (campaign sum = account total, diff GBP 0.00)
  4. Attribution split:        41.8% post-view (notably high)
  5. False retargeting:        0 flagged, 0 ASC exceptions, 8 valid RTG

SCORING:
  Engine:         55-check weighted scoring (V3)
  Overall:        59.9/100 (Grade D — Poor)
  Categories:
    Pixel/CAPI:   37.5/100 (F) [LIMITED DATA — 2/10 evaluated]
    Creative:     52.7/100 (D) [10/15 evaluated]
    Structure:    70.0/100 (C) [9/18 evaluated]
    Audience:     66.7/100 (C) [4/8 evaluated]
    Funnel:       80.8/100 (B) [3/4 evaluated]

KEY FINANCIAL METRICS:
  Spend:          GBP 106,755.09
  Revenue:        GBP 808,582.73
  ROAS:           7.57x (VERY PROFITABLE)
  Click-only ROAS: 4.24x
  Purchases:      22,675
  AOV:            GBP 35.66
  CPA:            GBP 4.71
  Net Profit:     GBP 701,827.64

YOY CONTEXT:
  Q4 2024 vs Q4 2025: Spend -61.9%, ROAS +75.0%, CPA -40.8%
  Q1 2025 vs Q1 2026: Spend +25.2%, ROAS +177.0%, CPA -70.6%

SEASONALITY:
  Q1 CPA improved 38% vs Q4 (GBP 4 vs GBP 7)
  Last year the same period saw a CPA rise — this year's improvement
  against the seasonal pattern is a strong positive signal.

Bug fixes applied (from prior audits):
  - funnel.py M-AT1: Fixed double-counting (use 7d_click sub-field, not value)
  - funnel.py M-FA1: ASC campaigns excluded from false RTG flags
  - funnel.py: Only count purchase action_type (not omni_purchase)

Data pull optimization applied:
  - Use insights endpoint to discover entities with spend > 0
  - Batch-fetch metadata via ?ids= endpoint (50 IDs per call)
  - Reduced API calls from 200+ to 23
  - Fixed ad_id KeyError on paginated responses (filter with .get())