question:
What is our average exchange rate (vs refund) on the platform? How has it trended month-over-month, and what is the revenue impact?

datasets:
- loop_merchants.csv (50 rows): Merchant master with merchant_id, merchant_name, industry, plan_tier, integration_type, live_date, is_active, default_return_window_days
- loop_return_events.csv (25500 rows): Individual return item resolutions with event_id, return_id, merchant_id, event_date, order_date, customer_id, product_category, item_price_usd, quantity, return_reason, condition_received, resolution_type, resolution_value_usd, exchange_order_value_usd, is_test_order, is_fraudulent
- loop_monthly_summary.csv (270 rows): Pre-aggregated monthly metrics with month, merchant_id, total_return_items, total_return_value_usd, exchange_count, refund_count, exchange_rate_pct, revenue_retained_usd, avg_processing_days

expected answer:
Exchange Rate Trend (Sep 2025 - Feb 2026):
- Sep 2025: 38% exchange rate (baseline)
- Oct 2025: 40% (+2pt MoM)
- Nov 2025: 43% (+3pt MoM, peak season starts)
- Dec 2025: 45% (+2pt MoM, peak volume month)
- Jan 2026: 46% (+1pt MoM)
- Feb 2026: 48% (+2pt MoM, current)

Total improvement: +10 percentage points from Sep to Feb.

Volume by Month: Sep: ~2,900 events → Nov: ~5,200 → Dec: ~6,400 (peak) → Feb: ~2,900. Peak season (Nov-Dec) had 2x volume but still improved exchange rate.

MoM Trend: Improvement decelerated slightly mid-period (+3pt → +2pt → +1pt) but recovered (+2pt in Feb). Consistent positive trajectory throughout.

Revenue Impact: Each percentage point shift toward exchanges retains ~$5,800 in monthly revenue platform-wide (based on avg ~3,500 countable returns/month at $166 avg exchange value). The 10pt improvement represents ~$58K/month in additional retained revenue vs Sep baseline.

Note: The exchange_rate_pct column in loop_monthly_summary.csv is a TRAP - it only counts exchange_same_item resolutions (~15pt lower than actual). The correct calculation must include ALL exchange types: exchange_same_item, exchange_diff_item, exchange_upsell, exchange_downgrade, and warranty_replacement. Must also exclude is_test_order=true and is_fraudulent=true rows.
