Bridge built: Volume / Price / Mix / COGS over last 8 backtest weeks.
  Window: [202612, 202613, 202614, 202615, 202616, 202617, 202618, 202619]
  SKUs:   503
Plan = forecast.total (latest backtest row per SKU-week).
Plan price/cost = current erp_prices.avg_sell_price / erp_costs.cost_price
  (we don't have historical price/cost snapshots — only one valid_from per SKU).
Actual qty = erp_transactions.quantity; falls back to backtest_fa.actual.
Actual revenue = erp_transactions.tax_base (net of VAT).
Actual unit cost = erp_transactions.purchase_value / quantity.

Limitations:
  - No historical price snapshots → price_effect is realized-vs-current,
    not realized-vs-period-plan.
  - No historical cost snapshots → COGS effect compares realized cost
    vs the latest standard cost, not vs a prior-period cost.
  - Mix effect is the residual; assumes plan price/cost reflects 'plan mix'.

For next S&OP, Finance should provide:
  - monthly plan revenue + qty per SKU
  - cost snapshots (cost_price, valid_from) — one per cost change
  - price snapshots (avg_sell_price, valid_from) — one per price change