# Methodology note

## Purpose

The control tower provides one governed chain from point-of-sale identity and market sell-out submission to trusted actuals, one-to-three-year plans and executive consumption.

## Data model

- 12 markets, 240 current POS, 262 SCD2 versions and 48 SKUs.
- 45,891 raw sell-out rows and 45,708 unique curated rows across January 2025-June 2026.
- Synthetic sources for SAP, Data Entry v2, retailer feeds, stock, SimpliField, Anaplan and Newviz.
- Raw, curated and mart layers with explicit grain and source ownership.

## Core formulas

- Net sell-out EUR = (gross local - discount local - returns local) / local-per-EUR FX.
- Coverage = distinct reporting POS / eligible active POS.
- Ending active POS = beginning POS + openings - closures.
- Reconciliation delta = (GCP EUR - Newviz EUR) / GCP EUR.
- WAPE = sum absolute forecast error / sum actual.
- Bias = sum(forecast - actual) / sum actual.
- Approved plan = baseline + network uplift + baseline × (market override + category-mix override + central challenge).

## Controls

Eight sell-out rules cover uniqueness, completeness, POS/product reference integrity, non-negative sales, arithmetic consistency, FX consistency and M+3 timeliness. Critical failures block publication. Corrections require an approved record-level log. POS attributes use SCD Type 2 with one current version and no date overlap. Consumption refresh requires an absolute GCP-Newviz delta at or below 0.10%.

## Forecast

The Gradient Boosting baseline uses market, category, calendar seasonality, active POS, lag-1, trailing-three-month mean and lag-12. March-June 2026 is held out. The model reaches 9.21% WAPE and +2.34% bias. Long-range recursion blends 35% model signal with a 65% prior-year seasonal anchor grown by 5.5%.

## Results

- €112.50m synthetic sell-out analysed.
- Coverage increases from 80.34% to 97.81%.
- Latest on-time submission reaches 97.68%.
- Record-level critical quality increases from 98.11% to 100.00% after controlled remediation.
- The latest GCP-Newviz delta moves from 0.1761% before replay to approximately 0.0000% after replay.
- 30 openings and 15 closures are planned through 2029.
- 12 automated tests, 30 UAT cases, 240 SCD2 controls and 9 Excel release checks pass.

All data is synthetic. The project is independent and has no affiliation with or access to any company system.