Data Governance: The Master Data Is the Compliance System

Count the framework's data consumers: the technical documentation and declarations, the EPR registrations and filings in twenty-seven accents, the marketplace verifications, the deposit registries, the label registers, the customers' specification systems, the company's own sustainability reporting. Every one renders from the same underlying facts - what each pack is made of, what it weighs, what it contains, where it is placed. Where those facts live in one governed dataset, every rendering agrees and every change propagates once. Where they live in spreadsheets per department and per market, the renderings disagree on a schedule set by entropy, and the reconciliations of the audit guide become a permanent remediation programme.

This guide specifies the governance: fields, owners, changes, validation, history.

The core record per packaging component (not per product - components are where the framework counts) carries: identity (component, parent SKU, version), composition (materials and shares, substances flags, recycled content with its certification references), physical data (weights, dimensions), regulatory renderings (grade computation inputs, label set per market, deposit status), commercial placement (markets, channels, producers of record), and lifecycle (valid from/to, superseded-by links). The model's granularity is the recurring design decision: too coarse and the declarations approximate; too fine and maintenance collapses. The test is the framework's own questions - if a register's field cannot be answered from the record without a phone call, the record is too coarse.

Packaging data changes when reality does - a resin switch, a weight reduction, a supplier's formulation move - and the governance question is whether the change enters once, with its regulatory consequences assessed, or five times partially. The controlled pattern: changes proposed by the data's owners (packaging engineering for composition, procurement for supplier-driven changes), impact-assessed against the renderings (declarations to re-issue, registers to update, artwork to review), approved and versioned with effective dates, and propagated by interface rather than by email. The supplier-change clauses of the sector stations (notification before change) exist to feed this constitution; the constitution exists so their notices land somewhere with a process attached.

Two mechanical disciplines keep the dataset honest. Validation rules at entry and on schedule: weights that sum, shares that total, certificates in date, markets consistent with registrations, impossible combinations refused by the system rather than discovered by the audit. And the retained history: every rendering reproducible as at its date (the declaration of 2027 regenerable in 2031 from the data as it then stood), because authorities and customers ask historical questions of historical placings. Retention follows the documentation clocks of the conformity station; the dataset's history is part of the technical documentation in everything but name, and should be governed - backed up, access-controlled, format-migrated - with the same seriousness.

When packaging compliance fails publicly, the visible artefact is a pack or a fine; the actual failure, traced, is almost always a datum - wrong in the master, propagated faithfully everywhere, discovered by a reconciliation the business or the authority finally ran. Governance of that dataset is therefore not IT housekeeping at the edge of the programme; it is the programme's foundation course, and the resources it asks (field ownership, change control, validation) are small against every other station's budget. Fund the foundation first. The stations above it are only as true as it is.

A reconciliation found a flagship SKU declared at three different weights across the EPR filing, the deposit registry and a retailer's data feed - each defensible in isolation (gross, net, and an old specification's ghost), indefensible as a set. The governance fix was not a correction but a constitution: one weight field with a named owner and a definition, the renderings re-pointed to it, the ghost retired with its history retained. The incident's cost was a fortnight of clerical repair; its value was the discovery, made internally, that the dataset had no constitution at all. Most portfolios learn this from an authority. The lucky ones learn it from their own audit, which is why the audit guide runs the reconciliations first.