A multi-brand service can receive devices from several manufacturers, but attribution must be limited by identification confidence and privacy controls.
Neutral intake does not mean unstructured data
The collection point can accept an eligible product category without promoting a specific manufacturer. Intake rules, excluded items and custody records still need to be consistent across brands.
Confidence belongs beside every attribution
A visual match, readable label and verified product code do not provide the same level of certainty. Manufacturer-level reporting should include an identification method, confidence state and an unclassified bucket.
Share the minimum useful dataset
Counts, weight, condition and treatment evidence can be useful without direct household identifiers. Access rights, retention periods and permitted uses should be fixed before any manufacturer-specific export is offered.
Separate observation from interpretation
The operator can provide observed counts, weight, timestamps, condition and custody records. A manufacturer or assurance party may then apply its own methodology to calculate recovery rate, avoided emissions or target contribution. Preserve the raw observation, method version and derived result as separate fields. When one organisation silently controls both the measurement and the benefit interpretation, correction becomes difficult and an estimate can harden into reported performance.
Use explicit identification states
Assign each item one of four states: verified code, verified label, visual indication only or unclassified. A verified code must match an agreed identifier and reference table; a verified label requires a legible manufacturer or product marking; visual indication is a review queue, not manufacturer data; unclassified remains in the programme total but outside any manufacturer subtotal. Retain the identification method, reference version and reviewer where manual judgment was used. Do not convert confidence into a precise percentage unless it has been validated against a labelled sample.
Minimum event and export schema
For each event record event ID, stream ID, item or aggregate ID, event type, timestamp, controlled location, custodian role, quantity and unit, identification state, condition, evidence reference and exception code. Keep household identity outside the operational export. A manufacturer file should contain only permitted rows, the applicable identification states, field definitions, units, reporting period, corrections and an unclassified total. A dashboard total is not a chain-of-custody record unless the underlying event history is retained.
Export-control checklist
Before a manufacturer-specific export, confirm purpose, lawful basis where personal data could be involved, recipient, allowed fields, retention period, onward-use restriction, correction method and deletion or archival rule. Confirm separately that the manufacturer name may be used to label returned observations; data receipt does not imply programme participation or endorsement. Stop the export if direct household identifiers are present, the identification state is omitted or rows cannot be reconciled to the neutral programme total.
Primary sources and limits
GS1 EPCIS 2.0.1 provides a normative model for event data and contextual master data: https://ref.gs1.org/standards/epcis/2.0.1/. GDPR Article 5 states purpose limitation, data minimisation, accuracy, storage limitation and accountability for personal data in its scope: https://eur-lex.europa.eu/eli/reg/2016/679/oj. EPCIS does not decide waste classification or privacy law, and GDPR does not automatically apply to every dataset or jurisdiction. The four identification states above are an operating control proposed by UAMKT, not a GS1 confidence standard.