A pilot should fail on the same defects as a real shipment

The pilot begins with a buyer question and a claim card, not a software shortlist. ISO describes chain of custody as applicable across materials and products, while ISEAL explains that custody models transfer materials and associated attributes through certified supply chains.[S01][S02] GFT would therefore state the material, claimed attribute, geographic and temporal boundary, product unit, custody model, claimant and evidence owner before selecting identifiers or infrastructure. The question might be whether a defined garment batch can carry a bounded statement about the country in which its cotton was grown. That question does not automatically include customs origin, certified content, labour conditions or environmental performance.

A suitable first chain is narrow enough to reconstruct and real enough to contain consequential handovers. The proposed scope would name one product family, the participating receiving and processing boundaries, the relevant transformations and the final dispatch unit. It would include at least one conversion and one custody transfer, because a pilot confined to a single stock ledger cannot test genealogy across organisational boundaries. It would also state which activities remain outside the claim. Thread, trims, packaging and non-cotton components should not disappear into an ambiguous whole-product statement. The claim card must make those exclusions visible to the reviewer.

Participation is an entry condition rather than an assumption. Each proposed evidence holder would need to agree which original records can be inspected, which personal or commercially sensitive fields require restriction and who can explain an exception. GFT would not treat a verbal expression of interest as evidence access. The pilot charter should identify the accountable operating role, evidence approver, data steward and independent reviewer for every boundary. If a necessary upstream holder cannot provide admissible evidence, the team should narrow the claim or stop at that boundary rather than fill the gap with a default value.

The custody model must match the intended wording. A physically segregated chain and a quantity-allocation chain can produce different defensible statements even when they use the same event fields. The pilot charter should name the selected model, its scheme or contract version, permitted mixtures and the treatment of unknown inputs. It should also specify whether the observation unit is a bale group, production lot, fabric roll, cutting batch or garment batch. A serialised garment can serve as a query key without implying that every earlier operation was observed at item level.

Initial volumes, durations and performance thresholds remain illustrative planning assumptions until the sites agree them. They should appear in a scope register with an owner and a reason, never as silent promises in a demonstration. The register should distinguish rehearsal fixtures from live evidence, and proposed release wording from wording approved for actual sale. During rehearsal, public-facing examples should carry a visible synthetic status. A real origin claim requires actual supply evidence and the ordinary release process, regardless of how convincing the rehearsal appears.

Scope control is itself a test. Adding another supplier, product family or claim during the exercise changes the expected genealogy and may invalidate earlier comparisons. GFT recommends a versioned scope baseline and a change record stating the reason, affected fixtures and required regression tests. The reviewer should be able to determine exactly which chain was evaluated. Expansion before the initial chain closes its adverse cases would conceal rather than resolve uncertainty.

Ordinary transactions provide the control against which injected faults become meaningful. GS1 EPCIS provides visibility-event semantics, including transformations and aggregations, while the Core Business Vocabulary supplies shared business meanings used within those events.[S10][S11] GFT would translate those concepts into a proposed fixture set covering receipt, storage, conversion, split, shipment, return and balance close. The mapping is a design choice to be tested. It is not a claim that every partner already uses EPCIS or that a particular scheme requires this implementation.

Each fixture needs an independently prepared expectation. Before the software processes a transaction, the test owner should record the expected input identifiers, output identifiers, custody state, quantity effect, evidence references and admissible claim. Another person should be able to prepare that expectation from the fixture narrative without seeing the system's answer. This separation prevents a demonstration from becoming circular, where the same event interpretation both produces and validates the result. The expected genealogy and expected quantity ledger are acceptance artefacts in their own right.

A receiving fixture should distinguish the physical observation from its later entry. The proposed record identifies the supplier, receiving site, material unit, observed quantity and basis, event time, recording time, actor, transport reference and source evidence. A storage fixture preserves the material identity while changing its location or custody status. A conversion fixture consumes declared inputs and creates declared outputs. A split changes the division of units without pretending that a new production process occurred. These distinctions give the reviewer a physical explanation for every digital transition.

The shipment fixture should pair dispatch and receipt rather than treat an invoice as proof of both. The return fixture should identify the original movement and the condition of the returned material, then apply the appropriate claim and quantity treatment. A balance-close fixture should reconcile opening quantity, receipts, consumption, outputs, returns, recorded losses and closing quantity on a compatible basis. Where a correction is required, the expected history retains the original observation and its authorised correction. Deleting the inconvenient event would remove the very evidence the pilot is meant to preserve.

Normal does not mean idealised beyond recognition. The reference set should include ordinary variations such as a document arriving after the material, a permitted roll split and an authorised return. Each variation needs a stated expectation so that the team can distinguish a supported exception from an uncontrolled defect. The pilot should not inject every fault at once. A stable reference history establishes that the basic chain works before a challenge case shows which control fails under pressure.

Synthetic fixtures are useful at this stage because their complete truth is known by construction. They allow the test owner to specify exactly which inputs were consumed, which quantities should remain and which claim should survive. That certainty belongs only to the exercise. When live capture begins, the reviewer must replace assumed truth with inspected evidence. A successful synthetic genealogy is not an origin certificate, and a balanced synthetic ledger is not proof of physical stock.

The reference history also establishes the query cases used later. A reviewer should be able to ask which inputs support a garment batch, which outputs consumed a bale group, which evidence justified a release and which corrections changed the record. The expected answers should include absent and restricted information, not only successful paths. By defining these queries before fault injection, the pilot can test whether the system preserves a useful history when the answer becomes uncomfortable.