Better data means representative data, not simply more data

GHG Protocol recommends screening before committing to detailed collection, considering emissions significance, reduction opportunities and business relevance. Its calculation guidance permits different methods for different activities, rather than requiring the same precision everywhere.[S23] GFT would translate that principle into an evidence worklist. Each line would identify the activity, the current estimate, the weakness that matters, the decision exposed to that weakness and the person able to resolve it. The worklist should describe a decision problem, rather than reward whichever supplier submits the most attachments.

Begin with a process map and a completeness register. For every included process, record whether the model uses measured activity, an engineering estimate, an industry average or a scenario. Record exclusions separately. An empty field must remain distinguishable from an observed absence of activity. GFT recommends a specific missing-data flag, with a temporary proxy and an owner where a proxy is justified. A blank emissions cell must never become an apparently measured zero merely because the calculation software expects a numeric input.

Next ask how a weak input could affect the result. A large estimated contribution is an obvious collection candidate, but it is not the only one. A smaller contribution may determine whether an apparent improvement survives, whether a contractual boundary is satisfied or whether an operating action can be checked. Conversely, a large contribution whose uncertainty would not change the immediate choice may be suitable for a bounded screening estimate while a different question is resolved. This is GFT's proposed decision discipline, rather than a universal materiality threshold.

The evidence ladder in Exhibit 15 expresses that discipline. It is a proposed workflow informed by the screening guidance, not a statement that its sequence has been implemented. The final review can send the account back to collection when the remaining uncertainty still affects the decision. Completion means the evidence is adequate for the declared purpose, rather than that every possible variable has become a direct measurement.

A useful request states the uncertainty it is intended to reduce. Instead of asking for a complete sustainability report, ask for the fuel record covering the same production interval as the product output, or the site allocation schedule behind a supplied fabric factor. GFT recommends checking whether the requested evidence already exists in production, maintenance or purchasing records before creating another reporting form. Collection cost belongs in the decision, but it cannot justify hiding an unresolved hotspot or replacing an unavailable observation with a favourable assumption.

The worklist should distinguish a data owner from an action owner. GFT proposes that the holder of the original record confirms what it measures, while the carbon analyst confirms how it enters the model. A purchasing team can identify a supplier and a transaction period without being asked to approve a technical emission factor. A production supervisor can explain unusual operating conditions without being made responsible for an entire life cycle account. This division gives each request a practical route to resolution and avoids an unexplained transfer of judgement between departments.

Priority should also survive a change in the headline result. GFT recommends retaining the earlier screening estimate beside its replacement and documenting whether the activity moved up or down the collection worklist. A revised estimate may reveal that a formerly minor stage deserves attention. It may also show that the initial hotspot was largely a mismatch in the proxy. Neither result is a failure of screening if the method made the uncertainty visible and the replacement evidence is properly tested. Screening should guide learning rather than defend its original ranking.

The worklist also needs stopping rules. GFT would close a collection action when the source is readable, its boundary matches the model, the replacement calculation is reproducible and the remaining limitation is recorded. Where evidence is inaccessible, retain a dated screening estimate with a sensitivity test and a further collection trigger. Reopening should follow a changed process, a newly material uncertainty or a new intended use. A model adequate for internal investigation may need further evidence before it supports an external comparison.

Ecoinvent describes its processes as geographical production averages rather than company-specific or site-specific conditions.[S16] That distinction gives background data a useful role and a clear limit. A background process can supply a transparent starting point where direct evidence is unavailable. It cannot establish that a particular mill uses the represented technology, energy mix or production yield. GFT recommends making every proxy visible in the model so that the analyst can replace it without reconstructing the account.

The first selection question is process identity. Write what enters and leaves the dataset, its reference product and its declared unit. Check whether it represents fibre, yarn, fabric or a finished good before attaching it to a purchasing quantity. GFT would also record moisture or conditioning assumptions where these affect the mass basis. A factor with the right material name but the wrong processing stage should be held as a candidate, not accepted automatically. The account needs a matching process boundary as well as a matching commodity description.

Geography should describe the process represented, rather than merely the address of the trader supplying it. GFT recommends keeping production location, purchase location and dataset geography in separate fields. Where the true production location is unknown, state that gap and test plausible alternatives. A purchase invoice can prove that a transaction occurred without proving the origin of its upstream electricity or heat. This distinction is particularly useful in a textile chain where intermediate goods can pass through several commercial owners before reaching the garment maker.

Quantis describes WALDB as an apparel and footwear database with disaggregated processing information and geographical modelling.[S17] Its sector coverage can help an analyst locate relevant candidate processes, but coverage alone does not settle transferability. GFT recommends requesting the dataset documentation before accepting a process match, including the model version and restrictions on use. The chapter relies on the provider's description of the resource, not on an inspection of its licensed underlying inventory. No proprietary process value is reproduced here.

Method consistency remains a separate test. Textile Exchange cautions that methodological choices, assumptions and boundaries affect LCA results, and advises against comparisons between unlike studies or datasets.[S22] GFT would therefore keep a candidate-factor register rather than assemble a ranking from unrelated published numbers. The register should show why a dataset was selected, what it substitutes for and which sensitivity case tests its weakest match. A common reporting unit cannot by itself make inconsistent models comparable.

GFT proposes recording both the foreground and background versions when a model combines direct factory observations with database processes. Otherwise a later software update may change the total even though the factory file is unchanged. Preserve the selected process identity, system model, impact method and any modifications made by the analyst. The record should explain whether a supplied value is a direct inventory flow or an already characterised impact result, so the receiving calculation does not apply characterisation again to an impact total.

Selection should include a rejection record. A plausible candidate may be unsuitable because its energy route is unknown, its product output differs or its boundary duplicates another included stage. GFT recommends recording the specific reason rather than merely marking the candidate as poor quality. That record prevents the next analyst from reinstating the same unsuitable factor during an update. It also makes a collection request more focused: the supplier can be asked for the missing process description or coverage statement that would allow a decision, instead of being asked to repeat its entire submission.

A proxy replacement should leave an audit trail. GFT recommends retaining the original dataset identifier, the replacement identifier, the reason for change and the affected model lines. Do not overwrite a global factor because a supplier provides a more specific record for a particular purchase. Create a scoped replacement for the relevant product, site and period. That approach allows the analyst to discover whether an apparent improvement came from production change, a better process match or a changed modelling convention.

For a Cameroon application, the absence of local process evidence should remain a visible limitation, not become a claim of either advantage or disadvantage. GFT proposes selecting a documented screening proxy, identifying its mismatch and checking whether reasonable alternatives change the immediate decision. If they do, prioritise local collection before advancing the claim. If they do not, retain the proxy provisionally and revisit it when the purpose changes. This is an evidence strategy for a prospective account, not a measured conclusion about Cameroonian manufacturing.