Inventory earns its place only when it protects an explicit promise

The programme design base for this paper is cut, make, trim and finishing. Spinning and knitting are conditional partnered stages, not installed GFT operations. This boundary matters because the garment maker's immediate stock decisions concern approved fabric, trims, work in progress and finished garments, supported by dependable specifications and receipts. It does not begin by deciding how much seed cotton or spinning capacity to own. The three-part programme connects feasibility, performance management and digital strategy at this operating boundary without asserting deployment.

An integrated mill would face a different set of decoupling points and physical conversions. It would need to coordinate fibre, yarn, greige or finished fabric, wet-processing queues and downstream garment requirements. That may be a valid partnered configuration after its technical and commercial case is proved, but importing its stock architecture into the base case would obscure the first decisions a garment maker must control. A model should fit the operation being governed, not the most expansive possible value chain.[S07]

Within the garment-making boundary, three illustrative contract archetypes create different promises:

These archetypes can coexist in one factory. That does not permit one undifferentiated stock pool. The same nominal cloth may be forecast-owned for a local range, reserved for a uniform order, or supplied under custody for an assembly customer. Planning may identify technical commonality, but commercial release requires compatible ownership and approval rights. Later chapters examine segmentation, pooling and postponement in detail; the opening rule is simply that physical similarity does not erase a contractual boundary.[S01][S08]

The archetype also changes what counts as a shortage. In make-to-stock apparel, an unavailable requested variant can become a lost sale. In make-to-order production, a missing component can delay a confirmed order and trigger expediting or renegotiation. In customer-supplied assembly, an unusable receipt can create a custody discrepancy and a demand for replacement outside the factory's ordinary purchasing authority. A single service formula applied without the contract context would report different failures as if they were the same event.[S05][S07]

A decision paper must distinguish a method from an observed result. An accounting standard can define measurement, a teaching text can establish a conditional model, a field study can report what happened in its sample, and a research manuscript can examine an algorithm under stated assumptions. None of these, alone, proves that a proposed policy will save money in a Cameroon garment factory. The wording of a recommendation should carry the strength and boundary of its evidence.[S01][S09][S22][S23]

Official and foundational sources are indispensable but limited. IAS 2 supports the distinction between accounting valuation and operational decision cost; it does not prescribe a stock quantity.[S09] MIT OpenCourseWare provides methods for segmentation, deterministic and probabilistic inventory models, dependent requirements and contracts. Those methods allow transparent calculations, but their assumptions must be checked against local demand, supply and contract conditions before their outputs become orders.[S01][S02][S05][S07][S08]

Field evidence answers a different question. The ILO's 2017 purchasing-practices survey reports 1,454 supplier responses across 87 countries.[S10] That sample supports attention to the relationship between purchasing and working conditions, not an estimate of current prevalence in Cameroon. Likewise, a fashion-retail conference study comparing an RFID cycle count with a physical count supports validating a capture method against physical truth. It does not establish a savings rate or read accuracy for a garment factory with different packaging, layout and processes.[S15]

Historical place-specific evidence needs equal care. The World Bank's Douala publication provides a historical diagnosis of port efficiency and investment constraints.[S14] It justifies measuring clearance, transport and receipt as separate stages. It does not provide a current lead-time parameter for a new purchase policy. Using an old corridor observation as if it were a present supplier distribution would convert a useful diagnostic prompt into an unsupported buffer calculation. Current quotations, route observations and usable-receipt timestamps must replace inherited assumptions.[S14][S30]

Research manuscripts help formalise failure modes without supplying local operating proof. Work on lost sales and uncertain supplies identifies censoring: observed demand and supply can be truncated by availability.[S22] A simulation-optimisation manuscript illustrates repeated stochastic comparison of inventory policies.[S23] These contributions inform what an evaluation should measure. They do not make unobserved demand known, validate a factory's input distribution or establish that a simulated ranking will survive real supplier and customer behaviour.

Exhibit 4 sets these evidence classes beside their permitted use. The table is a source-bound reading guide, not a quality ranking. A carefully executed simulation can be more informative than a poorly matched field sample, but only for the question its design actually addresses.

This discipline also determines what must be absent. Missing annex references are excluded rather than cited, and unsupported tariff or training claims are not carried into the stock argument. The source register's verification limitations remain visible instead of being converted into present-tense certainty. Where no current local observation exists, the chapter uses a labelled illustrative case or states the input required; it does not fill the gap with a precise number.

A forecast estimates uncertain demand; an inventory policy converts that estimate into a commitment. WP04 owns demand distributions and forecast residuals. This paper begins at the decision interface and chooses quantity, timing, location, material form and approval conditions. It also decides what not to buy, which uncertainty to carry, and when a planner must return to the commercial owner.[S05][S06][S07]

That separation prevents two common errors. The first is to treat a central forecast as a deterministic order, ignoring lead-time variation, usable stock and lot constraints. The second is to add an unexplained buffer to every forecast in the name of caution. A high expected demand does not justify a purchase if the required material cannot arrive in time or the cash limit is binding. A low expected demand does not imply zero stock when a confirmed promise, a long protection period or a line-stopping component changes the consequence of shortage.[S03][S05][S07]

The decision interface must therefore carry more than a demand number. It needs the forecast vintage, requested service event, stock and reservation boundary, open supply, usable-receipt distribution, approved substitutions, order bounds, cash limit and expiry of the recommendation. Later chapters derive specific replenishment and seasonal rules. Their common obligation is to expose the assumptions that transform an estimate into an action.[S05][S22][S30]

Other papers retain their own subjects. WP05 owns detailed traceability qualification, WP09 labour and supervisory capability, WP07 energy and infrastructure architecture, and WP13 financing structures. Their constraints remain present here: stock cannot be promised without usable identity, service cannot be protected by assuming harmful working practices, and a materials commitment is not funded merely because a future receivable is imaginable.[S10][S18][S20]