Cutting-room marker loss and sewing defects consume 15 to 25 percent of purchased fabric

The inherited evidence register summarises inspection literature in which manual visual inspection achieves 60 to 70 percent accuracy and misses 20 to 30 percent of defects. It records computer vision above 90 percent accuracy across more than 40 defect types and line-scan inspection at up to 60 metres a minute. These are literature benchmarks retained in the register, not a current acceptance test for a specified camera, fabric or factory.[R29]

Detection accuracy is not itself a waste-saving percentage. For an illustrative population of 100 defects, a 65 percent detection rate intercepts 65 and a 90 percent rate intercepts 90. The difference is 25 earlier interceptions, not a 25 percent reduction in purchased fabric. The saving depends on defect location, size, severity, the length that must be removed and whether the fault would otherwise have caused a recut panel, a repaired garment or a rejected unit.[R29]

Location determines how much value remains at risk. A fault mapped on an incoming roll can be cut out as an unsewn strip and logged against the supplier. The same fault found after cutting may require a replacement panel and another cutting operation. Found at final audit, it can consume the panel, sewing labour, thread, trims and pressing already applied. Early inspection therefore protects conversion cost as well as material. The business case should count the route each defect would otherwise have taken rather than applying one average saving to every detected fault.[R29][S33]

A usable roll map contains more than a defect count. It needs roll identity, position along and across the fabric, defect class, dimensions, image or inspector confirmation, and the disposition decision. During spreading, the operator can then place a splice deliberately or show that a fault falls inside an inter-piece void. The quality team can compare supplier lots on defects per inspected length, while the cutting team can measure metres removed by cause. These are different decisions supported by the same observation.

Automation also needs a manual exception path. Fabric texture, colour, reflectance and defect presentation can affect classification, and the register's benchmarks do not establish performance on every Cameroon cotton construction. A deployment should begin with labelled local rolls, an agreed defect taxonomy and blind comparison against competent inspectors. Acceptance should measure false negatives and false positives by defect class, not only aggregate accuracy.[R29]

The earlier knit study assigns 6.91 percent of intake to panel checking, 4.31 percent to sewing and 1.72 percent to finishing. These figures are not directly transferable to another factory, but they demonstrate why material loss cannot be managed only in the cutting room. A factory may improve its marker and still lose the benefit through damaged panels, repeated handling, irreparable seams or final rejection.[S0303]

Rework, seconds and scrap must remain separate categories. Rework returns a garment to the process and consumes additional labour, but it does not necessarily destroy the fabric. A second-quality garment may remain wearable and saleable through an approved channel. Scrap is material that cannot re-enter the intended product route. Treating every first-pass failure as a total fabric loss overstates waste; hiding irreparable failures inside rework understates it.

For this chapter's calculation, first-pass yield means units accepted without rework divided by units started, expressed as a percentage. It needs a companion material measure: accepted garment mass plus documented seconds, repairable work and scrap. This pairing stops a line from improving its reported yield by reclassifying defective units. It also lets the plant value the labour and material consequences separately.

The cost record should follow that disposition rather than applying one defect cost to every failure. Repairable rework carries additional handling and labour while preserving the garment. A wearable second may carry a price reduction and an approved alternative destination. An irreparable unit carries its material and accumulated conversion cost, less any verified recovery credit. This separation prevents two common errors: counting repaired garments as total waste, and assuming that every rejected garment can recover its full fabric value.

Defect Pareto analysis should use the same dispositions. A frequent minor seam fault may consume more labour than a rare catastrophic fault, while the rare fault may destroy more material. Ranking only by defect count can therefore direct attention to the wrong loss. The review should compare frequency, affected mass, rework time and final route, then assign the corrective action to the process that created the defect rather than the station that happened to find it.

Overlock and safety-stitch trimming creates another distinct stream. Where polyester or core-spun thread is used, the trim combines cotton lint with non-cellulosic thread. It should not enter the clean inter-pattern clip container by default. The practical control is a dedicated receptacle at the relevant station, with the thread specification linked to the product bill of materials. A cellulosic thread specification may simplify later recovery, but the actual bale still requires inspection rather than an assumed purity claim.

Destructive testing belongs in the same material ledger, even though its purpose is quality assurance rather than process failure. Where tensile, seam, pilling or other buyer-required tests destroy a sample, the laboratory should issue a sample identity and record its mass and disposition. Predictable test remnants can then be routed correctly, and testing cannot become an unexplained drain on production stock. This is an operating control recommended by GFT, not a claim that testing forms a fixed share of fabric loss.

Supervisor action should follow the defect's first occurrence. A skipped-stitch cluster calls for machine and operator checks. Shade mixing calls for bundle and roll identity control. Panel damage calls for handling changes. Repeated recuts from one supplier lot call for an incoming-quality exception. The measure is useful only when it triggers a named response before more material follows the same route.

Pre-consumer loss also includes prototypes, samples, cancelled orders and unsold production. These streams sit outside marker efficiency but still embody purchased fabric. The European Environment Agency estimates that 82 percent of EU textile waste in 2020 was post-consumer, leaving 18 percent derived here as manufacturing or unsold textiles. The agency cautions that data on pre-consumer categories such as unsold retail textiles are limited, so the residual must not be presented as a precise factory-waste share.[S12]

Demand forecasting addresses a different failure from marker nesting. A perfect marker cannot rescue an order whose size, colour or quantity was wrong. The inherited register summarises McKinsey's State of Fashion 2025 as reporting forecast-error reductions of 20 to 50 percent and inventory-cost reductions of 35 to 40 percent from AI at scale in fashion. These are potential effects reported across the sector, not guaranteed results for a new Cameroon plant, and they depend on usable sales and order history.[R27]

A cold-start factory should consequently begin with smaller, traceable order batches and explicit assumptions rather than a sophisticated forecast built on sparse data. Record the forecast, the actual order, the size and shade curve, cancellations, replenishment and unsold balance by style. Once that history exists, a model can be tested against a naive baseline. Forecast value should be judged by avoided overproduction and stockouts, not by model complexity.

Digital design and sampling can prevent another early loss. The inherited register describes apparel AI work in 2023 and 2024 as mainly design-side image generation, followed by later multi-model pipelines covering fabric construction, shade, weave and silhouette, with physics-constrained simulation emerging in 2025 and 2026.[R30][R31] This supports virtual review and fewer unnecessary iterations, but it does not prove that every physical fit or performance sample can disappear.

The operating rule is to give every sample a decision purpose. A digital silhouette review may remove one physical iteration. A buyer fit approval may still require cloth. A durability claim still needs appropriate testing. Recording metres and garments used by sample type reveals which approvals genuinely change and which are habitual. This chapter therefore treats forecasting and sampling as prevention controls, while later chapters examine unsold-stock rules and digital design in greater detail.