The bottleneck is a system property

The minimum-capacity result follows from conservation in a restricted serial chain. Every accepted tonne must pass through each stage, and no stage can advance more than its effective annual capacity. Delivered flow therefore cannot exceed any stage capacity. It also cannot exceed eligible demand under the model's definition. The smallest of those bounds limits the chain. This is an upper-bound argument first; treating the bound as attainable requires further assumptions about coordination, continuity and the ability to hold material between stages.

For the canonical app, those further assumptions are built into the steady-state convention. The chain has a single route, consistent cargo quantity, no processing loss and no time-resolved queue or inventory. Each stage is represented by an annual effective capacity rather than a dispatch schedule. Under that convention, the minimum becomes the calculated deliverable throughput. The term deliverable should retain its illustrative model basis. It does not mean an operator has demonstrated the physical, administrative or contractual ability to maintain that service in an actual corridor.

A binding stage is one whose effective capacity equals the calculated chain flow within the engine's comparison convention. When demand is lower than all stage capacities, the model has no capacity bottleneck. When several stages share the smallest capacity, they are jointly binding. These are different states and should not be collapsed into a single recommendation to improve the weakest asset. Demand limitation points toward a commercial or service question. Tied capacities point toward a coordinated intervention question. A unique constraint supports a single-stage sensitivity test but does not establish its economic desirability.

The derivation explains why local spare capacity is not necessarily waste. A nonbinding stage can provide recovery capability, accommodate batch arrivals or protect a service during another stage's interruption. The annual formula only records the difference between effective stage capacity and common flow. It does not value resilience. Calling every positive difference stranded capacity would therefore overstate the model. The useful distinction is between capacity that cannot raise steady-state flow on its own and capacity that has no justified role under the full service objective.

The frozen base chain delivers 11.90 Mt/year against illustrative annual demand of 20.00 Mt/year. The binding stage is Rail corridor. Its nominal capacity and availability are explicit assumptions rather than measurements. The result is a deterministic gap under those assumptions. It does not estimate the likelihood of missed delivery, the frequency of congestion or the level of profitable customer demand.

Illustrative. The vessel-loading intervention changes only its nominal capacity. Calculated flow remains 11.90 Mt/year and the binding stage remains Rail corridor. This is the clearest form of the paper's thesis: enlarging a visible stage does not increase annual flow when another unchanged stage sets a lower bound. The comparison is useful precisely because it holds demand, the route and other capacities constant. It says nothing about whether the loader expansion improves berth occupation, safety or recovery during a concentrated vessel window.

Illustrative. The rail intervention produces 12.00 Mt/year and shifts the binding stage to Vessel loading. Relieving the original constraint therefore changes the result, but it exposes a nearby limit. The coordinated rail and loading intervention produces 14.40 Mt/year with Port stockyard binding. This sequence shows why an expansion plan should identify the next constraint before committing to a local improvement. A solution that merely shifts the problem can still be useful, but its extent must be stated honestly.

The operating unit costs change with the common throughput denominator while the input operating costs remain fixed. That isolates a mechanical effect rather than a procurement conclusion. A committee should not infer that the coordinated intervention is the best investment solely because its illustrative operating cost per delivered tonne is lower. The cases omit intervention capital costs, timing and feasibility. They establish the physical ranking under the model's restricted assumptions, then hand the economic and operational questions back to the relevant decision process.

A measured shortfall is not automatically intrinsic capacity. An idle rail unloading system can be starved because trains arrive late, blocked because the yard has no eligible space, held because documentation is incomplete or unavailable because it is under repair. Each mechanism can reduce observed output, but each has a different remedy. An annual effective-capacity input can absorb all of them and still produce the right historical flow. That fit would not identify which intervention will improve future service.

The diagnostic map should therefore distinguish capability, eligibility, coordination and demand. Capability asks what a ready stage can safely perform under the declared configuration. Eligibility asks whether cargo and permissions allow performance. Coordination asks whether adjacent stages and shared resources are ready together. Demand asks whether the resulting service has a user. These categories are original analytical distinctions for the current study. They are not empirical findings or a claim that every corridor failure can be assigned cleanly to a single box.

Some mechanisms cross categories. A stockyard may block a train because its physical space is full, because segregation rules reserve that space or because a vessel cancellation changes the intended inventory pattern. The correct response depends on the cause, even though all cases present as no room. The analyst should record the immediate stop condition and the upstream reason separately. This protects the intervention from addressing the symptom while leaving the mechanism intact. It also gives managers a fairer allocation of responsibility than a single utilisation score.

A constraint can move without any asset changing. A demand shift, product mix or operating calendar can alter which resource governs the service. An intervention that improves the original stage may therefore encounter a different system by the time it is commissioned. The committee should treat the map as a living hypothesis tied to a service edition. It should request renewed evidence when the relevant conditions change rather than preserve the original bottleneck label as a permanent characteristic of the corridor.

A parallel route breaks the single-path assumption. Cargo may use several paths, each with its own capacities and shared links. The useful system constraint can then be a network cut, a dispatch allocation or an access condition rather than the smallest stage in an arbitrarily drawn route. Adding the capacities of alternatives is also unsafe when they share a bridge, border process or marine approach. A broader network method would need explicit topology and constraints. The present app should not be stretched into that role by adding imagined stages.

Finite buffers and varying calendars break the attainability assumption. Annual capacity may exist at each stage while cargo cannot be transferred at the required time because a yard fills, a gate closes or a shared handling resource is occupied. A deterministic annual minimum may then overstate realised service. It remains a useful screening upper bound, provided it is named as such. A time-based operating study should carry the chronology needed to decide whether that bound can be approached under the actual service requirement.

Quality restrictions can make gross flow a poor objective. More movement may create off-spec cargo, damage or rework that reduces accepted delivery. Legal and safety requirements can be noncompensable: a faster rate cannot offset a prohibited movement or unsafe operating configuration. A capacity calculation should therefore sit behind an eligibility gate. The model result can rank feasible options only after those conditions are reviewed, and even then the committee must decide whether increased accepted service is worth its full incremental costs.

The practical decision is to use the minimum result as a disciplined question generator. It locates which annual assumption governs the fictional flow and tests what happens when that assumption changes. It does not replace the service map, operating evidence or feasibility review. The strongest use of the bottleneck thesis is conditional: relieve the constraint that actually governs the declared service, then check the next limit and the consequences outside the annual throughput metric. An unqualified promise to optimise the bottleneck would be less defensible than this narrower, testable recommendation.