A measurement plan can find the wrong constraint

A measurement programme should begin with the disputed decision, not with the easiest available sensor. If the question is whether a loading asset lacks intrinsic capacity, vessel turnaround alone is too broad. It includes waiting for cargo, nautical services and clearance. If the question is whether users receive a dependable end-to-end service, the loader productive rate is too narrow. The analyst should first describe the mechanism that could change the decision, then select events that distinguish it from plausible alternatives.

An event is a recorded transition, while a state is the condition between transitions. Arrival at a queue, eligibility to enter service, commencement of service, completion of service and acceptance by the next actor are distinct events. Combining them into arrival and departure may suffice to measure total residence, but it cannot diagnose why residence increased. The difference between arrival and eligibility might be documentary delay; the difference between eligibility and service start might be scheduling or resource contention. Treating every interval as equipment inefficiency creates a biased investment case.

The timestamp dictionary should state the physical location, triggering action, recording device and responsible actor for each event. It should distinguish an estimated event from an actual event and retain both rather than overwrite a forecast with an observed completion. A forecast revision is useful evidence of planning uncertainty. It is not a physical movement. If systems transmit only the latest estimate, an analyst cannot later reconstruct how much notice an operator received or whether a missed handoff arose from inaccurate information or inadequate response.

A record should also identify the cargo or service object that persists through the chain. A consignment can split between trains, a train can combine consignments and a vessel can accept several batches. The mapping between these objects is necessary to avoid duplicate quantities and missing intervals. A convenient vehicle identifier may support road-cycle analysis while failing to reconstruct accepted cargo delivery. The unit of observation should therefore follow the decision and be reconciled with each local operational identifier.

Selection bias can enter before a timestamp is recorded. A gate log may include vehicles admitted to the terminal while excluding vehicles waiting outside. A completed-service report may exclude unfinished jobs, giving the appearance of improving average time during a growing backlog. A vessel database may record only calls that reached the berth and omit cancellations. The data can be internally clean while answering a different question from the service promise. The collection plan must state what enters the observation universe and what remains outside it.

Unfinished movements need explicit treatment. At an observation cutoff, cargo in progress has an incomplete duration rather than a short completed duration. A performance report should show the unfinished population and its elapsed waiting states. It should avoid comparing a completed-only average with a later complete cohort as though the bases were identical. This is especially important after disruption: the most delayed cases may still be unresolved while the reported completed movements recover quickly. A reassuring average can therefore coexist with persistent service failure.

Missing timestamps are not automatically zero intervals. A device failure, manual omission and unobserved event have different implications, but none justifies quietly substituting immediate movement. The analyst can mark the duration unavailable, bound it using adjacent evidence or commission a reconstruction. Each treatment should remain visible. Imputation may support a particular analysis if its rule and sensitivity are declared. It should not create an appearance that every movement was directly observed, especially where the missingness relates to difficult or congested operations.

The same discipline applies to sampling windows. A short quiet-period survey can establish a handling process but cannot represent an operating year containing maintenance, mixed cargo or concentrated demand. A campaign should deliberately include the conditions that could change the constraint classification. When those conditions cannot be observed within the available period, the report should narrow its conclusion. It may establish performance during the sampled operating envelope without asserting an annual capability or a tail-delay estimate.

Clock alignment is a substantive measurement issue at institutional boundaries. Systems can use different time zones, device clocks, manual entry conventions or batch upload times. A negative transfer duration can indicate incompatible clocks rather than physical impossibility. The dictionary should distinguish event time from record creation and transmission time. For a manually reported event, the observer and reporting delay should be known where possible. Correcting clocks should preserve the original record and the documented correction rather than erase the discrepancy.

Cause codes should be designed to support action. A generic delay code assigns no mechanism. Conversely, a long menu of mutually unclear codes invites inconsistent use. The programme can record an immediate stop condition, an attributed cause and the evidence supporting that attribution. A downstream yard full condition may be directly observed while its underlying reason remains disputed. Preserve that distinction. A model that treats an allegation about another actor as a measured cause is likely to reinforce local conflicts rather than resolve the corridor constraint.

Quantity reconciliation should track accepted movement separately from gross handling. A conveyor total can count material that later returns to the stockpile; a weighbridge total may need an agreed cargo basis; a train departure total may differ from released receipt. The end-to-end ledger should document where quantities are measured, what corrections occur and whether reconciliation differences remain unresolved. It should not force agreement by deleting the inconvenient measurement. A discrepancy is evidence of a measurement or process problem that may itself matter to the service.

The analyst should produce a lineage sample before collecting a large dataset. Select a clearly labelled fictional or consented test object and reconstruct its event sequence across the relevant systems. Check whether each handoff has an identifiable actor, timestamp basis and quantity basis. This small exercise can reveal that the intended end-to-end duration cannot be assembled from existing records. Discovering the gap early is more useful than acquiring a large extract whose columns never represented the required service transitions.

Consider a fictional appointment trial at a receiving terminal. The gate admits only vehicles with confirmed unloading readiness. Inside-terminal residence becomes shorter because ineligible vehicles remain outside or are asked to delay departure. A local report may correctly show improvement at its declared boundary. The corridor report must ask whether the driver and cargo service improved across the whole journey. Without the external and upstream holding record, it cannot distinguish removal of waiting from relocation of waiting.

The analyst should preserve the original appointment, its revisions, actual eligible arrival, gate entry and accepted receipt. A rescheduled movement is not automatically a failure, but its effect depends on the user's commitment and ability to adapt. If the carrier receives early reliable notice, upstream holding may be less burdensome than a terminal queue. If notice is late, the same policy can leave vehicles waiting on the approach. The intervention benefit therefore depends on the communication transition as well as the physical gate process.

The collection design should include movements refused entry and those not yet completed at cutoff. It should also retain the operational reason for rescheduling and the readiness confirmation that supported it. A terminal can legitimately restrict admission to protect safety or usable space. That condition belongs in the feasible operating plan rather than being described as avoidable inefficiency. The evaluation should distinguish the necessary restriction from the proposed information and appointment remedy.

A useful conclusion may be qualified: the trial improves local receipt coordination, while its end-to-end effect remains unresolved because upstream holding was not measured. That is an actionable finding. The next trial should collect the missing states and test notice quality. The team should resist both declaring universal success from the local metric and rejecting a useful local improvement merely because the wider evidence is incomplete.

Evidence quality should affect the strength of the recommendation. A reliably measured queue at a release gate may justify an operating trial even when the wider annual capacity basis is uncertain. It cannot by itself justify a major physical expansion if the delayed population is unrepresentative or the cause unresolved. The committee should receive the mechanism hypothesis, supporting observations, unresolved alternative explanation and proposed next action together. A dashboard without that chain of reasoning can make weak evidence look decisive.

The evidence gate can require a defined universe, consistent event semantics, acceptable reconciliation and an explanation of missingness. The threshold depends on the decision. An exploratory trial can tolerate unresolved intervals if it is designed to learn about them. A service guarantee needs a much stronger basis. The gate should state what would block the planned decision and what would merely require qualification. This prevents managers from treating every data imperfection as either immaterial or fatal.

An intervention evaluation also needs a comparison design. If a scheduling change is followed by shorter waiting while cargo mix, vessel calls and weather change at the same time, the observation does not isolate the scheduling effect. The team can still learn from the episode, but it should describe the competing explanations. A staged trial, matched operating envelope or mechanism-specific process record may provide a better basis than a broad before-and-after average. The design should be proportionate to the consequences of being wrong.

The final measurement product is a usable record of what happened, what it means and what remains unknown. It should permit an independent reader to distinguish an equipment limit from a handoff failure without trusting a local actor's preferred story. The corridor manager should be able to update the hypothesis when new events contradict it. That is the value of the timestamp dictionary: it turns a collection of local logs into a common language for service, while leaving uncertainty visible enough to influence action.