One-sentence answer: Scheduling works by assigning tasks to time and resources while respecting release times, durations, dependencies, capacities, deadlines and uncertainty, then continuously comparing the plan with the world and replanning when assumptions change.
A schedule is not a calendar with boxes filled in. It is a model of temporal feasibility. It answers a harder question: given limited time, resources, dependencies and uncertainty, what can happen when without breaking something that matters?
A good schedule is not the one with no empty space. It is the one that survives contact with reality without sacrificing the wrong receiver.
Quick Read: the causal chain
OBJECTIVE → TASKS → RELEASE TIMES → DURATIONS → DEPENDENCIES → RESOURCE NEEDS → CAPACITY → DEADLINES / WINDOWS → FEASIBLE SEQUENCE → BUFFERS → BASELINE → EXECUTION → DELAY / ARRIVAL / FAILURE → REPLAN → RECEIVER DEADLINE → WORLD RETURN
1. Scheduling begins with a temporal job
Different schedules solve different problems. A school timetable tries to place classes, teachers and rooms. A factory schedule sequences jobs through machines. An airline schedule coordinates aircraft, crews, gates and passengers. A revision plan allocates attention across subjects before an examination. A computer scheduler allocates processor time. A hospital schedules theatres, clinicians, equipment and patients.
Before building a schedule, ask:
- What outcome matters?
- Which deadlines are hard and which are preferences?
- Which receivers are affected by lateness?
- Which resources are actually scarce?
- Which tasks depend on earlier tasks?
- How uncertain are durations and arrivals?
If these are unclear, the schedule may optimise an irrelevant metric.
2. A task needs more than a duration
A useful scheduling record distinguishes several time and constraint fields:
| Field | Meaning |
|---|---|
| Release time | Earliest time the task can begin. |
| Duration | Expected amount of processing time. |
| Deadline / due date | Time by which completion is required or preferred. |
| Precedence | Tasks that must occur before another can start or finish. |
| Resource need | People, machines, rooms, tools or capacity required. |
| Setup / changeover | Time or cost created by switching states. |
| Pre-emption rule | Whether a task may be paused and resumed. |
| Priority / class | Ordering rule when demand exceeds capacity. |
Collapsing these fields into one “time needed” estimate hides why a schedule becomes infeasible.
3. Dependencies create the shape of the schedule
Some tasks cannot start until others finish. Concrete must cure before another construction step. A student may need algebraic manipulation before tackling a dependent topic. A software deployment may require database migration before application release.
Dependencies form a network. The order is therefore not arbitrary.
Typical relationships include:
- finish-to-start;
- start-to-start;
- finish-to-finish;
- minimum or maximum lag;
- shared-resource dependency;
- conditional dependency triggered by an event.
An incorrect dependency can make an impossible schedule look feasible or a feasible one look blocked.
4. The critical path identifies the current completion bottleneck
In a dependency network, the critical path is the chain of activities that currently determines the earliest completion date under the scheduling model. If one critical activity slips and nothing else changes, the project completion slips.
GAO’s Schedule Assessment Guide treats a valid critical path as a core property of a well-constructed schedule and warns that broken dependencies or hard date constraints can corrupt float and critical-path calculations.
Critical path is not a permanent label. If durations, resources or dependencies change, the critical path can move.
5. Float is time flexibility—not automatically spare capacity
Float or slack describes how much an activity may move without violating a particular downstream date under the current network.
It does not automatically mean the associated people or machines are idle. A non-critical task can still use a resource that is overloaded elsewhere. Time flexibility and resource capacity are different dimensions.
This distinction links scheduling to How Capacity Works.
6. Capacity converts logical feasibility into operational feasibility
A schedule can satisfy every dependency and still be impossible because too many tasks require the same resource at the same time.
Examples:
- two classes need the same teacher;
- three jobs need one machine;
- two surgeries need the same theatre;
- several services share one database migration window;
- multiple revision tasks assume the same two-hour evening block.
Scheduling therefore requires resource feasibility, not just precedence feasibility.
7. Utilisation near 100% can make delay explode
Systems with variable arrivals or service times often become fragile when utilisation approaches full capacity. There is no recovery room for random variation. Small delays queue behind one another.
This is why “fill every available slot” can be a bad scheduling objective. Buffers may look inefficient locally while protecting the end-to-end receiver.
Queueing logic helps explain why: waiting depends not only on average workload but on variability, service distribution, arrival pattern and capacity.
8. Buffers absorb uncertainty
Durations are estimates. People get sick. Machines fail. Traffic changes. Tasks are discovered late. Information arrives slowly.
Buffers can appear as:
- time contingency;
- spare capacity;
- inventory or material reserve;
- extra route time;
- recovery windows;
- protected transition time;
- fallback resources.
A buffer is not automatically waste. It is often a deliberate response to uncertainty and consequence.
9. A schedule is a baseline, not a prophecy
A baseline schedule records the approved plan against which progress and variance can be measured. It should be stable enough to provide accountability but not so rigid that the system refuses to acknowledge new reality.
GAO describes a reliable schedule as comprehensive, well-constructed, credible and controlled. In 2026, GAO continues to use this framework when assessing major modernisation programmes, showing that the guide remains an active benchmark rather than historical advice.
Control therefore means:
- record the baseline;
- record actual progress;
- explain variance;
- update forecasts;
- preserve reasons for approved changes;
- avoid silently rewriting history.
10. Static, rolling-horizon and event-driven schedules solve different worlds
| Mode | Best fit | Main risk |
|---|---|---|
| Static schedule | Stable tasks and known horizon | Becomes stale when conditions move. |
| Rolling-horizon | Plan near-term detail, revise later periods as evidence improves | Repeated changes can destabilise commitments. |
| Event-driven | High-variability environments where specific events trigger replanning | Too many triggers can cause thrashing. |
| Real-time / dynamic | Continuous arrivals and rapidly changing capacity | Optimising the instant can damage longer-term fairness or stability. |
The planning mode is itself a design choice. A hospital emergency queue and a school examination timetable should not use the same replanning logic.
11. Pre-emption changes the scheduling problem
Some work can be interrupted and resumed. Other work cannot.
A CPU task may be pre-empted frequently. A chemistry reaction may not be safely paused. A student can switch revision subjects, but excessive switching can create cognitive setup cost. A surgery cannot be treated like a spreadsheet task.
A schedule therefore needs to know:
- whether interruption is allowed;
- what state must be preserved;
- the cost of pause/resume;
- whether partial completion has value;
- what happens to downstream commitments.
12. Priorities determine who waits
When demand exceeds capacity, a priority rule decides who receives service first. First-in-first-out, earliest deadline, shortest job, highest consequence, highest value and triage-based ordering create different distributions of waiting.
There is no universally neutral priority rule. Every rule privileges a feature.
That means scheduling is partly governance. The rule should state which receiver outcomes matter and what recourse exists when a case is delayed.
13. Optimising average completion can create unfairness
A scheduler that minimises average completion time may favour many short tasks while starving a long but important task. A hospital that optimises throughput alone could mishandle severity. A school schedule that maximises room utilisation might create exhausting back-to-back teaching loads.
Scheduling objectives may include:
- makespan;
- lateness;
- throughput;
- waiting time;
- cost;
- utilisation;
- fairness;
- deadline reliability;
- human workload;
- risk;
- schedule stability.
How Optimisation Works owns the general objective/constraint/trade-off mechanism. Scheduling owns its temporal expression.
14. Time windows can make routes feasible or impossible
A destination may be physically reachable but temporally unreachable. The bus leaves before the feeder arrives. The clinic closes before the patient can reach it. The loading bay has a restricted window. A visa appointment must occur before another deadline.
Temporal feasibility therefore belongs inside routing:
LOCATION + TRAVEL TIME + TRANSFER WINDOW + CAPACITY + OPENING WINDOW + RECEIVER CAPABILITY → ACTUALLY REACHABLE
This connects scheduling to logistics and route planning without duplicating those domains.
15. Stale schedule state is a hidden failure mode
Schedules depend on current state: task completion, resource availability, arrivals, outages and deadlines. If one part of a distributed system replans while another still operates on the old schedule, conflicts appear.
World-class scheduling systems therefore manage:
- schedule version;
- effective time;
- acknowledgement;
- cancellation;
- supersession;
- reservation state;
- commitment state;
- exactly-once or idempotent actions where needed.
The data layer matters here: a correct schedule computed from stale state is operationally wrong.
16. Replanning should change when evidence changes—not every time noise appears
If the system replans too rarely, the schedule drifts away from reality. If it replans after every tiny fluctuation, commitments become unstable and people cannot coordinate.
Useful replanning triggers include:
- a critical-path delay;
- a missed or threatened receiver deadline;
- resource failure;
- new high-priority arrival;
- material change in duration estimate;
- constraint removal or addition;
- schedule version conflict;
- risk threshold crossing.
The trigger should be large enough to matter and fast enough to prevent cascade.
17. Worked example: a revision schedule
A student has seven days before an examination. A weak schedule fills every hour with subjects. A stronger schedule asks:
- Which topics have prerequisites?
- Which weak links are earliest and most consequential?
- Which tasks require fresh attention?
- Which deadlines are fixed?
- How much recovery time is needed?
- Which skills require retrieval after delay rather than same-day repetition?
- What happens if one evening is lost?
The schedule may deliberately leave buffer time. That is not laziness. It protects recovery and creates space for evidence-driven replanning after practice tests.
18. Worked example: a production line
A factory has three machines and several product types. Jobs have different processing times and changeover requirements. If the schedule minimises machine idle time, it may create downstream inventory or delay urgent orders.
A stronger scheduler considers:
- due dates;
- setup sequences;
- bottleneck machine capacity;
- maintenance windows;
- material arrival;
- quality hold states;
- rework probability;
- customer priority;
- recovery after machine failure.
The receiver is not “machine utilisation.” The receiver is the production system and ultimately the customer receiving acceptable output at the required time.
19. Scheduling across domains
| Domain | Scheduling problem | Critical boundary |
|---|---|---|
| Education | Study, classes, assessment, teacher/room allocation | Learning sequence and human recovery matter. |
| Manufacturing | Jobs, machines, setups, materials | Throughput can conflict with due dates and resilience. |
| Healthcare | Patients, staff, theatres, equipment | Clinical priority and safety cannot be reduced to generic queue rules. |
| Transport | Vehicles, crews, transfers, windows | Network delays and missed connections cascade. |
| Computing | Processes, processors, memory, deadlines | Fairness, latency and starvation differ by workload. |
| Construction | Tasks, dependencies, crews, materials | Critical path, weather and procurement uncertainty matter. |
| AI agents | Tool calls, jobs, dependencies, retries | State freshness, idempotency and authority matter. |
20. Common scheduling failures
| Failure | What goes wrong | Repair |
|---|---|---|
| Calendar thinking | Tasks are placed without dependency logic. | Build the network first. |
| False duration certainty | Point estimates are treated as guaranteed. | Use ranges/risk analysis and buffers. |
| Resource blindness | Logical schedule exceeds actual capacity. | Model constrained resources explicitly. |
| Hard-date masking | Artificial constraints hide the true critical path. | Use logical relationships and transparent constraints. |
| 100% utilisation | No recovery room remains for variability. | Protect buffers and effective capacity. |
| Average-only optimisation | Critical or weak receivers are delayed. | Include fairness, deadline and consequence constraints. |
| Stale state | Old schedule remains active after reality changes. | Version, acknowledge and supersede schedules explicitly. |
| Replanning thrash | Constant updates destroy stability. | Use material triggers and protected commitments. |
| Local pass/global fail | One resource is optimised while end-to-end delivery worsens. | Measure receiver completion and cascade effects. |
21. Hostile test: perturb the assumptions
- What task network is being scheduled?
- Which deadlines are truly hard?
- What resources are scarce?
- Where is the current critical path?
- What float exists, and is it confused with capacity?
- Which durations/arrivals are uncertain?
- What happens if each critical duration grows by 20%?
- What happens if one key resource disappears?
- Which receiver becomes late first?
- What buffer absorbs the disturbance?
- When does replanning trigger?
- Does the new schedule preserve fairness, safety and authorised commitments?
22. Where Scheduling fits in the wider How Things Work map
Scheduling connects Capacity, Optimisation, Logistics, Maintenance, Work, Risk, Uncertainty and Data.
Its distinct public job is: How do tasks, dependencies, resources and deadlines become a feasible, resilient sequence through time?
23. What this article does not claim
- A full calendar is not automatically a good schedule.
- Float is not automatically spare resource capacity.
- Critical path is not fixed when assumptions change.
- Maximum utilisation is not always efficient once variability and queues are considered.
- A mathematically optimal schedule is not automatically fair, safe or legitimate.
- A static schedule should not be treated as current after material world changes.
- Generic scheduling logic does not override clinical, legal, educational, safety or operational owners.
24. Observable mastery test
You understand scheduling when you can take an unfamiliar plan and identify task durations, release times, dependencies, resource constraints, deadlines, critical path, float, buffers, uncertainty, pre-emption rules, priority logic and replanning triggers—and explain which receiver fails first when one assumption changes.
Authoritative source corridor
- U.S. GAO Schedule Assessment Guide: Best Practices for Project Schedules — integrated schedule logic, critical path, risk and schedule control.
- GAO-26-107629 (2026) — current application of the Schedule Assessment Guide and its four reliability characteristics.
- Little (1961): A Proof for the Queuing Formula L = λW — foundational queueing relationship.
- How Capacity Works and How Optimisation Works — neighbouring public mechanism owners.
Governing idea: A schedule is good when it preserves feasible order through time and still reaches the receiver after uncertainty starts moving the plan.