UML

A class box stacks three compartments: the class name on top, the attributes in the middle, the operations at the bottom. A leading +, -, # or ~ on a member marks it public, private, protected or package-visible.

A lifeline is one participant in the interaction — an object, component or actor — drawn as a labelled box above a dashed vertical line down which time runs. The activation bar marks the stretch where that participant is actively executing.

A state is a named condition the object waits in over time. An event is a discrete occurrence the object perceives. A transition is the arc that fires when that event arrives and moves the object from one state to another.

Structural UML diagrams show what a system is made of — its parts and how they are arranged, frozen in time. Behavioural UML diagrams show what it does — how the parts act, react and change as time passes.

The system boundary separates what you are building from everything outside it: use cases sit inside, actors outside. An actor is any external role that interacts with the system — a person, another system, or a scheduled trigger.

Generalization is a solid line with a hollow triangle pointing at the more general class. Realization uses the same triangle on a dashed line. A plain association is a solid line; a dependency is a dashed line with a thin open arrowhead.

A synchronous call is a solid line with a filled arrowhead and the sender waits. An asynchronous send is a solid line with an open stick arrowhead and the sender continues. A reply is a dashed line with an open arrowhead.

Behaviour belongs on the entry action whenever every way into the state must run it. A transition effect covers one path only, so work duplicated across several incoming arcs is exactly what an entry action replaces.

Start from the question, not the notation. Say in one sentence what you need to know — what the parts are, in what order they interact, or what one object's lifecycle permits — then draw the family that answers it.

Include means the base use case always performs the included behaviour, and the arrow points from base to included. Extend means optional behaviour attaches at a named extension point under a condition, with the arrow pointing from the extending use case to the base.

A decision node routes the arriving token down the single outgoing edge whose guard is true. If no guard holds, the token stops there and the path dies silently; if two hold, the model permits either path and no longer has one meaning.

Multiplicity at an association end says how many objects of that end's class may be linked to one object at the far end. It is a range: 0..1 optional, 1 exactly one, 1..* one or more, * any number.

A combined fragment is a labelled box drawn over part of a sequence diagram, with an operator in its top-left pentagon. alt holds several guarded, mutually exclusive operands; opt holds one that runs only if its guard holds; loop repeats its operand.

It exits them. Leaving a composite state first exits whichever substate is currently active, running that substate's exit action, then runs the composite's own exit action — so one arc drawn on the boundary covers every substate inside it.

Partitions assign every action to the party that performs it, so each control flow that crosses a partition boundary is a handoff. Counting and timing those crossings shows where work waits, rather than where it is being worked on.

A sequence diagram shows one ordered trace of one scenario. It cannot assert completeness, structure, data shape, real elapsed time, or what holds when nobody is calling. Transcribing every call produces a diagram that is unreadable and wrong within a sprint.

Model explicitly when an object has several named conditions, when the legal moves between them matter, and when the same rules are being re-checked in scattered places. Two or three values with no ordering rules need no diagram at all.

Interaction diagrams answer one scenario at a time, so everything that quantifies over all scenarios is missing: the type-level rules and multiplicities, which lifecycle transitions are legal, what the full set of user goals is, and what is co-located.

The diagram has slid into functional decomposition: those are steps, not goals. Test each entry by asking whether a primary actor would walk away satisfied when it completed, then collapse the steps into the handful of goals that pass and push the rest into the written flows.

Buy modelling where ambiguity is concentrated and expensive, rather than working through the taxonomy. Most teams need two or three families. Draw in answer to an open question, treat most diagrams as sketches, and keep only what is costly to rediscover.

An activity final node ends the whole activity: reaching it stops every flow still running. A flow final node ends only the one flow that arrives at it, and any concurrent flows carry on unaffected.

Shallow history remembers only which substate of its own region was last active. Deep history remembers the active configuration all the way down through every level of nesting. Both restore that memory when the composite state is re-entered.

Generalization is a solid line with a hollow triangle pointing at the more general element. A specialised actor can participate in everything the general actor can, plus its own; a specialised use case is a complete variant that can substitute for the general one.

A class diagram earns its keep when the shape is spread across many files or does not exist yet, and when a reader outside the code must follow it. Reading the source is faster whenever the diagram would restate one file.