A list of tasks is a wish. A sequence is a claim about the world: what must finish, what may overlap, and what is waiting on concrete to cure. Get the dependencies wrong and your Gantt chart will still look decisive. It will also be fiction.
The four dependency types — FS, SS, FF, SF
Finish-to-Start (FS) is the default: B cannot start until A finishes. Design finishes, build starts. Most schedules are a pile of FS links because that is how people think when they are tired. Use it when it is true. Do not use it because the software offered it first.
Start-to-Start (SS) means B cannot start until A has started — laying cable can begin once trenching has begun, not only when the whole trench is done. Finish-to-Finish (FF) means B cannot finish until A finishes — you cannot close testing until the last fix is in. These two are how you model genuine overlap instead of pretending work is a single-file queue.
Start-to-Finish (SF) is rare and easy to fake. B cannot finish until A starts: the night shift cannot stand down until the day shift has begun; the legacy system cannot be decommissioned until the replacement is taking traffic. If you think you need SF, write the sentence in English first. If the sentence is really FS wearing a hat, use FS. Clever networks that nobody can explain will not survive the first change request.
Leads, lags, and the temptation to hide time
A lag is waiting time on a link. FS plus five days of lag: the successor starts five days after the predecessor finishes. Use it for things the calendar owns — concrete curing, statutory notice, a vendor's contractual review window. Do not use lag as a drawer for 'we are not sure,' or you will lose the uncertainty inside a number that looks like a fact.
A lead is an overlap: a negative lag. FS with a two-day lead lets B start two days before A finishes. That is a decision to accept risk — you are building on work that is not done. Write why the lead is safe. A schedule full of undocumented leads is a fast-track you have not admitted to, and it will take its payment in rework.
Prefer an explicit activity over a mysterious lag when someone has to do something in the gap: 'procurement approval,' 'environment refresh,' 'legal redlines.' Lags do not get owners. Activities do. Owners are how waiting time gets managed instead of merely displayed.
Build a network, not a pretty list
The precedence diagram (activities on nodes, arrows as dependencies) is the model underneath the Gantt chart. Start with the work packages, add the logic, then apply durations. If you start with dates and work backwards to 'dependencies,' you are drawing a picture of a promise, not a network of work.
Every activity except the start needs a predecessor; every activity except the end needs a successor. Dangling tasks are how work falls out of the critical path calculation and then surprises you from a side alley. Open ends are not flexibility. They are unanalysed work.
Sequence from the work, not the org chart. 'Design then the whole of engineering then the whole of test' is a departmental parade. Real products have components that can move in parallel and integration points that cannot. The network should embarrass false sequential thinking. If your diagram is a single snake, either the work is truly serial or you have not yet asked 'what could start once this has started?'
Summary & next steps
Model the true relationship: FS, SS, FF, or the rare SF. Use lags for real waiting; use leads only as a named risk. Prefer owned activities to anonymous delays. Build a closed network from the work, then hang dates on it — not the other way round.
Next: the critical path method — the longest path, the two kinds of float, and why shortening the wrong task is an expensive hobby.
Practice this
WBS / Project Schedule
Turn theory into a real deliverable. Generate a wbs / project schedule with the AI tool and adapt it to your own project.
Open the generator