Schedule development and analysis

A schedule is a model of how work connects. Most schedules in circulation are a list of dates someone typed, which looks the same in a status deck and behaves completely differently when anything moves.

Build it from logic, not from dates

Every task except the first and last should be connected to something. A task with a date and no predecessor is an assertion; a task with a predecessor is a plan, because when the predecessor moves the task moves with it and the schedule tells you so.

• Finish-to-start is the default and covers most real dependencies.
• Start-to-start and finish-to-finish describe overlapping work honestly and are worth using rather than faking the overlap with an earlier date.
• Lags should represent something real — curing time, a notice period, a review window. A lag used to make dates line up is a hidden assumption.

Hard-coded date constraints are where schedules go to die. Each one severs the logic at that point, so slippage upstream stops propagating and the schedule keeps reporting a date it can no longer support.

Float is the information, and the critical path is the consequence

Float is how long a task can slip before it moves the end date. The critical path is the chain with none. Attention belongs there and almost nowhere else — expediting work with three weeks of float buys nothing and consumes the same management capacity.

Two readings worth taking regularly: near-critical chains, which become critical after one bad week and are where surprises come from, and negative float, which means the schedule is already telling you the date is impossible and someone has not looked.

Load resources before you believe the dates

A schedule that assumes the same three people work on five parallel tasks is fiction that will resolve itself as a delay. Load the named constraint — the one team, the one approver, the one environment — and level it. Leveling extends the schedule; that extension was always there, it was just unrecorded.

Analyze schedule risk rather than adding a buffer at the end

Durations are ranges. Running the schedule with ranges rather than single durations shows something a buffer cannot: paths that merge. Where several chains converge on one milestone, the milestone waits for the latest of them, so the probability of hitting the date is far lower than the probability of any single chain hitting its own — which is why plans with many parallel workstreams slip even when no individual stream looks late.

A buffer at the end of the plan protects the end date. A buffer in front of each merge point protects the plan.

Measure progress against the baseline, not against effort

Keep a baseline and compare to it, or "we are 70% done" has nothing to be 70% of.

The useful comparison is three-way: what was planned to be complete by now, what is actually complete, and what it cost to get there. Those three answer different questions — the first two give you schedule performance, the last two give you cost performance, and a project can be ahead on one and badly behind on the other.

Percentage complete on in-progress work is where reporting rots. A task reported at 90% for three weeks is not 90% complete. Binary reporting — not started, in progress, done — with small enough tasks that "done" arrives often is more honest and takes less time to maintain.

Auditing a schedule you inherited

Look for these in order, because each one invalidates everything after it: tasks with no predecessor or successor, hard date constraints, negative float, durations longer than a reporting period, tasks that have been 90% complete for more than one period, and a baseline that has been re-set so often it no longer means anything.

Sources

references/sources.md in this skill lists the outside authorities that settle the questions here — what each one is authoritative for, and what you may do with it. Check them before answering on anything they cover, and cite what you used. Most are free to read and not free to reproduce; the use note on each is binding.

Never

• Accept a date on a task with no logic behind it.
• Hard-code a constraint to make a date hold. It stops the schedule from telling you the truth.
• Report percentage complete on a task whose remaining work nobody has re-estimated.
• Re-baseline to make variance disappear. Re-baseline when scope genuinely changed, and say which.