A schedule that ignores who does the work is a thought experiment. Resource leveling is the moment the thought experiment meets the timesheet. The network may still be right. The dates often are not.
Overallocation is a fact, not a motivational opportunity
If the same engineer is assigned to two full-time critical activities in the same week, you do not have a commitment problem. You have a model that assumed a 160-hour human. Tools will show a nice peak. Reality will show a quiet delay on one of the two tasks and a person who stops answering Slack.
Find overallocation before you baseline. Histogram the scarce roles — the architect, the tester, the only person who understands the general ledger. Peaks above capacity are decisions you have not yet made: slip a task, split work, add a person, or cut scope. Leaving the peak in the plan is how you discover the decision in week six, with less dignity.
Named resources hide the problem better than generic roles. 'Developer' at 300 percent is obvious. 'Priya' at 100 percent on paper and 100 percent on a production-support rota is not in your file. Ask what else they do. The calendar you were not shown is still part of the project.
Leveling versus smoothing
Resource leveling resolves overallocation by delaying work until a person is free. The end date may move. That is the honest trade: you can have this logic and these people, or you can have that date, not both as a matter of willpower. Leveling is a schedule change. Treat it as one.
Resource smoothing (sometimes called resource allocation within float) reshuffles work to flatten peaks without moving the project finish. You spend free float and unused total float. When the float runs out, smoothing cannot finish the job. Anyone who says 'just smooth it' on a schedule with no float is asking you to level and lie about it.
Do not confuse either with 'smoothing the politics.' Spreading a person across six projects at 0.2 FTE each is not leveling. It is fragmentation. The switching cost will eat the 0.2. Prefer fewer assignments and a clear priority when two managers want the same Tuesday.
After you level, the critical path may change
Resource-constrained scheduling creates a new kind of criticality: a task is late because the person was elsewhere, not because the predecessor was late. The path that matters is now the chain of scarce people. Recalculate. Report the new date and the new critical work. A baseline taken before leveling is a souvenir.
Watch for leveling that creates idle gaps you could have filled with discretionary work — documentation, refactoring, a spike. Smoothing leftover troughs is useful. Inventing busywork to make a utilisation chart look 'efficient' is how you destroy float you will want when the vendor slips.
Document the rule you used. Priority to the critical path, then to near-critical, then to the work the sponsor named. If leveling is a late-night argument in the tool's options menu, nobody will be able to defend the date. A one-line rule in the schedule notes is how you survive the meeting where someone asks why their favourite task moved.
Summary & next steps
Overallocation is a modelling error until you decide. Leveling delays work and may move the finish. Smoothing uses float and cannot invent capacity. After either, recompute the path — resource constraints rewrite criticality.
Next: schedule compression — crashing and fast-tracking, and the quality and risk bill that arrives after the date looks better.