Construction schedule delay analysis almost always starts from the same document: the float table. Which activities had slack, how much, and therefore who owes whom the days. It is the number that decides whether a delay was excusable, concurrent, or somebody’s fault, and it is on the front page of every claim I have ever read.
So I tested it. I took a 21-task renovation with three general crew, delayed every single task a day at a time, and re-solved the whole schedule after each one to see when the finish date actually moved. Twenty-one tasks, a few hundred solves. Three of them turned out to have nine days of float on paper and no tolerance at all.
What the test was
The plan: 21 tasks, five trades, three general crew, two electricians, two plumbers, two painters, one excavator. Its critical path finishes 25 November 2026. Scheduled so no day ever needs more people than exist, it finishes 2 December. That second date is the one a delay analysis is really about, because it is the one in the contract.
For each task in turn: push its earliest start one working day later, re-solve the entire schedule optimally, and check the finish. Still 2 December? Push it another day. Keep going until the finish moves. Whatever number you stopped at is how much lateness that task can genuinely absorb — not in theory, not holding everything else frozen, but with the whole plan free to rearrange around it.
That is a deliberately generous test. It gives the schedule every chance to recover. If a task still cannot absorb a single day under those conditions, it really cannot absorb a day.
The three that fail
Roofing, driveway and landscaping each carry nine days of total float. Each absorbs zero. Lose one day on the roof and 2 December becomes 3 December, while the float table sits there saying you had over a week in hand.
The reason is not subtle once you see it. Total float is computed on the dependency network alone, and on that network roofing genuinely can start any time in a nine-day window without troubling its successors. But roofing needs three general crew, and in the schedule that actually gets built, those three people are booked solid on either side of it. There is no window. The float exists in a plan where labour is free, and nobody is building that plan.
Driveway and landscaping fail the same way and for the same reason — both are general-crew work sitting in a stretch where the crew is the binding constraint, not the logic.
It runs the other way too
The interesting half is that float is not simply optimistic. It is uncorrelated.
Rough electrical has zero total float — it is on the critical path, the sort of activity a claim would treat as automatically compensable — and it absorbs three days without moving the finish. Rough plumbing shows one day of float and absorbs four. Windows shows two and absorbs five. Interior painting and flooring both show zero and absorb two each.
Eight of the twenty-one tasks tolerate more delay than their float suggests, mostly because the crew-constrained finish already sits five working days beyond the critical path date, and that gap is real slack that the CPM table never accounts for. Three tolerate dramatically less. Only a handful match.
If you are using float as the test for whether a delay was critical, you are wrong in both directions on this plan, and you have no way of telling which kind of wrong you are looking at without doing the experiment.
Why the standard methods do not catch it
The recognised forensic methods are catalogued in AACE International’s recommended practice 29R-03, Forensic Schedule Analysis — as-planned versus as-built, impacted as-planned, collapsed as-built, windows analysis, time impact analysis. They are careful, well-argued, and widely used in arbitration, and every one of them is built on a critical path model.
Which is fine when labour is not the constraint. On a job where the crew is what everyone is actually fighting over, inserting a delay fragnet into a CPM network and reading off the new critical path answers a question about dependencies when the argument is about people. The methods are not wrong. They are answering the question they were designed for, and on a resource-constrained job that is not the question in dispute.
None of this is exotic. It is the same gap I keep running into across this series: float measured on the network and room measured in a crew-constrained schedule are different quantities, and the second one is the one that governs a real site.
Two numbers, and the difference between them
Worth being precise, because there are genuinely two different questions here and they have different answers.
- How far can this bar slide without disturbing anybody? That is a property of the schedule as drawn — slide it further and it shoves the next crew hand-off. Useful when you are looking at the chart and asking whether you can move something today.
- If this task runs late, can the job still finish on time? That allows everything downstream to be re-planned around the problem. It is the delay-analysis question, and it is the one I measured above.
The second is always at least as large as the first, and on this plan they differ on two tasks: exterior painting can slide six days but tolerate fifteen, and windows can slide none but tolerate five. Both facts are true. If somebody quotes you a tolerance figure in a dispute, it is worth asking which of the two they computed, because the answers are not interchangeable and the gap can be nine days.
Running it on your own schedule
You do not need forensic software for the version of this that is useful before a dispute rather than during one. The test above is a loop anyone can run:
- Solve the schedule inside your real crew numbers and note the finish.
- Pick a task. Set it to start one day later. Re-solve. Did the finish move?
- Repeat until it does. That number is the task’s real tolerance.
- Do it for every task on the job and compare the column to your float column.
The free scheduler takes a task list with durations, predecessors and a crew per day, plus a sheet of what you have, and returns a proved-optimal schedule in a fraction of a second — which is what makes the loop practical, because a few hundred re-solves is a couple of minutes rather than an afternoon. It also matters that each answer is proved rather than approximated: when seven commercial packages were benchmarked on this kind of problem, the median schedule came back 3.6% above optimal, and a tolerance figure measured against a schedule that was not optimal in the first place is not measuring much. If your plan is still in a spreadsheet, the construction schedule template with crew sizes is the same workbook with the trades already filled in.
Do it before the job starts and you get a list of the tasks where a single lost day costs you a date — which on this renovation was the roof, the driveway and the landscaping, none of which any float table would have flagged. That list is worth more on a Tuesday morning than the whole forensic exercise is worth eighteen months later.
One honest limit. This is one plan, and a fairly crew-heavy one. On a job with deep float and slack trades the float table and the real tolerance will agree, and everything above becomes an academic point. The test tells you which kind of job you have, and it takes about a minute. That seems like a reasonable trade for finding out whether the number your entire delay position rests on means anything.
Things that I use, like, and am affiliated with:
Mint Mobile offers great cell phone service for $15 flat, get $15 off using the link. Get discounted phones with service activation and no contract.
I never spend money before I check Mr Rebates or Rakuten to get cashbacks, rebates, discounts, coupons or cheaper gift cards.