Why a Collapsed As-Built Can Show a Negative Delay

What a Collapsed As-Built Actually Compares

A collapsed as-built analysis takes the completed record, removes the time consumed by a chosen delay event, recalculates the network from logic and each activity’s as-built duration, and reports how much earlier the project would have finished. The natural way to score it is subtraction: the real as-built finish minus the recalculated “but-for” finish.

That subtraction quietly assumes something: that if you recalculated the as-built schedule with nothing removed, you would land exactly on the real as-built finish. On real projects you often don’t.

Why Actual Dates Don’t Rebuild From Logic

• Out-of-sequence work — activities started before their logical predecessors finished, so the recorded dates are earlier than the logic allows.
• Waiting that isn’t in the logic — an activity that sat idle for weeks for a reason no relationship captures, so the recorded dates are later than the logic implies.
• Calendar and duration effects — an as-built duration measured across non-working time doesn’t always convert back to the same dates.

Any of these opens a gap between the real finish and a logic-only rebuild. That gap exists whether or not anything was removed, and it has nothing to do with the delay being analyzed.

A Real Example of the Problem

While testing Claims Digger’s own collapsed as-built engine against a large test schedule whose recorded dates did not follow its logic, removing a single 40-hour activity produced a delay of −83 calendar days — the project appeared to finish nearly three months later without the activity. The logic-only rebuild of that same schedule with nothing removed already landed 90 calendar days away from the real finish. The 40-hour activity was innocent; the rebuild baseline was off.

The Fix: Measure Against a Rebuild, Not Against the Record

Run the same rebuild twice — once with nothing removed, once with the delay removed — and score the delay as the difference between the two rebuilds. Whatever systematic gap the method introduces appears in both and cancels. Then report the gap between the real finish and the unmodified rebuild separately, so a reader can see how far the logic and the record disagree.

The corrected figure matches what you’d expect: removing a 40-hour activity from the critical chain shortens the project by 40 working hours, which is seven calendar days across a weekend.

Frequently Asked Questions

Does a negative result mean the delay wasn't real?

Not by itself. It means the number was computed by comparing two things that were never comparable -- real recorded dates on one side, a logic-driven recalculation on the other. Until the comparison is made like-for-like, the sign of the result says nothing about whether the event delayed the project.

Should the removed activity be deleted from the network or set to zero duration?

Both are used in practice. Deleting it forces the analyst to decide how its predecessors and successors should be re-linked, and a real activity with several of each has no single obvious answer. Setting its duration to zero while keeping its relationships intact preserves the original sequencing and removes only the time it consumed, which is easier to explain and reproduce.

What if the schedule isn't finished yet?

A collapsed as-built needs actual dates to work from. On a schedule that is only partly complete, only the activities with recorded finishes can be measured, so the result describes the work completed so far, not the project's eventual finish. Say so explicitly, and rerun the analysis when the record is more complete.