Granular attribution is one of the most useful promises of computational delay analysis. Instead of saying that logic changes moved a milestone by 30 days, we want to identify the relationship, activity, or duration change that caused the movement. Usually, the schedule supports that level of detail.
Some network revisions place a hard limit on it. Their component changes are interwoven so tightly that applying one without the others produces a cyclic network or an arbitrary intermediate state. In those cases, a smaller measurement would require a hypothetical schedule that could never have been calculated.
In our earlier post on structural impacts in an evolving network, we described why changing topology makes attribution so difficult. This is the next question: what should the analysis do when the network itself says that several changes belong together?
Start With the Smallest Possible Swap
Consider two activities. In the base schedule, A precedes B. In the revised schedule, their order is reversed.
One revision, four possible network states
The reversal appears to be two granular changes: delete A→B and add B→A. Apply the addition against the base schedule, however, and both relationships exist at once. A drives B while B drives A. The resulting loop has no valid CPM dates, so there is no milestone movement to measure.
Applying the deletion first avoids the loop, but creates a different problem. A and B are temporarily disconnected. That network may be calculable, yet nothing tells us the planner intended it as a schedule state. Its dates depend on open ends and other logic that the complete revision was meant to replace.
Choosing one edit order would therefore create the attribution. It would not discover it. An invalid network has no CPM result, and an unintended network has no defensible factual basis. The complete reversal is the smallest coherent unit available for measurement.
Think About the Moment You Press F9
This has a familiar parallel in P6. A scheduler rarely treats every relationship edit as a finished scheduling decision. Reworking a sequence may require removing an old tie, adding its replacement, reconnecting surrounding work, and checking the final logic before running the schedule calculation.
- − Remove the old relationship
- + Add the replacement
- + Reconnect surrounding work
P6 calculates the completed revision as one network. Its dates reflect the full set of coordinated edits.
The meaningful boundary is the network presented when the scheduler presses F9. A half-edited network may briefly exist on screen, but it does not become a calculated update unless someone chooses to schedule and preserve it. Forensic analysis should be equally careful about inventing calculations between coordinated edits.
Let the Network Set the Measurement Grain
Most structural changes remain independently measurable. A single deleted tie can release downstream work. A new activity can extend a path. A replacement relationship can change the date at which a successor becomes available. When each change has a coherent individual measurement state, FPM preserves that granularity.
Coordinated revisions receive different treatment. When additions and deletions depend on one another to keep the network valid, FPM measures their combined effect as a single event called Network restructure. The underlying changes remain available for inspection, while the impact belongs to the complete revision.
The change can stand alone in a valid network, so its milestone effect can be measured on its own.
The changes need one another to form a coherent network, so their milestone effect is measured together.
Once the restructure has been measured, the coherent revised structure becomes the context for the remaining changes in that update. Progress, duration changes, and separable logic revisions are then evaluated in the network where they actually belong.
What the Analyst Can Defend
Grouping interwoven changes does not make the evidence disappear. The analyst can still inspect the relationships and activities that compose the restructure, identify when they entered the schedule, and trace the combined effect to the affected milestones. What the analysis declines to do is assign a portion of that effect to an arbitrary intermediate network.
That changes the forensic question in a useful way. Instead of asking which keystroke caused which fraction of the movement, the analyst can ask who directed the sequence revision, why it was introduced, whether it was approved, and what effect the complete decision had on the project.
More rows do not automatically mean more truth. Granularity is valuable only while each measurement describes a coherent schedule state. Once the network stops supporting that state, further division creates false precision.
A defensible analysis follows the same discipline as a scheduler preparing to press F9: assemble a valid network first, then calculate its effect. When several structural changes must travel together, the impact should travel with them.