The 118-Day Schedule Nobody Believes

Sort the four features by whether they push duration up, hide float, or both.

1. Dangling 'Final QA' (no successor) - a modeling error, but a neutral-to-low one on duration. A dangling activity is not connected to the finish milestone, so the tool may exclude it from the driving path. Effect: duration looks shorter and the activity's float looks meaningless. Fix: link it to the project finish milestone.
2. Hard constraint on 'Permit approval' - inflates apparent float and suppresses real float. A hard constraint overrides logic, so downstream activities get float values that reflect the constraint date rather than the network. This also explains the 31 zero-float activities: constraints propagate zero float backwards and forwards and make the float field uninformative.
3. Rework ↔ Review loop - a circular logic error. Tools either flag it or silently break the loop; if silently broken, sequencing follows the wrong order and duration is unpredictable. This must be removed before any calculation is trusted.
4. Missing 5-day overlaps on three design tasks - pure inflation. Modeling three overlap-able tasks as sequential adds roughly 2 × 5 = 10 schedule days of false duration (and more if the overlaps compound through successors).

Net: the loop makes 118 uninterpretable; sequentialization inflates it; the dangling activity and constraint mask truth. The 23-day gap to the engineers' 95-day estimate is more than fully explained by items 2 and 4 plus schedule lag, which is why the file cannot be submitted as-is.

No. A hard constraint is a declaration, not a de-risk. It does not move work; it moves numbers.

Why the float report is misleading under a constraint:

• Total float is computed as LS − ES, and a constraint fixes LS (or LF) for the constrained activity and everything downstream of it. The float values then describe the gap to an arbitrary date rather than the gap to network logic.
• When the constraint is earlier than the logic-driven date, float turns negative, which many viewers read as 'zero' or 'critical' and stop. When it is later, activities show generous float that does not exist.
• Float on constrained activities no longer identifies where to manage.

What to do instead:

1. Remove the hard constraint from the activity and attach a deadline or a soft constraint (start-no-earlier-than / finish-no-later-than) so logic still drives dates and the constraint only flags violation.
2. Model the permit as a real activity or external dependency with its own duration and a finish-to-start or start-to-start link to the work it gates, so its uncertainty is visible.
3. Re-run the backward pass and confirm the constrained date shows as negative float on the affected path - that is the honest signal, and it is actionable.

Rule of thumb: constraints belong on deliverables and milestones that are genuinely externally imposed; they do not belong on ordinary activities to 'make the plan work.'

Order matters - later checks are meaningless until earlier ones pass.

1. Close all loops. Find and eliminate the Rework/Review cycle. Until the network is a DAG, no date or float is valid.
2. Connect all dangling activities to a successor or the finish milestone, and check for orphan activities at the start with no predecessor.
3. Re-express the three overlapping design tasks using start-to-start links with lag (e.g., SS+5d) rather than sequential FS, and confirm the overlap is physically available (same team? same inputs?).
4. Remove or demote hard constraints and re-run to see the logic-driven finish date.
5. Re-run the forward/backward pass and reconcile the new duration against the engineers' 95-day estimate. A gap that survives these fixes points to a genuine estimating or logic disagreement and is worth a working session; a gap that disappears was modeling noise.
6. Only then run risk analysis.

Diagnostic value: if the corrected network lands near 95 days, the 118 was artifactual and the file can be fixed in an afternoon. If it still sits at 110+, there is a real scope or sequencing problem and the constraint was hiding it.