A Primavera P6 schedule travels as an .xer file: tab-separated tables of activities, links, calendars and settings. Owners' reps, lenders' monitors and subcontractors receive them every month, and many of them have no P6 seat to open one. The usual quality screen is the DCMA 14-point assessment (missing logic, leads, lags, hard constraints, high and negative float, long activities, the critical path test, CPLI, BEI and the rest), and the tools that run it are mostly paid desktop or subscription products.
XER Schedule Check is a free page that runs the 14 points in the browser and lists the exact activities behind each failing one. Two of the points cannot be done by reading the file alone. The critical path test delays an activity on the longest path and asks whether the project finish moves, which needs a full critical path recompute; and a reviewer wants to know whether the float stored in the file still follows from its logic. So the page carries its own CPM engine, and the useful question is whether that engine computes what P6 computes.
The check
Every XER stores, per activity, the early and late dates and total float that P6 wrote at its last schedule run. Six public XER files (from four GitHub repositories, exported from P6 5.0, 6.0, 7.0, 20.12 and the 24.12 web client; 19,822 open activities) give a direct test: recompute, and count the activities where early finish, late finish and total float all equal P6's to the minute.
| File | Open activities | All three equal |
|---|---|---|
| Expressway (7.0) | 823 | 823 |
| Hotel, 4 projects (6.0) | 4,217 | 4,216 |
| Example (24.12 web) | 404 | 404 |
| Terminal building (5.0) | 66 | 66 |
| Sample (20.12) | 42 | 42 |
| CCA programme (6.0) | 14,270 | 12,582 |
| Total | 19,822 | 18,133 (91.5%) |
What it took
The first pass, a textbook forward and backward pass on each activity's own calendar, matched 10,701 activities. Each rule below came from one mismatch traced back to the first activity that went wrong, and was kept only if the totals went up:
- Planned start. An unstarted activity never starts before the project's planned start, even when the data date is two weeks earlier. One rule: 0 to 823 of 823 on the Expressway file.
- As late as possible. The constraint spends the activity's free float, worked from the last activity back so a chain of them moves together.
- Milestones across calendars. A finish milestone on an 08:00-16:00 calendar after a predecessor finishing at 17:00 on another calendar lands at 08:00 next morning, not at 16:00 the same day, which would be before its predecessor finished.
- Completed activities stay in the network. P6 keeps a finished activity as a zero-length node at the data date. If an unfinished predecessor (progress out of sequence) pushes it, the push carries on to its successors. A finished predecessor's lag still runs from its actual finish: a seven-day cure after a pour finished yesterday still holds the next activity.
- Start-to-start links into started work. A started activity's remaining work waits for a start-to-start predecessor's remaining work. If that predecessor has started, the lag is already used up; if not, the lag still applies. The backward pass mirrors this rule.
On the large programme those last two rules took the count from 10,190 to 12,582. The 1,688 activities still off all sit among its 2,134 out-of-sequence links. They are started activities with a mix of finish-to-start and start-to-start predecessors that are also started, where P6's retained-logic rule is not yet reproduced. The full trail, with the count after each rule, is in the eval note.
What the 14 points found
None of the six files passes all fourteen. The lightest schedule (52 activities) fails only on relationship types (57% finish-to-start, against a 90% floor) and resources. The 16,000-activity programme passes five. 62% of its open activities carry more than 44 working days of float, it has 544 with negative float, and 2,134 links were progressed out of sequence. Those numbers are not a verdict on the projects. The points are screening metrics, and some failures have good reasons. The value is in the lists: a contractor can be sent the 105 Expressway activities missing a predecessor or a successor (59 with no successor), not just "fails point 1".
The page is free, MIT licensed and nothing is uploaded. The next step is the residual retained-logic rule, then an update-to-update comparison (what moved, which float was used up, which logic changed), the question a monthly pay-application review asks.