P6 compatibility: settings, import, export and what round-trips
How closely os.construction matches Primavera P6. Which scheduling settings the engine honours, which P6 calculated fields it works out, what crosses an XER or P6 XML file in both directions, the measured agreement with P6's own dates and costs, and the limits that remain.
On this page
Most schedules you’ll receive come from Primavera P6, and many owners still want an XER back. This page says how far os.construction goes with P6 today. It covers the settings the engine honours, the P6 calculated fields it works out, what crosses a file in each direction, how closely the calculated dates and costs agree with P6, and what doesn’t carry yet. Every figure here comes from the product’s own test evidence, measured on 2026-11-10 and, for costs, units and earned value, on 2026-11-11. How we verify explains that evidence and its limits.
How closely the dates agree
The engine is checked against P6’s own stored answers, not against hand-made expectations.
| Measure | Result |
|---|---|
| P6 23.12 cross-validation fixture (13 single-purpose projects: every relationship type, negative float, several calendars, holidays, retained logic, out-of-sequence progress, mandatory and ALAP constraints) | 156 of 160 field comparisons agree. The 4 misses are one presentation convention: P6 writes a started activity’s actual start into its early and late start. |
| License-safe corpus of real P6 exports (39 projects, 26,337 comparisons), default import settings | 87.6% agree (23,083). That rises to 98.8% once you exclude cells P6 fills by a provable stored-row convention rather than by calculation. |
| The same corpus with whole-day precision forced | 81.5%. Automatic time precision recovers the difference, and no project gets worse. Auto now gives exactly the same answer as forcing Minutes on every corpus project. |
| Probe projects built and calculated in P6 Professional 23.12.1 (13 projects that isolate open ends, resource-dependent work, suspended work, expected finishes and other edge cases) | 1,962 of 1,968 values agree (99.7%), up from 1,844 and, before that, 1,620. Every value of the four Make open-ended activities critical projects and of the two expected-finish constraint projects matches P6. The six that don’t are one start-to-finish pair, printed on the other side of a non-working night; the float on both rows agrees. |
| Export round trip (import, calculate, export XER, re-import, recalculate) | 39 of 39 projects produce identical dates, float, criticality, longest path and finish. |
| Costs, units and earned value against the values P6 stored (34 projects: assignment prices and costs, unit roll-ups, expenses, actual this period, stored period earned and planned value, P6 XML activity and assignment roll-ups) | 67,756 of 68,660 agree (98.7%). Excluding values P6 stored by a provable convention, 99.94% agree; the remaining 41 have named causes on our side. |
20 of the 39 corpus projects agree on every cell. The remaining disagreement is concentrated in a few files. Some stem from P6 conventions for milestones and summary rows, and some from the gaps listed under Limits. See How we verify for where these numbers come from.
Scheduling settings
Open Schedule › Settings to see the project’s scheduling options with P6’s names and defaults. An imported XER or P6 XML file brings its settings with it. A new plan starts from the engine defaults, so check them before you compare dates with P6.
Honoured by the engine, with P6-equivalence tests:
-
Progress mode: retained logic, progress override and actual dates.
-
Relationship lag calendar: predecessor, successor, project default or 24-hour.
-
Define critical activities as total float at or below a threshold in hours, or longest path.
-
Multiple float paths, numbered after the calculation. P6 doesn’t publish its ranking rule, so the numbering is a documented approximation.
-
Make open-ended activities critical, measured in P6 23.12.1: an unfinished activity with no successor of any kind takes its own early finish as its late finish, so its float drops to zero or below and the work driving it follows. Activities with no predecessor are left alone, as in P6. The edge cases now match too:
- Completed open ends. Under retained logic, completed work stays a point in the backward pass and hands back the latest date its successors allow, no later than the project finish. With the option on, completed work with no successor hands back the date its own logic carries. Under progress override nothing passes through it.
- As Late As Possible open ends slide to the project’s early finish, while their late finish stays capped at the finish they had before sliding, so they show negative float as in P6.
-
Use expected finish dates, including how an expected finish interacts with the other constraints, measured in P6 23.12.1:
- On work not yet started, an expected finish drops the early side of Finish On or After, Finish On and Mandatory Finish. The work starts from its logic and finishes on the expected finish. Start constraints keep their early side.
- On work under way, the late side of Start On or Before, Start On, Mandatory Start and Mandatory Finish no longer applies, as in P6. Finish On and Finish On or Before still cap the late dates.
- No free float on work whose finish an expected finish holds, or on unstarted work under a Mandatory Start or Mandatory Finish. P6 stores zero free float on those activities, and the engine now does too.
-
Ignore relationships to and from other projects.
-
Suspend and resume. Started work with a suspend date and a resume date after the data date restarts its remaining work on the resume date, as in P6. A resume date with no suspend date, or one on or before the data date, changes nothing. The schedule log notes each activity that resumes.
-
Must finish by, the project’s default calendar, work weeks, shifts, holidays and part-day calendars.
-
All ten constraint types, primary and secondary.
-
Resource-dependent activities, on one resource calendar or several. Each driving assignment is scheduled on its own resource’s calendar for units ÷ units per hour (the remaining rate once started). The activity starts with its earliest assignment and finishes with its latest. The work opens at the activity calendar’s next working time, then each assignment at its resource calendar’s next working time, and total float is counted on the activity’s calendar. Also honoured:
- Assignment lag. An assignment’s planned lag, and its remaining lag once started, delays that assignment’s start on its resource calendar.
- Drive activity dates. Only assignments with P6’s Drive Activity Dates flag set the activity’s dates. With none driving, the activity keeps its own calendar and duration.
- Finish-to-finish and start-to-finish predecessors set a finish on the activity calendar, and each assignment has to reach it on its own calendar.
Assignment lag and the driving flag come in from XER and P6 XML and go back out on export. Edit them on the Activity Details Resources tab or in the Assignments view.
-
Units % complete: imported, kept and exported, and editable on the Activity Details Status tab and as a grid column. A new units % complete re-spreads each assignment’s actual and remaining units, and leaves remaining duration where it is, as P6 does.
-
Leveling options: leveling priorities, Level resources only within activity total float, Preserve minimum float, Max percent to over-allocate, Level all resources or a chosen list, and Preserve scheduled early and late dates. They shape the leveling proposal.
-
Earned value settings: the six earned value techniques and five ETC techniques, for the project on the Settings page and per WBS in the Planner’s WBS window, and When calculating earned value from a baseline use (P6’s administrative earned value preference: budgeted values with current dates, budgeted values with planned dates, or at completion values with current dates). See Period performance and earned value.
Duration type and % complete type aren’t inputs to P6’s calculation either. They decide what’s recalculated when you edit units, duration or units per time, and the Planner applies P6’s rules there. See Duration types.
Stored and exported, but without a scheduling effect yet: schedule automatically, level resources during scheduling, recalculate costs after leveling, project defaults, calculation flags, and leveling across other projects’ assignments. Edit them on the Settings page and the export writes your value. Leave them alone and the export writes back exactly what was imported.
Worth checking:
- Total float method. P6 defaults to finish float, but a new os.construction plan defaults to start float. All three methods are now checked against values P6 calculated, including probe projects set to start float and to the smallest of start and finish float (328 of 328 values agree on the open-end probe sets). They give the same float on most networks, but set the method to match the P6 project before you compare.
- Calculate start-to-start lag from early start or actual start. It’s carried and matches P6 on the 46 real links tested, but it only changes dates when retained logic holds a started predecessor’s remaining work.
- Hours per week, month and year and the start day of the week aren’t represented.
The settings inventory behind this page lists every P6 project setting with its status and the test that covers it. Leveling and earned value settings moved from stored only to applied in this release.
P6 calculated fields
Besides dates and float, the engine and the Planner work out the values P6 calculates and shows in its columns, using P6’s published definitions:
- Remaining early and late dates, start float, finish float and remaining float, actual and at completion duration, and duration % complete, which is always (original − remaining) ÷ original, as in P6.
- Schedule % complete, measured against the project baseline, else the planned dates.
- Cost and units roll-ups by resource type (labor, nonlabor, material and expense), with Total Cost columns that include activity expenses, as P6’s do.
- Cost and units % complete, % of planned, variances against the project baseline and against BL1 to BL3, the earned value indices (CVI, SVI, TCPI, Accounting Variance) and the labor units earned value family. See Earned value columns.
Where a P6 export stores the value, the cost oracle above scores it. XER stores no activity earned value fields (P6 derives them for display), so most indices and variances are checked by tests that write out Oracle’s definition rather than against stored P6 values. A few values aren’t calculated yet: scope % complete (a P6 EPPM scope feature), staffed and unstaffed remaining units and cost (role-only assignments aren’t imported), relationship early and late dates, the critical path number and the relationship float path.
Import
Upload a P6 XER, P6 XML or Microsoft Project XML file of up to 32 MiB at Schedule › Import file. See Import from P6 and Microsoft Project for the step-by-step flow. On adoption into the Planner, the engine reads activities, WBS, calendars, relationships with lags in hours, constraints, progress, resources and assignments, activity codes, user-defined fields, notebooks, steps and the scheduling options. It then calculates dates itself rather than trusting the dates stored in the file. P6 XML calendars write each shift’s end as its last working minute (for example 16:59); the importer reads that as the full shift and the export writes it back the same way, and a P6 XML lag is converted on the successor’s calendar, as from XER, so both formats schedule the same.
Columns that the canonical plan doesn’t model are kept as retained P6 metadata with the plan, so they can be written back on export. Retained metadata has a budget. On very large files it’s trimmed, and then dropped, before it could push the plan past its size limit, and the import says so.
Microsoft Project .mpp files aren’t supported; save them as XML first.
P6’s import options are available once the project has a saved plan: Update existing plan with an Update Existing, Insert New, Keep Existing or Delete Existing action for each object type, Replace existing plan, and Add into WBS node. Activities match by retained P6 GUID, else by Activity ID, and matched objects keep their identity. You review a preview before one new version is saved. See Update the current plan from a file.
Export
Export the current plan, or any saved version, as P6 XER, P6 XML or MSPDI. Use the Export XER and Export P6 XML buttons on the schedule Overview, Export Current Plan on an import’s review page, or the Schedule API. Export needs Schedule admin, because the file discloses the whole network. The version is recalculated first, and nothing is written.
Calculated values are written fresh, never copied from the source: dates, float, status, unit roll-ups, the remaining early and late dates of unfinished work (XER restart_date, reend_date, rem_late_start_date and rem_late_end_date, and P6 XML’s RemainingEarly* and RemainingLate* dates, blank on completed work) and the activity type, so an activity flagged resource dependent in the plan exports as Resource Dependent. Planned dates are deliberately left as they are: P6 moves the planned dates of unstarted work to the early dates when it schedules, but writing them would change the plan on re-import. Activity code types and values with blank names, which P6 allows, export exactly.
Your edits always win over retained metadata. The export includes edits to activity codes and their dictionaries, user-defined fields, steps, notebook topics, feedback and notes to resources, resource assignments, duration type and % complete type. See Activity details and P6 fields.
Stored period performance is written into both formats up to the version’s data date, and wins over any financial-period data retained from the source file.
The original uploaded file is still available byte for byte from the import review page. Use that when you need exactly what was received, and an export when you need the plan as it stands now.
Other P6 exchange
- Spreadsheets. Export the activity table to Excel with a layout’s columns, edit it, and import it back as a reviewed update, as with P6’s spreadsheet import. See Round-trip the activity table through Excel.
- Layouts. Layouts export and import as os.construction
.jsonfiles. P6.plflayout files import on a best-effort basis: columns, filters, Group & Sort, bars with their exact colours, relationship lines, sight lines and the timescale are mapped where there’s an equivalent, and an import summary lists what was mapped, approximated and left out..plfexport isn’t offered. See Layout files. - Store Period Performance. Financial periods and period performance are managed in os.construction. P6 financial periods and past-period actuals in an imported file are kept and written back on export, but aren’t editable. Re-storing a period appends a superseding run instead of overwriting it. See Period performance and earned value.
- Baselines. A P6 XML file’s baseline projects and its project, primary, secondary and tertiary baseline assignments are kept and written back to P6 XML. Importing doesn’t create os.construction baselines, which are versions you save. XER has no baseline object.
What round-trips
Field by field, the product’s parity matrix now covers 139 XER and P6 XML rows: 94 round-trip unchanged, 36 round-trip with a stated caveat and 9 are dropped. On the full-featured XER test file, every non-blank source column survives import and export unchanged, apart from the values that are recalculated on purpose (dates, float and status). A genuine P6 Professional 20.12 XML export is now part of the tests too, and its expenses, baselines, feedback fields and cost dictionaries round-trip.
| Carries both ways | Carries with a caveat | Doesn’t carry |
|---|---|---|
| Activities, WBS, relationships and lags, calendars with exceptions and part-day shifts, constraints, progress, scheduling options | Project defaults, activity priority, assignment rate type, resource details and WBS or resource UDFs: kept and re-exported, but not editable in the app | Activity discussion, WPs & Docs links and weekly work plans, which are os.construction records |
| Activity codes (global and project, hierarchical) and assignments | Code-list UDFs carry their codes in P6 XML, but XER has no documented code type, so they export to XER as text | Role-only assignments and assignment UDFs |
| UDF definitions and values on activities, including Indicator UDFs (red, yellow, green, blue) | A line break inside an XER notebook entry becomes a space (P6 XML keeps it) | Suspend and resume in P6 XML |
| Steps, in XER and P6 XML; notebook topics; resource assignments with units, rates, costs, remaining units per time, assignment lag, the Drive Activity Dates flag and cost account | Suspend and resume dates carry in XER and the resume date is scheduled, but the suspension isn’t taken out of elapsed start-to-start lag | The risk register, correlation groups and risk templates |
| Duration type, % complete type and units % complete | Financial periods and past-period actuals: kept and re-exported, not editable. Stored os.construction periods win on export | WBS, resource and project UDFs in P6 XML |
Activity expenses (XER PROJCOST, P6 XML ActivityExpense), the cost account tree, expense categories and resource curves |
P6 XML baseline projects and baseline slots: kept and written back to P6 XML; no os.construction baselines are created | |
Notes to Resources (XER TASKNOTE) and Feedback (XER TASKFDBK) with the New flag, as separate fields |
Cost accounts and expense categories named only in the plan are written as new ones on export | |
| Earned value settings per WBS (technique, ETC technique, custom percent and performance factor) | Time precision, as a project UDF, when it’s set to Days or Minutes | |
| Three-point risk estimates, as three numeric activity UDFs |
Earlier releases read TASKNOTE as activity feedback. It’s P6’s Notes to Resources. Plans imported from XER before this release show that text under Notes to Resources the next time they open, and the next save stores it there.
MSPDI export goes through the same engine, but P6-specific metadata isn’t mapped into it. Treat it as a schedule-logic export, not a P6 archive.
How we verify
The figures on this page come from four kinds of evidence, each with stated limits.
- P6’s own answers. A corpus of 39 real P6 projects whose licences allow testing is imported, recalculated and compared with the dates and float P6 stored, value by value (26,337 values). Agreement per project is pinned, so any change, including an improvement, has to be reviewed and re-pinned. A separate P6 23.12 cross-validation file (13 single-purpose projects) checks 160 fields, 156 of which agree.
- Probes calculated in P6. Thirteen small probe projects, published under the MIT licence, were built and calculated in P6 Professional 23.12.1 to isolate behaviour no corpus file exercises, such as open-ended activities, resource-dependent work, suspended work and expected finishes. 1,962 of 1,968 probe values agree. No corpus project is resource dependent or sets the open-ends option, so the probes measure those changes. The corpus confirms three of the newer rules on its own: P6 stores zero free float on eight activities that carry an expected finish, drops the late side of a Start On constraint on work under way, and passes dates back through completed work under retained logic. Each change was compared cell by cell with the previous engine, and no value that agreed with P6 stopped agreeing.
- Unit and contract tests. Engine and interchange tests run hand-worked P6 answers, round trips and P6 research cases (leveling options, units progress, program options, forensic methods, weather risk). Route and contract tests check the Schedule REST endpoints’ requests, refusals and responses, many of them against the published API contract.
- A real-environment run. On 2026-11-10 the full app ran locally against an in-process Postgres with every migration and the seed data applied. The schedule and reporting suites passed 2,979 of 2,979 tests in 209 files. Every failing browser test was then triaged as either a product bug or a test script that hadn’t kept up with the interface. Five product bugs were found and fixed (see the changelog), and the scripts were brought up to date without weakening any check. On the last full run, browser tests passed 51 of 52. The one failure was a timing step, a save still in progress after five seconds on the shared test server, and it passed every time when repeated alone. That database is Postgres 17 compiled to WebAssembly with a single session, so concurrency and locking tests prove nothing there; it complements continuous integration rather than replacing it.
Security and accessibility are verified separately. Security and accessibility summarises the Schedule security review, the tests that keep every route and action authorised, and the WCAG 2.2 AA audit with its known exceptions.
What isn’t verified yet: P6’s ranking rule for multiple float paths isn’t published, so float-path numbering is an approximation, and the remaining 6 probe values and 3,254 corpus values that differ from P6 are listed as open, not explained away.
Limits
- Two activity code types with the same name collide, because codes are keyed by name.
- Resource-dependent work: constraints, As Late As Possible and expected finishes on resource-dependent activities are approximated with the activity calendar and its longest assignment, and an assignment with no remaining work holds no window of its own.
- Suspend and resume: the suspension isn’t taken out of elapsed start-to-start lag, the resume date is read to the day (work restarts at the calendar’s first working time that day), and P6 XML doesn’t carry suspend and resume dates.
- The largest files can lose their retained metadata to the size budget. Their export then keeps only the fields the plan models. See Large schedules.
- Multiple projects in one XER can be read by the engine, but the field snapshot refuses them. Export one project per file.
- P6
.plflayout files import on a best-effort basis only. Bars whose filter has no equivalent here, such as custom criteria, are listed and left out (a Level of Effort bar filter becomes the bar’s activity type), and.plfexport isn’t offered. - Financial periods and P6 baselines from a file are kept, but you can’t edit them in the app.
- Import doesn’t re-rate assignments. On import, P6 fits each assignment’s remaining units to the activity’s window and recalculates units per time, and it re-sizes resource-dependent durations. The importer keeps the file’s values, so a file P6 has never re-imported can schedule slightly differently.
- Float path numbering agrees with P6 on 51 of 168 stored path numbers, because P6 doesn’t publish its ranking rule.
Mistake: Comparing a new plan with P6 without checking settings
In the app: Open Settings first. The total float method and progress mode are the usual differences.
Mistake: Sending an MSPDI file to someone on P6
In the app: Send XER or P6 XML. MSPDI carries the logic, not the P6 metadata.
Mistake: Expecting baselines in the XER
In the app: XER has no baseline object. Send P6 XML, or name the baseline in os.construction and export that version separately.
Mistake: Using a code-list UDF as a filter key after an XER round trip
In the app: Code-list UDFs come back from XER as text. Use activity codes, or exchange P6 XML.
Related
Keep exploring
Related across os.construction
- DocsViews and layouts: filters, Group & Sort, columns, Network and Trace LogicOpen and save P6-style layouts for yourself, the project or the company, export and import them as files, build filters with nested All/Any criteria and highlight mode, group and sort to five levels with totals, choose from 45 core columns plus baseline, cost, units and earned value columns, draw a progress line, read the logic in the Activity Network, Trace Logic and Assignments views, and roll the plan up in the WBS window.
- DocsScheduling: CPM planning, analysis and evidence in one workspaceA native critical path scheduling workspace with a P6-validated engine, P6-style layouts and views, Line of Balance and takt planning, Last Planner weekly work plans, activity discussion and a change feed, P6 import options, DCMA health, three-point risk, a reviewed update cycle, portfolio programs, reports, baselines, versions, resources, costs and period performance.
- GlossaryData DateThe status cutoff separating recorded progress from the remaining work forecast in a schedule update.
- GlossarySchedule CalendarThe working-time rules a scheduling calculation uses to place activity durations on actual dates.
- BlogP6 scheduling settings: the assumptions behind the finish dateReview retained logic, progress override, criticality and lag calendars before comparing Primavera P6 schedules or explaining forecast changes.
- ExplainerHow the critical path worksForward pass, backward pass and float on an eight-activity network, then one slip moves the critical path.
Was this helpful?
Run it on your own project.
Start a 14-day free trial. Want a say in what ships next? Join the founding program.
14-day free trial. Pricing for paid plans isn't set yet. Talk to us. Already have an account? Sign in →