Skip to content

Scheduler Conventions

This page is a reference for evaluators and library users who need to know whether trueppm-scheduler models their schedule the way they expect. Each row is one rule the engine implements today, with the issue or ADR that decided it. Rows marked Differs are the ones where a schedule built in MS Project or Primavera P6 can come out differently here. Everything else follows the conventions those tools use.

The same list ships in the trueppm-scheduler README for PyPI readers. The narrative for each rule lives on CPM Scheduler and Monte Carlo.

Rulevs MS Project / P6Decided in
Durations and three-point estimates count working days, in whole days only. A sub-day duration or lag is rejected with InvalidScheduleInput.Differs — both tools schedule in hours.#826
Calendar.hours_per_day and Calendar.timezone are inert. They round-trip through serialization and change no computed date.Differs — in MS Project, hours per day converts a day-denominated duration into working time.#4131
Lag counts calendar days. The resulting date always snaps forward to the successor’s next working day — a lead (negative lag) follows the same rule and never snaps backward — so a weekend can absorb a short lag or lead.Differs — MS Project counts lag in working time by default and treats elapsed days (2ed) as the opt-in. P6 counts lag on a working calendar.#2534; open question #2535
With per-task calendars, a task’s duration expands on its own calendar and a link’s lag is consumed on the successor’s calendar.Same as MS Project’s task calendars.ADR-0120
Rulevs MS Project / P6Decided in
All four link types (FS, SS, FF, SF) take a lead or lag.Same.—
A zero-lag FF and a zero-lag SS on a whole-day Standard calendar reproduce the Start/Finish MS Project itself computed and saved for the same tasks.Confirmed against a real MS Project export (LinkTypes.mpp), not just the documented rule.#4323
A task driven by an FF or SF link stays contiguous and is right-aligned on its pinned finish, so its start moves back. As a result, CPM is not monotone in duration on a network with FF/SF links: a longer task can start earlier and pull an SS successor with it.Same as MS Project.#3806 (decided: keep)
An SF link finishes the successor at the start of the predecessor’s start day. With zero lag, the successor’s last working day is the day before. A positive lag counts from the predecessor’s start itself, so a weekend just before that start cannot absorb it: a predecessor starting Monday with SF+2d finishes its successor on Tuesday. A lead counts back from the zero-lag finish, so a shorter lag never finishes the successor later. The rule is the same for a task predecessor and a milestone predecessor.Same — Verified for the zero-lag, ordinary-work case against a real MS-Project-saved schedule (link_types_2013.py, whose SF predecessor starts on a Tuesday), and now for a positive lag against four independently-corroborated MS-Project-saved alignments (task_links_cross_version.py, #4327) — including the Monday-start, weekend-adjacent-anchor case #4333 reports: all four agree with MS Project, confirming the #4333 fix against external file evidence rather than only the hand-built regression test in test_sf_from_work.py. A second, separate divergence remains (#2534): the engine counts lag in calendar days and snaps forward once, so a lag long enough to reach a second weekend from the anchor again finishes one working day short of MS Project — confirmed on all four alignments by raising their lag past what the saved files actually used (e.g. the Monday-start case agrees through lag=6 and diverges again at lag=7; a mid-week start diverges as early as lag=3). An SF link out of a milestone remains unverified — no MS Project sample found covers it (#4327).#4145, #4322, #4333, #2534
A task with an SF link may be placed before the project start: a predecessor on the project’s first working day puts a zero-lag SF successor on the working day before it. The task’s project-start floor is the earlier of the project start and the place its SF links put it, and since those links already hold it there, the project start never holds back a task with an SF link. Its other links still apply, so a binding FS, SS or FF lead can also place it before the project start. A task with no SF link keeps the project-start floor, so its leads stay floored. A data date or planned_start after the project start, and a recorded actual_start, hold every task. A data date on or before the project start holds nothing, SF-linked tasks included (#4635): it is the project-start floor, which those tasks do not take. A planned_start on an SF-linked task holds it wherever it falls, even before the project start, because it can only hold the task later than its links place it. The Schedule’s “snap to project start”, “move project start” and “start at the earliest” options write exactly such a planned_start. A shorter lag never finishes work later. On a network of tasks with work, adding an FS, SS or FF link that does not bind moves nothing. Adding an SF link can move its successor earlier, because it lifts that task’s floor. A zero-duration milestone held only by the project start is an exception: its successors measure from its unfloored instant (see the milestone row), so adding a link into it can move them.Same as MS Project for a task whose links are all SF, which it schedules ahead of the project start. Unverified for a task with mixed links and for leads (#4319): whether MS Project or P6 floors either at the project start has not been checked.#4218, #4220, #4319, #4635
A zero-duration milestone is an instant. A milestone driven by work sits at the end of its driver’s finish day, and a milestone held by a floor sits at the start of that day. When a lag places the milestone after a weekend or holiday, it is shown at the start of the next working day. A predecessor’s recorded finish on a non-working day does not: it is read as the end of the last working day before it (see Progress and actuals), so the milestone is shown at the end of that day. The project start only bounds where a milestone is shown. When a lead or an SF-placed predecessor puts a milestone’s links before the project start, the milestone is shown on the project’s first working day — unless it has an SF link of its own, which shows it where its links place it — but its successors’ lags still count from the earlier point, so the project-start floor alone never makes inserting a milestone into a finish-to-start link move anything. A data date or planned_start after the project start, and a recorded actual_start, still hold the milestone and every lag measured from it, so a milestone held by one of those can move a successor later than the same link without it. A floor is the same point in working time as every midnight back to the close of the last working day before it (Saturday 00:00 for a Monday on a Monday-to-Friday week, and for a recorded Sunday actual_start, which is kept verbatim), so a milestone held by one links from the later of its link and that earliest midnight. Recording an actual_start or a planned_start on the day a milestone is shown therefore moves nothing (#4690), and slipping a predecessor never moves the milestone’s successors earlier (this holds for an SNET, the data date and a recorded actual_start; moving the project start later can still move them earlier, #5055). The cost: a successor on a wider calendar can sit on the weekend before a Monday-floored milestone, and a calendar-day lag out of it counts from that Saturday, so M -FS+2cd-> B with M held at Monday puts B on Monday, not Wednesday. A Sunday actual_start links from that Saturday too, so recording a milestone’s start on Monday instead of Sunday never moves its successors earlier, and a lag out of a started milestone can count from before its recorded actual_start, and the successors of a completed (signed-off) milestone can likewise sit on the non-working days before its recorded date (#4883). A data date on or before the project start is the project-start floor, so like that floor it never moves a successor. The same holds for a planned_start on or before the start, unless the milestone has an SF link of its own, where it holds the milestone as on any SF-linked task. A project that starts in the future, whose default data date (today) is before the start, therefore keeps every lag out of its kickoff milestones (#4635). A milestone held at the project start keeps zero total float and reads as critical, even when the tasks on both sides of it have slack. A milestone’s start and finish are one instant, so a finish-to-finish link into a milestone places it exactly where the same finish-to-start link would, and a lagged start-to-finish link places it on its anchor plus the lag (a positive lag counts from the predecessor’s start itself). Neither is snapped as if the milestone had a finish day to land on. A finish-to-finish link out of a milestone measures a positive lag from the milestone’s own instant, the point a finish-to-start link measures from. With no lag the successor only has to finish by the end of the working day before the milestone, and a lead counts back from there, so a longer lead never finishes the successor later than a shorter one.Same for the instant and the weekend rule: MS Project moves an elapsed lag that ends on non-working time to the next working time — confirmed for the work-driven case against a real MS Project export with all 10 zero-duration milestones in a saved .mpp chain (#4323). Unverified for the project-start rule (#4225): passing the earlier point on to a floored milestone’s successors has not been checked against MS Project or P6.#4079, #4173, #4225, #4272, #4273, #4323, #4635
The only date constraint is start-no-earlier-than, via Task.planned_start, snapped to the next working day. There is no finish constraint, must-start-on, or as-late-as-possible. From 0.5, Task.deadline is a soft deadline that the engine does not read: it flags a late finish and never moves a date.Differs — MS Project has eight constraint types plus deadlines. P6 has primary and secondary constraints.#3345, #804
An SS or SF link from a summary task is rejected with InvalidScheduleInput. FS and FF links from a summary are expanded to its leaves. This rule belongs to the summary expansion step (expand_summary_dependencies), not to the CPM pass: the CPM pass has no summary concept and schedules every task it is given as a leaf. TruePPM runs the expansion before every schedule; a library user of trueppm-scheduler must call it first, or a phase row is scheduled as an ordinary task, with no error. Creating such a link is refused, and so is the indirect route: a structural edit that would make a task driving such a link a summary (indent, outdent, reparent, create under it) is refused, and an import (MS Project, CSV/Excel, template) skips the link and lists it in its warnings. Ships in 0.5; in 0.4 those paths stored the link and the project then failed every recalculation.Differs — MS Project accepts any link type on a summary.ADR-0370, #4830
Rulevs MS Project / P6Decided in
A completed task (actual_finish set, or percent_complete at 100) is pinned to its recorded dates verbatim, even when they fall on a non-working day. It is out of network logic and is never re-sampled in Monte Carlo. With no recorded actuals at all it is pinned too: its full duration is laid out from the project start, or from its planned_start (start no earlier than) when that is later, and no predecessor moves it. Its predecessors’ float is therefore measured without it, and it still drives its own successors. It never finishes after the data date, though: when that span would run past it, the task is laid back to finish on the data date (the working day before it, when the data date is not a working day), keeping its full duration even if that starts it before the project start. Recording no actuals therefore never slips the schedule further than recording an actual_finish on the data date would. A data date at or before the project start changes nothing. Ships in 0.5; in 0.4 such a task, and its successors, can be pushed past the data date (#4539). A completed milestone stays a milestone. With no actuals it is an instant at the start of its pinned day. With actuals it sits on its actual_finish day (else its actual_start day), early_start and scheduled_start included, even when an earlier actual_start is recorded (#4458), and keeps the edge of that day it had while live. A milestone that a link from completed work places at the end of that day stays there, so work after it starts the next working day. A link that places it on a later day — work recorded after the sign-off, out of sequence — does not count, so closing that work moves nothing (#4703). The link decides, not the finish date: an FS+1d link from work finishing on the milestone’s day places it on the next day and does not count, while an FS-1d link from work finishing the day after places it at the end of the milestone’s day and does, so closing that work moves the successors a working day later. The edge is decided by completed work, or, while that work is still open or completed work places the milestone after its day, by the edge the last schedule recorded for it on that day (the TruePPM API hands it back; library callers pass live_day_end_on). Once every predecessor is complete, the completed work decides again, so correcting a predecessor’s actual dates or removing a link is reflected. That also means a handed-back end-of-day edge is re-read: successors can still move a working day earlier when a closed late predecessor’s link places the milestone at or before its day, and several late predecessors with negative lag can still move successors a working day later (#4915). A milestone with no predecessor, or one the data date or its planned_start held, stays at the start of the day, so work after it starts that same day. Nothing about open work’s current forecast is read, so no change to open work ever moves a completed task, and closing a milestone on the day it was forecast keeps the edge of the day its successors measure from. Once every predecessor is complete, that covers the lags out of it too: a milestone shown at the start of a day may sit at an earlier midnight, such as Sunday 00:00 shown on Monday, and once closed it keeps linking from that midnight, read from its completed predecessors. Closing a milestone while one of its predecessors is still open can still move its lagged successors, as before. The open work’s forecast is not read, so the milestone links from the start of its recorded day (#4740).—ADR-0136, #4343, #4433, #4525, #4458, #4690
A link out of work whose recorded actual_finish or actual_start falls on a non-working day of that work’s calendar measures from the same point in working time as the nearest working day. A non-working actual_finish is read as the end of the last working day before it, and a non-working actual_start as the start of the next working day. This holds for all four link types at every lag, for in-progress and completed work, on the project calendar and on a per-task calendar. On a Monday-to-Friday week, recording completed work as finished on Saturday or Sunday therefore places every successor exactly where a Friday actual_finish does, and a weekend actual_start places start-to-start and start-to-finish successors where a Monday one does. The task’s own dates are still shown verbatim. Ships in 0.5. In 0.4 a link counted its lag from the recorded weekend day itself, so a finish-to-finish successor of a Saturday finish landed on Monday instead of Friday, and a lagged finish-to-start, start-to-start or start-to-finish successor could land a working day earlier or later than its weekday twin.Unverified — MS Project’s absolute-time model may place a weekend finish’s finish-to-finish successor on Monday.#4631, #4702, #4332
An in-progress task schedules only its remaining work, duration − floor(duration × percent_complete / 100) working days. That work runs forward from the data date (status_date) and is floored at actual_start, which is kept unsnapped, and at the task’s planned_start (start no earlier than), which acts as “resume no earlier than”: a planned_start after the data date holds the remaining work there, while scheduled_start stays on actual_start. Dragging an in-progress bar on the Schedule writes exactly that planned_start.—ADR-0132
A start-to-start or start-to-finish link out of an in-progress task measures from when the task started, not from where its remaining work starts. With a recorded actual_start, that is the actual_start, at any percent_complete, including 0% just after the task moves to in progress. That can differ from where the task’s bar is drawn, which for a task with no progress yet still starts where its remaining work does. A status update that moves only the data date leaves the successor where it was until the data date itself reaches it, since the data date still holds every unfinished task. Because that start is history, the link sets no late date on the task and does not limit its free float. With progress entered as a percentage only, and no actual_start, it is the task’s scheduled_start (the full duration counted back from its forecast finish, where its bar starts): the successor follows the task’s bar instead of sitting a full lag past the data date, and because that start moves when the remaining work slips, the link still bounds the task’s late dates and free float. Either way the link is reported as the successor’s driving link when it holds the successor. Finish-to-start and finish-to-finish links measure from the forecast finish in every case.Same as MS Project and P6, which measure a start-to-start successor from the predecessor’s actual start once it has one. MS Project records an actual start as soon as progress is entered, so it has no percentage-only case to compare.ADR-0132 §2a, #4523
Rulevs MS Project / P6Decided in
total_float is the working-day span from the early start to the late start, and is_critical means total_float == 0 on a task that is not complete (a completed task has zero float too, but is never critical). This holds for milestones too, because a milestone’s late start is shown the way its early start is. A milestone at the start of a day whose float runs to a non-working midnight (a weekend or holiday, such as Saturday 00:00, reached through a calendar-day lag, the data date or a non-working actual_start) shows its late start on the next working day after it (the Monday), not the working day before (the Friday). Saturday 00:00 and Monday 00:00 are the same point in working time, and pinning the milestone to that Monday moves nothing. That late start is the latest the milestone can start in working time when the milestone itself is held there. When its predecessor is what places it on that Monday, the project finish can still move: in P -SS-> M -FS+3cd-> S with a finish of Tuesday 03-16, holding M on Monday 03-15 with planned_start leaves the finish where it is, but holding P on Monday 03-15, so that the start-to-start link puts M at Monday 00:00, moves S to Thursday 03-18 and the finish by two working days, because the calendar-day lag now runs from Monday rather than Saturday. Whether lag counts calendar or working days is a separate open decision (#2535) that this rule does not settle. Ships in 0.5. In 0.4 such a milestone shows its late start on the Friday, one working day short of its total_float.—#4528
critical_path lists every critical task in topological order. It is not a single chain, so its last entry is not necessarily the task that sets project_finish: a critical task can have successors that all carry float. Two rules produce that. A milestone held at the project start reads as critical while its successors keep slack. A calendar-day lag across a weekend lets a successor slip a working day that the task before it cannot. Among critical tasks, the latest early_finish is project_finish when any live task decides the finish; critical_path is empty when completed work holds it, including in a project where every task is complete.—#4174, #4183, #4281
Float stops where a slip would move the shown finish. When the project ends at the end of a day (work finishing Friday), a milestone placed at the start of a day, by a start-to-start link from work or by a floor, can slip only to the start of that Friday: one more day puts it at the start of Monday and moves project_finish. A finish-to-start predecessor places the same milestone at the end of a day, so it keeps its float up to the finish. That holds when project_finish is a non-working day too (a completed task finishing on a Sunday): the milestone can slip to the end of the Friday before it. When a milestone at the start of a day sets the finish, anything may end as late as the start of the day that milestone is shown on, whatever calendar it works on: beside a milestone shown at the start of Monday, a task on a seven-day week can finish on the Sunday before it without moving project_finish, and its float runs to that Sunday. A milestone linked to another with no lag puts it on the same midnight and passes on its edge of the day, and each is shown on a day its own calendar works, so float stops where any milestone in that chain would be shown past the finish: a milestone at the start of Friday can be followed by one on a Monday-to-Thursday week, which is then shown on Monday. A start-to-finish link into a milestone places it the way a finish-to-start link does, and its predecessor’s float stops at the same point. None of this stops float short of a started milestone’s recorded actual_start, even on a non-working day: that start holds the milestone on its own day whatever edge a link would give it, so a predecessor can slip up to it without moving anything.—#4174, #4183, #4344, #4431, #4436, #4526, #4677
free_float inverts the forward constraint across all four link types, anchored on each successor’s early date, and is capped at total float. A task with no live successor falls back to its total float, and completed successors are skipped. A slip that leaves a milestone on the same midnight but changes the edge of the day it is shown on still moves it: when a milestone sits at the end of Wednesday and a start-to-start link would place it at the start of Thursday, the same instant, free float through that link stops a day short. A successor whose recorded actual_start falls on a non-working day is anchored on the next working day, the same point in working time, so a link reaching only the working day before it neither reads zero free float nor drives it. The same anchor bounds how far a milestone between them may move, so the milestone’s predecessors keep the same free float.Same definition.#1828, #4183, #4332, #4345
The order of ScheduleResult.tasks is unspecified. Look tasks up by id, never by position.—#1862
ScheduleResult.project_finish and a milestone’s early_finish are the day the finish is shown on, not an instant, and milestone_at_day_end says which edge of that day it is. The end of a Friday and the start of the following Monday are the same point in working time, so the shown day can move across a weekend or holiday while the working-time finish does not. To measure a slip, compare in working time and read a start-of-day milestone finish as the end of the working day before it. Never subtract two shown days. TruePPM measures a finish shift this way in the project end-date shift notification, the activity feed, the milestone forecast digest and sprint bridge card, the Monte Carlo what-if, the cross-project sprint-boundary check, baseline drift (the activity feed, the overview’s “Slipped vs baseline” list, a task’s Baseline tab deltas and a closed sprint’s milestone slip), the milestone rollup variance, the program rollup’s baseline and schedule variance, and every Monte Carlo delta — each percentile against the CPM finish (delta_vs_cpm), added time (risk_premium_days), run to run in the forecast history, and the what-if. A baseline or Monte Carlo run recorded before TruePPM kept which edge of the day each finish sat on is read as the end of its day. A task’s own schedule_variance_days (actual finish vs baseline) and baseline_finish_variance_days (forecast finish vs baseline — the baseline chip on a board card and in the task drawer) are measured the same way (#4203). Monte Carlo percentiles carry the same reading (#4204).—#4178
Rulevs MS Project / P6Decided in
A three-point estimate samples a Beta distribution, fitted by method of moments to the classic PERT mean (O + 4M + P) / 6 and σ (P − O) / 6. This is not the λ=4 Beta-PERT. Both have the same mean, but on a symmetric estimate the band here is about 12% narrower, so P80 and P95 land slightly earlier. On an estimate whose most-likely value sits at one end, the band is wider.Differs from the λ=4 Beta-PERT that @RISK and Primavera Risk Analysis default to. Neither MS Project nor P6 samples on its own.ADR-1245, #4133
Every sampled duration is floored at the task’s own duration, so the optimistic tail below the plan is not expressed. To model finishing early, lower duration.Differs — risk tools sample the whole estimate.#3765
A zero-duration task with a sampled duration is decided run by run. A draw under half a day rounds away: it is the instant the CPM pass places, links and lags included, and hands no fraction on to the work after it, so an O=0 / M=0 / P=1 estimate leaves P50 on the CPM finish. Three 0.2-day draws in a chain therefore cost nothing, although they add up to 0.6 of a day of work. A larger draw is simulated as work: with no recorded start, the same as a task of that length with the same link. On a network with FF or SF links a work-sized draw can back-solve a start, so a run can finish before CPM. Ships in 0.5; in 0.4 every run charges such a task at least one working day.Unverified — no MS Project or P6 sample covers it.#4698, #4822
Sampled durations are continuous, and a run’s finish is shown on the day its work ends in: work ending 0.3 of the way into a Monday finishes on Monday. A link carries the fraction, unless its lag lands the successor on a day it does not work, in which case the successor starts when its next working day begins. A milestone after such work sits at that point in working time, shown at the end of its day. A run with no spread reproduces the CPM finish exactly. An interpolated percentile is read the same way, so one that falls between two whole-day positions is shown on the later day. A zero-duration task with an estimate is exempt: its draws under half a day round away (row above). Ships in 0.5; in 0.4 the finish is read as the nearest day, and a lagged link can lose another day. Upgrading moves seeded percentiles. On networks built from FS and SS links no percentile lands before the CPM finish. Most percentiles move one working day later, and lagged links can move them up to four working days later, because the nearest-day reading could lose a day at each lagged link. In a few lagged cases a percentile moves one working day earlier, because that reading could also add a day at a lagged link. On networks with FF or SF links a percentile can also move earlier.Unverified against risk tools, which schedule in hours.#4697, ADR-1245
No percentile precedes the CPM finish on a network built only from FS and SS links. On an FF/SF network a percentile can land earlier, because CPM itself is non-monotone there (see the FF/SF row above).—#3806
P50/P80/P95 are ranked by each run’s working-time finish. Where runs at the same position are shown on different days (the end of a Friday, the start of the Monday after it), each is shown on the day its rank lands on with runs ordered by position and then shown day, so it stays a quantile of the distribution in both directions, with p50_at_day_start / p80_at_day_start / p95_at_day_start saying which edge of that day each one is. A run whose finish is a start-of-day milestone ranks level with a run ending the Friday before it. Compare a percentile with the CPM finish, or with another run’s, in working time. When tasks use different calendars, a percentile that falls between two runs can be shown on a day none of the tasks works (the end of a Friday when every task works Monday to Thursday). It still lies between those two runs. A completed milestone keeps the edge of its day here too, so a project that ends on a closed milestone with no predecessor, one placed after work that finished on an earlier day, or one with an open predecessor, unless it was last shown at the end of its day, reads its percentiles as the start of that day, matching the CPM finish. Ships in 0.5; in 0.4 such a finish is read as the end of its day, one working day after the CPM finish. Upgrading can also change the duration-sensitivity index of tasks in a project that has such a milestone, because runs are ranked by where the milestone actually sits in working time.—#4204, #4601
A fixed seed reproduces P50/P80/P95 exactly on the same numpy version and the same trueppm-scheduler version. Upgrading either one can move seeded percentiles.—#4099