A grounded survey of open-source project, program and portfolio management software, the standards it claims to implement, and the one capability you asked for that no repository on earth currently ships.
The body behind the PMBOK® Guide publishes standards, certifications and research. Its own literature observes that practitioners "repeatedly look for alternative tools and ask for information on project management tools tagged as agile, open source, web-based, or cloud" — PMI documents the demand rather than filling it. There is no PMI reference implementation to fork.
OpenPPM is the only open-source product that explicitly positions itself as project, program and portfolio management "in accordance with the PMBOK® Guide, from PMI®" — web-based, multi-tenant, with Project Manager, Team Member, Resource Manager and PMO roles. Its role model is worth stealing. Its codebase is not actively maintained.
PMBOK® Guide 8th Edition, released 2025, collapses the 7th Edition's 12 principles into 6, its 8 performance domains into 7, and reintroduces 40 processes embedded in those domains. EIA-748-E — the earned-value standard — was published February 2026. If your handbook is the 7th Edition, the data model we build will be one generation behind on vocabulary.
Every tool surveyed tracks change requests after a human has decided one is needed. None classify an incoming task against a baselined WBS. The commercial market agrees this is the gap — a vendor in the space states flatly that generic project management tools "do not solve scope creep because they are not designed for it." This is your differentiator, and it must be designed from first principles.
Critical-path calculation, resource levelling and working-time calendars are solved problems with hostile licensing. The MIT build of DHTMLX Gantt is explicitly "intended for schedule visualization and manual editing" — auto-scheduling, critical path, resource planning, calendars and baselines are all PRO-only. TaskJuggler gives you the maths free (it resolves resource assignment conflicts automatically and emits optimum start/end times) but as a GPL command-line compiler over a text file, not a library. Pick your poison deliberately, early.
Ranked by usefulness as a starting point for a hybrid, financially-aware PPM platform with gated phases.
You asked me to decide the classification logic. Here is the recommendation and the reasoning: deterministic rules first, language model as a tiebreaker, human approval always. Never let a model make the ruling alone — the output of this component is contractual.
Does the task map to a WBS element in the approved baseline? Deterministic: WBS ID, deliverable reference, or acceptance-criteria lineage. A hit ends the process — it is in scope, log it and move on.
No WBS hit, but is it work required to deliver an existing baselined outcome? Then it is discovery, not creep — the deliverable was always owed, the estimate was wrong. This distinction is the one PMs most often get wrong, and it changes who pays.
Ambiguous cases go to a model that must cite the specific clause of the scope statement it is reasoning from, and return a confidence score. No citation, no ruling. The PM sees the quoted clause, not a verdict.
Creep opens a change request pre-loaded with the schedule delta, the cost delta at the correct rate, the margin impact, and — your specific ask — the trade: what comes out if this goes in. Nothing enters the schedule until the trade is signed.
"Without a documented change request and the application of a formal change control process to assess that change request, there is an increased risk of scope creep."
EIA-748 already demands exactly this discipline of a compliant system: incorporate authorised changes in a timely manner and record their effects in budgets and schedules; reconcile current budgets to prior budgets; control retroactive changes so previously reported cost, earned value or budget figures move only for error correction, routine accounting, directed changes, or baseline integrity; and prevent revisions to the program budget except for authorised changes. Your adjudicator is not an invention on top of the standard — it is the standard, made automatic.
I found no published accuracy benchmark for automated scope classification, because no one has published one. Treat the classifier as a triage instrument that raises the right question with the right evidence attached — not as an oracle. Every ruling must be overridable, and every override must be recorded as training signal and as audit trail.
You said the handbook and agile methodologies are clearly defined standards and anything that exists should be an offering. These are the three that carry real, encodable constraints.
Released 2025 and built on more than 48,000 data points. Six core principles, seven performance domains, and 40 processes reintroduced inside those domains — a return to concrete guidance while keeping the principle-based frame. It adds expanded coverage of AI, PMOs and procurement.
Encodable as: the phase-gate model. The 7th Edition's domains were purely outcome-focused with no prescribed processes; the 8th Edition's are function-oriented, each embedding processes with inputs, tools and outputs. That is a state machine.
Thirteen pages, deliberately unprescriptive. Its load-bearing addition is commitments: the Product Backlog commits to a Product Goal, the Sprint Backlog to a Sprint Goal, the Increment to the Definition of Done. Work cannot be part of an Increment unless it meets the Definition of Done.
Encodable as: required fields, not optional ones. A sprint without a Sprint Goal cannot open. An increment without a satisfied DoD checklist cannot close. Note the trap Sutherland describes — teams were criticised for delivering a point less or a point more, and velocity became a weapon. Do not surface velocity as a per-person metric.
32 guidelines in five groups: defining the work and the organisation doing it, planning and budgeting, work accounting, analysis and reporting, and revisions and data maintenance. Guideline 1 is the WBS; Guideline 32 is documenting performance-measurement-baseline changes. Latest revision published February 2026.
Encodable as: your EVM module's acceptance criteria. Scale it — the standard is explicitly meant to be applied in proportion to the size, complexity and type of work. For DoD work the thresholds are real: compliance required at $20M, formal validation at $50M.
Your stated frustration is not that Microsoft Project lacks features. It is that "you need a PhD to keep it error free and accurate." That is a diagnosis of an interaction model, not a feature gap — and it means forking a tool that shares MS Project's model reproduces the problem.
The recommendation is therefore a hybrid: take OpenProject's data model and portfolio hierarchy as the substrate, take Plane's cycle-and-module patterns for the adaptive lens, wrap a scheduling solver rather than writing one, and spend all original engineering effort on the two things nobody has built — the scope adjudicator with its trade-off ledger, and a reporting layer where the standard outputs are fixed and everything else is composed on request.
Gated phases enforce the consistency you said was missing. If the app will not let a PM open execution without a baselined WBS, a signed charter and a resource plan with rates attached, then the tool is the standard, and consistency stops being a training problem.
Consulted 1 August 2026. Figures are dated to their publication, not to today. Where sources disagree, both are shown.
Scope-creep prevalence figures vary widely by source and method — PMI's 52% sits against third-party aggregations claiming 70%, and industry-specific studies putting software and creative work at 60–70%. I have used PMI's figure because its method is documented; treat the higher numbers as unverified. I have also not read the source code of any repository above — GitHub is not yet connected, so all repository claims come from project documentation, which is marketing. Verify OpenProject's cost model and OpenPPM's schema against the code before committing architecture to either.