Planning and Scheduling Business Processes under Control-Flow Uncertainty: Extended Version
Michel Kunkler, Stefanie Rinderle-Ma
cs.AI
Quick verdict
Formulates chance-constrained planning/scheduling of business processes under control-flow uncertainty; solid formalization, honest but small-scale evaluation.
Summary
The paper addresses scheduling business-process activities when the required activity sequence is uncertain (control-flow induced uncertainty from exclusive choices and loops), using probabilities replayed from event logs. The authors formalize the problem on stochastic process trees as a chance-constrained optimization: select branches and loop-iteration counts so that case feasibility meets a threshold while minimizing expected superfluous (planned-but-never-executed) activities. Two CP-SAT implementations are compared: a decomposed two-stage approach (planning, then makespan-minimizing scheduling that lets mutually exclusive activities share resources) and an integrated single formulation that additionally allows non-exclusive activities to overlap ('overbooking') with per-case selections, computing a lower bound on expected feasible cases via scenario enumeration. On a synthetic repair process plus BPIC-14I and BPIC-17W (up to N=50, 20-minute limit), the integrated approach yields shorter makespans on the configurations it solves (e.g., Repair Shop N=5: 624s vs 960s) but finds no feasible solution in most larger configurations; the decomposed approach finds feasible schedules nearly everywhere. Limitations: structurally small real-world processes, single runs without variance reporting, a weak most-probable-path baseline, and uniform branch selection across loop iterations.
Review complete
Assessment at a glance
4
risk flags identified
● high 0
● medium 2
● low 2
4 strengths · 5 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- Activity selection under control-flow uncertainty can be cast as a chance-constrained optimization on stochastic process trees that produces full a-priori schedules until case completion (Sect. 1-2).
- The decomposed two-stage approach finds feasible schedules in every tested configuration except BPIC-17W with N=50, i.e., it scales to 50 concurrent cases (Sect. 4.3).
- Where the integrated approach solves, it achieves shorter makespans than the decomposed approach, typically by overbooking low-probability non-exclusive activities, but it is intractable at larger scales (Sect. 3-4).
- The scenario-based expected feasible cases Efc (Def. 14) is a valid lower bound under overlapping non-exclusive activities, using plan-bounded scenario sets with tail-absorbing loop weights.
Strengths
Genuinely novel problem framing with full-horizon schedules
Unlike prior work that schedules only up to the next uncertain decision point (Havur & Cabanillas 2019) or allocates resources online (Kunkler et al. 2024; Middelhuis et al. 2025), this paper produces a-priori schedules until case completion with an explicit probabilistic feasibility guarantee (Sect. 1, Sect. 5).
Rigorous probability accounting for the overbooking semantics
Definitions 9-14 carefully handle overlapping non-exclusive activities: a static priority order, plan-bounded scenario sets with tail-absorbing loop weights, and an explicit argument (with a 0.5-probability counterexample in Sect. 3) that the expected feasible cases Efc is a valid lower bound rather than an approximation.
Both formulations implemented and benchmarked on solver infrastructure
Honest negative-result reporting
Weaknesses
Real-world testbeds are structurally minimal
Single runs, no dispersion for time-limited solves
Weak and structurally unfair baseline
Uniform selection across cases and iterations
Runtime semantics of overlaps are unspecified
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
This extended paper (arXiv:2609.05578v2, cs.AI) by Kunkler and Rinderle-Ma (TU Munich) addresses scheduling business-process activities when the exact activity sequence of a case is unknown at scheduling time — "control-flow induced uncertainty" from exclusive choices and loop iterations — using branch/loop probabilities replayed from event logs. The contribution is twofold: (i) a chance-constrained formalization on stochastic process trees that selects branches and loop-iteration counts to meet a case-feasibility threshold while minimizing expected superfluous (planned-but-never-executed) activities (Sect. 2.1, Eq. 6); and (ii) two CP-SAT-backed approaches — a decomposed two-stage pipeline (Sect. 2) and an integrated single model that additionally lets low-probability non-exclusive activities overlap on a resource ("overbooking", Fig. 1) with per-case selections (Sect. 3). Central claims: the integrated approach yields shorter makespans where it solves but is intractable at scale; the decomposed approach scales to 50 cases.
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 4 strengths & 5 weaknesses, analyzed in detail
- 4 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What is control-flow induced uncertainty in business processes?
It is the uncertainty about which sequence of activities a case will require, arising from runtime decisions such as exclusive choices and loop repetitions whose outcomes are unknown when the schedule is made. This paper handles it by planning with probabilities estimated from historical event logs, so the schedule covers enough execution paths to reach a target probability of successful case completion.
What is the difference between the decomposed and integrated approaches in this paper?
The decomposed approach first picks the activities to plan (minimizing expected superfluous activities subject to a feasibility threshold), then schedules them to minimize makespan, letting only mutually exclusive activities share a resource. The integrated approach solves one combined model that also lets low-probability non-exclusive activities overlap (like airline overbooking) and selects branches per case; it gives shorter makespans on small instances but fails to scale beyond them within a 20-minute limit.
Is code or data available for this paper?
Yes. A prototype implementation based on Google OR-Tools CP-SAT is linked in the paper (github.com/ltsstar/BusinessProcessPlanningScheduling/). The BPIC-14I and BPIC-17W event logs are public, and the synthetic repair process with its branching/loop probabilities is fully specified in the paper's figures.
What were the main experimental results?
On three processes with up to 50 concurrent cases and a 20-minute time limit, the integrated approach produced shorter makespans on the configurations it could solve (e.g., 624s vs 960s makespan on the synthetic repair process with 5 cases) but found no feasible solution in most larger configurations; the decomposed approach found feasible schedules in all configurations except BPIC-17W with 50 cases.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- PDF extraction flattened mathematical notation in Eqs. 1, 4 and 14 (superscripts and subscripts run together); the derivations were assessed from the surrounding prose and closed forms, and residual notation artifacts cannot be excluded.
- Table 2 is extracted as a dense flattened block; column-to-value alignment was reconstructed carefully, but minor misreadings of adjacent cells in individual rows cannot be fully ruled out.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.