Publications · 2026

Is the Optimiser Needed? Auditing a Digital Decision Support System for Necessity, Decision Stability and Input Fragility Under Turbulence

Vladimir Stantchev, Peyman Rabiei and Daniel Arias-Aranda

Systems, 14(12), 1509, 2026 · doi:10.3390/systems14121509 · Open access

Abstract

Optimisation-based decision support systems are adopted under turbulence on two untested assumptions: that the problem needs their stochastic optimiser, and that their committed decision is stable. We audit both, and input fragility, with a five-step protocol on our own published fuzzy relief-allocation system on synthetic instances. Its model is separable: a deterministic rule computes the exact optimum, which dominates 97–100% of the published front points; the check also shows that its prioritisation objective, as written, rewards directing the people of lowest priority. Runs with different seeds commit to allocations differing for 83–94% of the people directed, because the solver stops where the model is flat; an exact decider with a stated tie-breaking rule commits to one allocation. On that decider, regret from degraded data grows with turbulence; which inputs cost most depends on the objective, and correcting it cuts the cost of person-side inputs by 59–74%. A load-aware repair stays exactly solvable at the target sizes; beyond them, with hard capacity, a capacity-aware local search stays on average within 1.5% of the exact method on seconds of solver time, and the evolutionary system trails at every size. Two further models receive different verdicts. Economic turbulence is engaged by analogy of mechanism only.

Abstract reproduced from the published article under CC BY 4.0.

Read the full text at the publisher

Software and data

Cite

@article{stantchev2026optimiser,
  title = {{Is the Optimiser Needed? Auditing a Digital Decision Support System for Necessity, Decision Stability and Input Fragility Under Turbulence}},
  author = {Stantchev, Vladimir and Rabiei, Peyman and Arias-Aranda, Daniel},
  journal = {Systems},
  volume = {14},
  number = {12},
  pages = {1509},
  year = {2026},
  publisher = {MDPI},
  doi = {10.3390/systems14121509},
  url = {https://doi.org/10.3390/systems14121509}
}

Related work

All publications · Vladimir Stantchev