Decision intelligence

Translate rules into verifiable decisions

A work instruction is not yet an executable schedule. Terms, deadlines and dependencies must become explicit before a proposal can be checked for feasibility and optimality within a model.

  • 4 min read
  • Updated on
  • smartlytics editorial team

The short answer

Rule-based optimisation separates hard constraints from an objective to improve. The demo transparently maps known text presets to constraints and checks every integer start-time combination for three tasks. It reports either a proven model optimum or infeasibility.

Open interactive exampleFree · in your browser · no sign-up

At a glance

  1. Hard rules determine feasibility; the objective compares the remaining schedules.
  2. Translating text into unambiguous constraints is a separate, reviewable task.
  3. Optimality applies to the specified model. Conflicting or overly restrictive rules may leave no feasible solution.

Why a well-written schedule is not enough

“A must finish before B, C cannot start until hour four, and everything must be completed before the shift ends.” This instruction contains several different conditions. A proposal can sound convincing while violating one of them. A verifiable decision therefore requires terms such as “before”, “start” and “finish” to have precise computational meanings.

Scheduling models commonly represent tasks through start times, durations, dependencies and limited resources. The OR-Tools job-shop example demonstrates this separation using precedence and machine constraints (OR-Tools: Job shop). Our demo uses a smaller case: three tasks on one machine. Every possible integer start-time combination can therefore be checked exhaustively without calling an external solver.

From known text presets to constraints

Task A takes two hours, B takes three and C takes one. Tasks cannot be interrupted. Start times are nonnegative integers, and every task must finish by the chosen deadline HH. For task jj with duration pjp_j,

sj≥0,ej=sj+pj≤H.s_j\geq0,\qquad e_j=s_j+p_j\leq H.

The machine can process only one task at a time. For each pair, either the first task must end before the second starts, or the reverse must hold. The “A before B” preset adds eA≤sBe_A\leq s_B. The third preset also adds sC≥4s_C\geq4.

This mapping is fixed and visible. The demo does not claim to understand arbitrary text. That is precisely its educational purpose: the translation can be checked independently of the eventual calculation. In a larger process, a language component might draft the interpretation, but domain meaning, units and exceptions would still need review.

A worked schedule

The first objective minimises the weighted sum of completion times. A has weight one, B three and C two. These weights specify how strongly each task’s later completion counts in the objective:

J=eA+3eB+2eC.J=e_A+3e_B+2e_C.

Without a precedence constraint and with deadline six, the optimal order is C, B, A. C finishes at one, B at four and A at six. This gives J=6+3⋅4+2⋅1=20J=6+3\cdot4+2\cdot1=20. All three tasks fit into the available time with no idle periods.

Adding “A before B” makes that schedule infeasible. The optimum becomes C, A, B: completion times of one, three and six give J=3+18+2=23J=3+18+2=23. The worse objective value is the cost of the additional rule within this model. It is neither a calculation error nor evidence that the rule itself is unjustified.

If makespan is minimised instead, several orders can be equally good. For equal objective values, the demo shows the first schedule checked. A different objective can therefore favour different schedules even when the hard constraints remain unchanged. Choosing the objective is a substantive process decision, not merely a display preference.

Change feasibility and the objective

Interactive

Check rules and optimise a schedule

The calculator checks every start-time combination and displays either an optimal feasible schedule or an infeasible model.

Choose “A before B” and set the deadline to six. Switch between weighted completion times and makespan. Then reduce the deadline to five: the three durations already sum to six, and the machine cannot process anything in parallel. No feasible schedule exists.

The output reports the numbers of combinations checked and feasible combinations. “Infeasible” here means the complete finite search found no valid combination. The display does not substitute an apparently best schedule that violates the rules. A change must instead address the task set, capacity, deadline or a constraint that can legitimately be negotiated.

What the proof covers

The search covers only integer start times, known durations and one machine. It ignores breakdowns, setup times, breaks and uncertain processing durations. An optimal model schedule is therefore only as useful as those assumptions. Units and boundary cases belong in model review before the result controls a real process.

For larger problems, solvers explicitly distinguish optimal, merely feasible, infeasible and unknown results (OR-Tools: CP-SAT). A time limit can prevent a proof from being completed. Our small exhaustive enumeration avoids confusing “nothing found yet” with “proved impossible”, but it does not scale to large scheduling spaces. The transparent calculation is an educational model rather than a production scheduling engine.

Connect language, rules and execution

Retrieval-augmented generation helps locate the source of a rule. Explicit modelling records how that rule was interpreted. Optimisation compares feasible alternatives, while later execution must still account for current conditions and responsibilities. This chain is a concrete application of Decision intelligence.

Frequently asked questions

Does the demo understand arbitrary work instructions?

No. It offers three known text presets with fixed, visible mappings to constraints. There is no free-language processing or hidden model call.

What does optimal mean in this example?

Every permitted integer start-time combination is checked. Among feasible schedules, the one with the smallest selected objective is displayed. This proof applies only to the small specified model.

Does failing to find a schedule always prove infeasibility?

No. It does here because the finite search is completed exhaustively. A time limit in a larger procedure may instead leave an unknown status.

Sources and further reading

  1. SourceGoogle OR-Tools team (2026): The Job Shop Problem. Google for Developers.
  2. SourceGoogle OR-Tools team (2026): CP-SAT Solver. Google for Developers.

Content last reviewed on 14 November 2026. Please let us know about any errors: impressum@nyxai.com.

From NyxAI

smartlytics is the learning hub of NyxAI GmbH.

Explore the methods we work with, explained openly, with sources and calculations you can reproduce.

Decision Core

NyxAI is developing a Decision Core for Decision Intelligence that combines optimisation, simulation, statistics and reinforcement learning: compare options, make uncertainty visible and learn from outcomes, with people in control.

DeepSquare

Explainable chess AI with human playing styles: NyxAI’s testing ground for understandable decisions under time pressure.