Interactive planning lab / Entirely mechanical
Bottleneck
A machine is unavailable. Which jobs should move? Explore the knock-on effects, rebuild the order, and inspect every decision.
Independent synthetic shop. Three stages, up to ten jobs, one planned outage. No AI calls. Your job data stays in this browser.
- 1. DisruptChoose the unavailable machine.
- 2. ReplanSearch for a better job order.
- 3. InspectFollow the work and compare delivery times.
The outage is applied. Replan to compare a different job order.
Lime blocks: selected job · Other blocks: work · White cursor: inspected time · Outlined red area: machine unavailable · Dashed line: 16:00 reference, not a shift limit.
Follow one job
Why did this move?
Compare job completion times
| Job | Due | No outage | With outage | Replanned |
|---|
Inspect or replace the sample jobs
All jobs follow Cutting → Machining → Inspection. Times are whole minutes. Due is minutes after 08:00. Load 2–10 jobs with unique short IDs; no upload is made. Reorder the CSV rows to try your own sequence before running the search.
Planning workspace / Go beyond one schedule
Find the weak spots. Compare your options.
Run 27 what-if cases: the same outage length at nine start times on each machine. Each cell is a separate scenario, not simultaneous failures or a prediction.
Uses the jobs and outage length currently loaded above.
Keep a decision trail.
Save up to six named scenarios and compare their results. Snapshots live in this tab until you reload. Export a file to keep them; imported files are validated before replacing the collection.
How to read the analysis
Lower late-job counts are better under this model. A zero does not establish that a real shop can meet its commitments. The search is bounded and is not a proof of optimality. Compare total late minutes and final finish as well: saving one deadline can delay another. Changing jobs or outage duration clears the stress test; saved snapshots remain unchanged. Open any cell to examine its exact scenario in the timeline.
The model / Know its boundaries
A useful plan must hold together.
What is enforced
One operation per machine at a time. Each job completes its previous stage first. Operations cannot overlap the planned outage or be interrupted. Every machine follows the same job order; all jobs are ready at 08:00.
What the search does
A deterministic beam search keeps 120 candidate orders at each depth. It favors fewer late jobs, then less total lateness, then an earlier final finish. The original order remains a candidate. Every result is checked for conflicts. This is a valid best-found plan, not a proven global optimum.
What it does not know
No setup times, staffing, material shortages, breaks, alternative machines, or actual machine connections. This is planning before work begins; it does not reschedule completed or in-progress work. Review these assumptions before using an export for any real decision.
Built with D3 scales and a browser Web Worker. Computation is isolated from the interface; there is no model inference, telemetry, or server processing of job data. Read the method and provenance.