Start with something familiar
If a school changes tomorrow’s timetable, moving every class by a few minutes may be harder to manage than moving one class by a larger amount. Families and staff need to know both how many promises changed and by how much. Factory schedules behave similarly. A small gain on paper can require new kits, calls, and instructions. Stability makes those changes visible without pretending they should always outweigh urgent service needs.
Words to know: Stability, Frozen horizon, Tardiness, Dispatch.
Define the baseline and measure
Stability needs a reference schedule. Here it is the last communicated start for each existing job. Sum the absolute differences between new and old starts to measure total movement in minutes. Do not add signed changes: a job moved earlier and another moved later do not cancel the effort of changing both instructions.
Also count changed jobs. Two revisions can have the same total movement but affect different numbers of teams. A newly inserted job has no old start, so exclude it from movement and report it separately. These conventions are a declared local policy for the exercise, not universal definitions that every planning system must follow.
Treat a freeze as a governed rule
A frozen horizon can reduce routine churn, but its meaning must be precise. State whether it protects start times, sequence, quantity, or all three. Northstar protects starts strictly before the boundary. A job exactly at the boundary is outside this freeze. That boundary choice must be recorded because a vague phrase such as next two hours is hard to audit.
When a hard outage breaks the freeze, record the affected jobs and approve a new version through the stated owner. A policy should not require an impossible schedule. Equally, an attractive score alone should not authorize every change. Separate the physical feasibility test from the decision to disturb a commitment.
Compare options with visible tradeoffs
A combined score can add tardiness and a chosen multiple of movement, but that multiplier is a business judgment. If it doubles, the preferred option may change. Present the two raw measures beside the score so a reviewer can see whether a conclusion depends on that judgment. Do not hide a late customer behind a single index.
The lab asks for both feasible alternatives and a short release note. The purpose is to practice controlled revision: what changed, why it changed, and what evidence could cause another review. A stable schedule is not one that never changes. It is one whose changes have a clear reason and an understood cost to coordination.
- Validate inputs
- Place legal work
- Check every resource
- Compare outcomes
- Release with conditions
Northstar worked schedule
All data are fictional. Time is elapsed minutes from the stated shift origin; unlisted transfer times are zero. Use the explicit resource and setup assumptions below.
| Job | Old start | Duration | Due |
|---|---|---|---|
| A | 0 | 50 | 60 |
| B | 50 | 40 | 200 |
| C | 90 | 30 | 100 |
- 1. At time0, compare old A-B-C with A-C-B. All jobs are released0; no setups.
- 2. Old completion50,90,120 gives C tardiness20 and movement0.
- 3. Revised completion A50,C80,B120 gives no tardiness; B moves30 and C moves40.
A-C-B removes20 tardy minutes but changes two starts by70 total minutes. A frozen boundary60 protects B at50 and prevents this routine swap.
Assumptions
- All times and quantities are deterministic inputs unless explicitly labeled expected or scenario-dependent.
- Operations cannot be split; touching interval boundaries are allowed. Additional constraints are only those stated in the lesson.
Where this can fail: These small cases teach traceable decisions, not a promise of performance in an unmodeled plant.
Build, test and explain a plan
Use example data and one machine open0-200. Compute score total tardiness + alpha×absolute movement for alpha0.1,0.3,0.5. Compare A-B-C and A-C-B. Repeat with C due80. Apply freeze boundaries0,50,60 with starts strictly before each boundary protected. Assume no disruption grants an exception. Draft a release note only for an allowed change. The eight hours are self-paced estimates for modeling, scenario work, checks, and explanation; they have not been experimentally validated.
Use paper or a spreadsheet. The estimates below include the study and worked example already on this page.
| Estimated hours | Activity |
|---|---|
| 0.5 | Study the lesson and make a one-page input dictionary. Define the clock, units, resource names, release rule, and all assumptions. Mark which inputs are fixed and which will change in the scenarios. |
| 0.5 | Reproduce the worked example without looking at its result, then compare every intermediate start and finish. Explain any difference in a short error log before building the larger table. |
| 3 | Build a complete operation-by-operation planning table for every requested sequence or scenario in the lab brief. Include starts, finishes, blocked intervals, due comparisons, and a resource timeline. Cross-check each row against its input and reconcile resource totals. Use paper or a spreadsheet; no special software is required. |
| 2 | Run the specified stress cases independently, restore the base between cases, and compare each changed result with the original. Create a small sensitivity table and identify which assumption caused each difference. Audit feasibility before ranking the outcomes. |
| 1 | Write a decision memo with the selected feasible option, numeric evidence, the losing option’s advantage, and any unmet promise. State the condition that would cause you to reconsider. Do not invent additional capacity or missing data. |
| 1 | Teach the result aloud or in writing to an imagined operator. Rebuild one key check from a blank page, answer the recap questions, and revise the memo if your explanation reveals an unsupported assumption. |
Your deliverables
- Input dictionary and checked schedule table.
- Scenario comparison with job-level or quantity-level evidence.
- Decision memo and teach-back record.
Check your practice work
- At alpha0.1, old score20 and revised7, so revised wins when permitted.
- At alpha0.3, old20 and revised21; old wins. At alpha0.5, revised35.
- With C due80, old tardiness40 and revised0; revised score21 at alpha0.3, so revised wins if allowed.
- Freeze0 protects none; freeze50 protects A only; freeze60 protects A and B, so A-C-B is disallowed without an exception.
Keep these three ideas
- Use an explicit change measure and exception process to compare revisions.
- Which constraint or metric changes the choice in this case?
- What assumption must be confirmed before this plan can be released?
Common mistakes
- Comparing plans before checking every stated constraint.
- Changing several assumptions at once and losing the explanation for the result.
Optional video study
Use an explicit change measure and exception process to compare revisions.
- Introduce the specific Northstar inputs
- 1. At time0, compare old A-B-C with A-C-B. All jobs are released0; no setups.
- A-C-B removes20 tardy minutes but changes two starts by70 total minutes. A frozen boundary60 protects B at50 and prevents this routine swap.
Second half of module narration
Open the module media & quizPut this idea to use
Carry your assumptions into a practical tool, then compare your result with a short guide.
Open a related planning toolRead the related planning guidePractice with the planning gameThis saves a completion marker in this browser only. It does not verify time spent.