A Feasible Schedule Is Not a Reliability Work Acceptance

A documented Maximo Optimizer lab moves from infeasibility to a valid solution while leaving work unresolved. Reliability teams should treat solver status, excluded work and dashboard provenance as separate acceptance questions before trusting a maintenance plan.

Share
A Feasible Schedule Is Not a Reliability Work Acceptance

A Feasible Schedule Is Not a Reliability Work Acceptance

A maintenance schedule can be mathematically feasible and still fail to answer the reliability engineer's most important question: did the work that protects the plant actually make it into an executable plan? Solver status answers a narrower question about the model, its inputs and the solution found. It does not independently certify that the selected work captures every important failure risk or that the published plan matches the scenario someone reviewed.

A hands-on IBM Community lab by Mohamed Chihaoui makes that distinction visible. Working on Manage 9.2.4 with Optimizer 9.2.3 and demonstration data, the author documents an initial INFEASIBLE_NO_SOLUTION result, corrects schedule inputs, and obtains OK_SOLUTION while still reporting unresolved scheduling issues. The exercise is a practitioner account on an invented demonstration setting, not an IBM benchmark or a promise about another site's results. [1]

Its value for reliability teams lies in the sequence of evidence. Missing shift selection and work without labor plans explain the first failure. A later valid run still contains task-level problems. Further investigation exposes duration and skill-availability issues. The dashboards add another complication: they read different preparation steps and calculations rather than forming one automatically synchronized statement of maintenance performance. [1]

That is enough to build a more useful operational acceptance process without inventing a performance statistic. Separate the completeness of the work population, the validity of resource assumptions, the disposition of unresolved work and the provenance of management indicators. Then ask whether the selected scenario has actually been accepted and published through supported processes.

The sections below use the lab as an observed example and propose a reliability-oriented review method. They do not prescribe production database changes, weaken qualification requirements, or claim that every infeasible solve is caused by bad data. The purpose is to turn an optimizer result into an accountable maintenance decision.

1. Define the work population before celebrating a better solve

The lab's work query selects approved or waiting-to-schedule work for a demonstration site and adds an existence condition against planned labor. That condition is technically meaningful for resource leveling: a work order without a labor requirement gives the model no planned craft demand to level. The author identifies missing labor plans among the causes of the unsuccessful first attempt. [1]

There is also an important reliability consequence. Excluding records without labor plans can make an input population more suitable for optimization while leaving real maintenance obligations outside it. The exclusion may be entirely appropriate for this model, but it is not evidence that the excluded work is unimportant. A planner needs a disposition for that work, not merely a query that stops it disturbing the solve.

As a recommended control, maintain two views during schedule acceptance. One is the candidate work population under the organization's planning rules. The other is the actual model input after every query predicate and readiness filter has been applied. Reconcile the difference by record and reason. Work missing a labor plan should become a planning action or a documented exclusion, rather than disappearing from the reliability discussion.

Use the same discipline for date windows, status filters, site boundaries and parent-task selection. These are ordinary modeling choices, but they express an operational scope. A filter intended to avoid historical work could be correct; a filter that unintentionally removes an urgent corrective task could be consequential. The opened source does not establish a universally correct production query, so copy its logic only after review against local requirements and supported database behavior.

Preserve the unit of analysis. The demonstration discusses work orders and task activities, and its issue totals do not all refer to the same level of record. A manager hearing that several activities remain unresolved should not automatically interpret that as the same number of unplanned parent work orders. An acceptance record should identify whether each count concerns parent work, tasks, labor requirements or assignments.

The reliability owner can then ask a concrete question: which failure-prevention or corrective work is outside the executable plan, and who owns the next action? That question requires local criticality and maintenance-strategy context; the optimizer lab does not supply it. Keep the assessment explicit instead of treating solver success as a substitute for asset-risk review.

A useful output is an exclusion register tied to the planning snapshot. It need not be elaborate. It should show the work identifier, why it was not admitted to the model, the responsible owner and the agreed disposition. The point is not to force every record into one solve. It is to prevent input cleanup from being mistaken for completed maintenance planning.

2. Diagnose feasibility by layer, not by the word failure

The first unsuccessful run in the community example reports missing work periods, missing global capacities, missing requirements and an inability to schedule the work. The author's diagnosis connects those messages to an unselected shift and records without labor plans. The practical sequence is important: inspect model inputs before changing solver limits or escalating every failure as an infrastructure problem. [1]

It is equally important not to overlearn that example. An infeasible model can represent contradictory constraints or genuinely inadequate availability; the lab does not establish that every INFEASIBLE_NO_SOLUTION result is a data-quality defect. A processing failure can also belong to a different layer entirely. Maintain a distinction between inability to construct a valid model, inability to find a feasible plan under the chosen constraints, and inability to execute the job technically.

For a proposed troubleshooting runbook, begin with the schedule identity and selected scenario. Confirm that the work list was refreshed after an input change. Verify the schedule window, selected calendar and shift, and the relevant work periods through supported application screens. Then compare planned resource requirements with the availability the model actually received. A staffing spreadsheet is context, not proof that the optimizer received those people and hours.

The lab's missing-capacity symptom illustrates the distinction. People existed in the demonstration roster, but the schedule's shift selection did not produce the expected availability. Hiring more staff would not have fixed that configuration problem. Conversely, a correctly configured model can expose a real shortage. Diagnosis must establish which condition is present before the team proposes either a technical repair or an operational capacity decision. [1]

After correcting an input, preserve enough evidence to explain why the result changed. Record the changed assumption, refreshed population, new solve status and remaining issues. Avoid changing the date window, skill constraints and work query simultaneously if the purpose is to understand a specific failure. Controlled comparisons are an engineering recommendation, not a documented product requirement, but they make the result reviewable.

Infrastructure investigation remains valid when the evidence points there. The practitioner account discusses processing jobs separately from input alerts. For a production environment, use the supported operational runbook and product documentation rather than copying namespace or pod assumptions from a single lab. A missing calendar period and a job that cannot reach its execution service are not the same incident simply because both delay the planning meeting.

A feasibility review ends with a plain-language explanation. Either the inputs were incomplete, the modeled constraints could not accommodate the work, or a technical execution problem prevented evaluation. If the evidence is inconclusive, say so and retain the unresolved diagnosis. Do not make a red status disappear by assigning it a convenient but unproved cause.

3. Keep unresolved work visible after the status turns positive

The second resource-leveling run in the lab reaches OK_SOLUTION but still shows scheduling issues. The author then investigates task durations and remaining skill demand. This is the most useful reliability lesson in the exercise: a positive solution status and a complete operationally acceptable work plan are different deliverables. The published demonstration numbers belong to that dataset, not to an expected result for another plant. [1]

Review unresolved activities individually or in meaningful groups. Determine whether the problem is a missing requirement, an inconsistent duration, an unavailable skill, a date constraint or another documented condition. Link each issue back to its parent work and asset context. Without that linkage, a task-level exception can be mistaken for administrative noise even when it affects an important maintenance intervention.

The source includes a revealing skill example. A remaining requirement asks for a specific mechanic class that has no availability on the selected shift. The demonstration discusses experimenting with matching constraints, but that experiment is not permission to relax production competency requirements. A numerical improvement obtained by turning off a safety-relevant restriction may produce a less acceptable plan, not a better one. [1]

A proposed acceptance rule is to distinguish data correction from operational concession. Correcting an incorrectly entered craft requirement is a data correction, subject to the normal approval process. Supplying another genuinely qualified resource changes available capacity. Relaxing a model constraint changes what solutions the model will permit. Those actions may all alter the issue count, but they have different owners and different risk implications.

Require the relevant process owner to approve a change to required qualifications, execution duration or availability assumptions. The planner should not silently shorten a task just to fit a shift, and the reliability engineer should not assume that a higher-level skill automatically satisfies every configured class relationship. Validate the supported matching rules and the real competency requirement in the local environment.

The issue register should record the selected action and what remains open. Possible dispositions include replanning, providing suitable capacity, moving work under approved prioritization, correcting an erroneous requirement or escalating a genuine inability to execute important work. The options are management decisions supported by the model; the model does not decide the organization's risk tolerance.

This also improves the planning meeting. Instead of announcing that optimization succeeded and introducing exceptions later, present the result and the unresolved-work disposition together. Identify which work is executable, which work remains blocked, and which assumptions were changed since the previous review. A feasible plan with openly governed exceptions can be useful. A nominally clean plan produced by hiding requirements cannot be trusted merely because its solver status looks better.

4. Separate scenario review, dashboard nomination and publication

In the community exercise, the optimizer is run on a created scenario rather than directly from the initial schedule view. The scenario provides a place to experiment and compare before publishing changes back to work orders. The lab also documents that only one scenario per schedule can hold the Scheduling Dashboard Ready designation, with a warning when that designation moves. [1]

Those facts describe separate control points. A scenario may be mathematically valid. A scenario may be nominated for the dashboard. A scenario may have been accepted by the planning authority. A scenario may have been published. None of those labels should be used casually as a synonym for all the others. A dashboard can be understandable and still be showing an experimental choice rather than the approved operating plan.

As a proposed operating control, include the schedule identifier, scenario identifier, source-data refresh, solve result and approval state in the schedule-review record. Before a management meeting, confirm which scenario the dashboard currently represents. Before publication, confirm that it is the accepted scenario and that its open issues have agreed dispositions. Use supported application actions and the organization's change controls rather than altering underlying schedule relationships directly.

This is especially important after comparison runs. A planner may test a longer window, a different capacity assumption or a changed template to understand a shortage. Those experiments are useful precisely because they are not automatically commitments. The review should state which assumption set was accepted and why. Otherwise a promising experimental view can acquire authority simply because it is the one left on screen.

Publication also deserves a readback. The practitioner lab describes publishing as the action that writes scenario decisions back into operational work records. The engineering recommendation is to verify the intended effect on representative records and the supported publication status, particularly when other users or processes can update the same work. Do not treat a saved scenario as proof that field execution records were updated. [1]

Where conflicts or concurrent updates occur, follow the supported resolution process. This article does not recommend blind retries, direct SQL changes or forced updates. Capture the actual conflict, preserve the accepted scenario and investigate which record changed. The relevant question is whether the published operational plan still reflects the reviewed decision, not whether someone eventually obtained a success notification.

The reliability benefit is traceability. When a maintenance intervention slips, the team can distinguish work excluded before modeling, work left unresolved by the model, an experiment never approved, and an approved plan that was not successfully published. These are different failure modes with different corrective actions. Combining them under the phrase scheduling problem makes both root-cause analysis and accountability weaker.

5. Read dashboard indicators with their calculation history attached

The lab makes clear that the Scheduling, Planning and Dispatching dashboards do not all consume the same preparation steps. The Scheduling dashboard reads a nominated scenario. The Planning dashboard requires compliance and cost-and-usage calculations. Dispatching is connected with Graphical Assignment and a distinct preparation path. An optimizer run alone does not establish that all management indicators are current. [1]

For reliability reporting, that is a provenance problem before it is a visualization problem. A utilization value needs a defined resource population and time window. A compliance value needs a defined baseline and measurement period. A cost comparison needs the relevant planned and actual transactions and a completed calculation. Attractive cards do not remove those dependencies.

The practitioner example starts a compliance period to establish its planning snapshot, then finishes that period and runs cost-and-usage calculation separately. It explains that demonstration actuals can include pre-existing sample transactions. Consequently, the numbers on that system are unsuitable as benchmarks for maintenance effectiveness. Their value is showing which process produces the indicator and why apparently surprising values may appear. [1]

A recommended reporting control is to attach calculation context to the review, even if the dashboard itself does not present every field together. Identify the represented schedule and scenario, the compliance period, the known calculation state and whether actuals for the review period are complete. If those facts cannot be established, label the interpretation provisional rather than inventing freshness from the time the browser was refreshed.

The Dispatching section of the source warrants particular caution. Its demonstration encounters a spatial configuration failure, and the author later describes a sandbox database reassociation of an optimizer-run record to populate a view. The source itself says that this is not something to do on a production database. A populated screenshot from that sequence is therefore not independent evidence of an end-to-end supported dispatching deployment. [1]

Production acceptance should require the supported path to generate its own correctly associated results. Do not reproduce the sandbox manipulation to make a management dashboard look complete. If a spatial prerequisite is missing, record the prerequisite and involve the appropriate owner. A visible dependency is a project issue to resolve; concealing it with altered reporting relationships only makes the final indicators harder to trust.

For the reliability meeting, present the decision that the indicator actually supports. A feasible resource-leveling view helps evaluate modeled workload and capacity. Compliance calculation helps assess execution against a baseline. Assignment information concerns named execution resources. None independently proves that asset failure risk is controlled. Together, with the correct population, calculation history and exception disposition, they can inform an accountable maintenance decision rather than a decorative status update.

Practical Implications

For reliability engineers, add a work-population reconciliation to schedule acceptance. Confirm where the risk-relevant maintenance work went before discussing utilization. Work excluded for missing labor plans may need planning attention even when exclusion is technically appropriate for the model. Keep unresolved activities linked to their parent work, asset context and approved disposition.

For planners and schedulers, preserve a small evidence pack for the accepted run: selected scenario, refreshed input population, important constraints, solve result and open issues. Separate input repairs from deliberate changes to operating assumptions. Do not weaken a competency or duration requirement merely to improve a result. Escalate the actual capacity or readiness decision to its owner.

For Maximo application owners, make dashboard nomination and publication part of the operating runbook. A successful solve is not a substitute for verifying the approved scenario and its published effect. Establish supported troubleshooting paths for conflicting updates and spatial prerequisites. The sandbox reporting manipulation described in the community source should remain a cautionary example, not a production fix.

For maintenance managers, ask for calculation context alongside indicators. Determine which scenario is represented, which baseline is being assessed and whether the relevant calculations and actual transactions are complete. These recommendations do not require a new analytics platform. They require the organization to stop treating a positive solver status, a populated dashboard and an accepted executable plan as the same fact.

Bottom Line

The optimizer lab is useful because it documents a progression rather than a miracle: incomplete inputs, a failed attempt, corrections, a feasible result and remaining work that still requires judgment. It also exposes the preparation behind the dashboards and a sandbox shortcut that should not be promoted into production practice. Those boundaries are as valuable as the successful screens. [1]

For a reliability team, acceptance starts before the solve and ends after the chosen plan has been reviewed and its operational effect verified. Reconcile the input population, investigate the actual failure layer, govern unresolved work, identify the accepted scenario and understand the calculations behind the indicators. Keep every experiment separate from a change to the operating commitment.

A feasible schedule is a useful model result. It becomes an accountable maintenance plan only when the organization can explain what work it contains, what work remains outside it, why the constraints are legitimate and who accepted the resulting exceptions. The strongest planning report is not the one without visible problems. It is the one in which every consequential problem has an accurate explanation and an owner.

Source: [1] Mohamed Chihaoui, Maximo Optimizer hands-on scheduling and dashboard lab. This is a practitioner demonstration using sample data on Manage 9.2.4 and Optimizer 9.2.3. Product observations are attributed to that environment; the acceptance controls in this article are editorial engineering recommendations, not claims of measured production improvement.