A component that has not failed still has a history

## Start with the histories that the average leaves out

The fifty-four completed failure histories were the easiest part of the 2023 study to explain. Each had an observed event and a time associated with it. A proposed management summary averaged those times and described the answer as expected component life across the fleet. The calculation looked accessible, and service planners needed an assumption they could use. The difficulty was that the study contained 360 histories, not fifty-four. The proposed average gave the clearest records authority to speak for populations it had left outside the calculation.

At the common cutoff, 246 components had remained under observation without the defined first failure. Sixty others had left the service record earlier for documented reasons. Neither group supplied a completed failure time in the same sense as the fifty-four observed events. That did not make their histories irrelevant. The period during which a component was observed without an event can matter to a reliability analysis. Discarding it simply because the desired endpoint has not appeared changes the evidence being used to answer the question.

The opposite shortcut would also have been wrong. We could not call the remaining 306 permanently failure-free. A component still functioning at a cutoff has an unfinished future. A component no longer observed may subsequently have failed, continued operating or changed use in ways the available record does not establish. A status at the end of observation is not a complete account of its lifetime. The distinction sounds elementary when stated plainly, yet a management table can conceal it behind a tidy binary label.

I wanted the study to retain the observation start, the event definition, the last supported date and the reason the history ended. Qualified reliability colleagues would then determine what methods and assumptions were appropriate. We did not need a data office to announce a statistical cure for every incomplete record. We needed an evidence account that did not erase the incompleteness before the specialists saw it. A technically sophisticated model can still produce an unhelpful answer if the history supplied to it misstates what was observed.

The service-planning sponsor was not unreasonable in asking for a number. Resource commitments cannot always wait for every component to finish its life. The useful response was therefore more demanding than rejecting the proposed average. We needed to explain why it did not describe the fleet and what further work could support a planning assumption with stated limits. That explanation should help the business choose its next step, rather than make uncertainty sound like a reason never to decide.

## Preserve why observation stopped

A common administrative cutoff is different from an earlier departure whose cause may be related to equipment condition. If a service relationship ends, the subsequent history may be unavailable. If a unit is withdrawn because of an emerging concern, treating its departure as an unrelated interruption could conceal evidence relevant to the risk being studied. The record needs to preserve that distinction even when both situations initially appear as an end date in the same column. A date alone does not explain why observation ceased.

Our sixty earlier departures were therefore retained with their documented reasons. We did not assume that every reason was unrelated to failure risk, nor claim that the existence of a code established the assumption. Reliability specialists needed to examine whether the available classification was adequate for the proposed analysis. Some questions might require sensitivity work, a narrower population or additional follow-up. The appropriate response depends on the actual evidence; it cannot be chosen merely because one method produces a convenient complete table.

We spent ₹49 lakh reconstructing starts, last-seen dates and the conditions attached to the histories. That cost did not purchase a new physical life for the components. It improved the basis on which a qualified analysis could be considered. The case does not contain a replacement expected-life estimate, because the immediate decision was to withdraw the unsupported fleet interpretation and preserve the evidence for further work. I would rather state that limit than attach a precise number whose derivation the account cannot defend.

Event definitions also deserve attention. A first failure under a study definition, a repair action and a replacement can be related without being interchangeable. A component may be replaced preventively, and a work order can close without the event the study is meant to measure. Those distinctions belong with the responsible engineering owners. The data function should make their definitions usable and retain changes over time, not silently substitute whichever field happens to be most populated. Otherwise the analysis may compare records that share a label while describing different occurrences.

The same discipline applies to the beginning of observation. A history entering our system does not necessarily begin with a newly installed component. We need to know what earlier service is documented and what remains unknown. I would not treat an acquisition date for the dataset as proof of equipment age. The useful record distinguishes the physical history from the period our company has observed. That may limit the study, but a limitation stated early is more useful than a confident conclusion built on an invented starting point.

## Give planning a usable account without inventing certainty

After the review, the company still had to plan service resources. The corrected account did not make that responsibility disappear. It changed the material management could responsibly use. The fifty-four observed first failures remained events with operational significance. The 246 histories reaching the cutoff and sixty earlier departures remained part of the evidence population. A qualified analysis could consider their exposure and limitations, while commercial and engineering owners decided what assumptions were suitable for the particular planning purpose.

A useful planning discussion can distinguish an observed fact, an analytical estimate and a deliberately cautious business assumption. Those categories need not be enemies. The problem arises when an assumption is presented as though it were directly observed, or when a study result is applied beyond the conditions it examined. I want the account to make it possible for a later reader to identify which kind of statement is being used and who is responsible for accepting it. That is especially important when the same figure moves between technical and commercial papers.

I would build the next observation design before commissioning the next lifetime study. The team should agree the population, relevant event, follow-up arrangements and handling of departures while it can still improve collection. That is more useful than discovering at the end that the cleanest records are not the records needed for the intended question. It also gives field colleagues a reason to preserve apparently inconvenient details. A reason for leaving observation may matter more to the later interpretation than another decimal place in a completed measurement.

For a board, the oversight issue is whether management's account reveals the limits that could materially change a commitment. Directors should not select a reliability model in place of specialists. They can ask whether a fleet claim relies only on completed failures, whether departures have been examined and how uncertainty affects the proposed course. Those questions connect evidence to responsibility without turning the board into an operating study team. They also make it harder for a simple average to acquire more authority as it travels upward through the organisation.

The case taught me to value the histories that do not end neatly. They are not empty space around the observed events. They describe what was seen, for how long and under which conditions the record stopped. Preserving them did not produce a triumphant performance improvement or a completed answer to every reliability question. It produced a more defensible starting point. A component that has not yet supplied the desired endpoint still deserves an accurate place in the account, and the people planning around that account deserve to know where observation ends.