The sample that found the fault could not measure the whole collection

A quality review can do its job well and still be unsuitable for the claim someone wants to make from it. Our selected set of 520 pages exposed fifty-two reading-order problems in a 24,000-page collection. The result justified action. It did not establish that one in ten pages across the collection had the same problem, or that correcting the observed paths made the entire delivery error-free. I wanted the report to preserve that distinction. A useful diagnostic finding should lead to a better decision without acquiring an authority that its selection method cannot support.

The collection had passed file-completeness checks. Reviewers then noticed that some search extracts combined passages from the wrong parts of a page. We deliberately included complex layouts alongside more straightforward examples to understand the conditions behind the issue. The fifty-two affected pages helped the team identify transformation paths needing attention. I authorised ₹46 lakh for four weeks of correction and review work, retaining the original scans as the reference. The account described the known faults and the response, while leaving broader quality questions open to an examination designed for that purpose.

## Choose the sample for the question being asked

A diagnostic selection seeks informative cases. It may concentrate on unusual layouts, ambiguous structures or conditions a reviewer suspects will expose a weakness. That can be exactly the right approach when the immediate question is how the process fails. A representative estimate asks a different question and needs a selection and analysis suited to the population it describes. I do not rank one above the other. I ask the team to state which job the review is meant to do, so the result is interpreted consistently when it moves from the specialist discussion into a management or customer summary.

Our mixed selection was useful because it brought different page structures into view. It was not presented as random, and its composition reflected the diagnostic purpose. The observed count could therefore describe the selected set without becoming a population estimate. That may sound like a narrow distinction, but the commercial consequences are substantial. A customer deciding how to use searchable text needs to know whether a number describes examined pages, a reasoned estimate of a wider collection or a correction applied to a particular mechanism. Those statements can all be valuable, but they are not interchangeable simply because each contains a percentage.

File completeness was another separate measure. The expected files could be present while their searchable text failed to preserve the source's reading order. A process that checks only delivery volume will not necessarily reveal that kind of problem. I want the quality plan to connect measures with the user's task. Can a reader find the relevant passage, associate it with the correct page and understand enough context to inspect the source? The answer may require several kinds of evidence. A single completion figure becomes misleading when it is allowed to imply that every useful property of the result has been examined.

## Report the correction at the scope examined

The correction programme addressed the observed transformation paths and reviewed the known examples with collection specialists. We retained the scan and made the relationship to corrected text visible. That was a meaningful outcome. It did not require us to claim that the original source had changed or that the searchable representation could replace it for every use. A user may still need to inspect the page, particularly where layout or interpretation matters. I want the product to support that inspection rather than let a convenient extract become the only visible account of the material.

The report also needed to distinguish a repaired known example from confidence about another condition. Passing the pages used to diagnose a fault can show that the correction addresses those cases. It may not reveal a different layout the team has not yet examined. Further review should be planned according to the intended use and consequences, with its scope explained. I would not want every correction to trigger an unbounded promise to inspect everything, but neither should the business use the cost of further work as a reason to state confidence it has not earned. The decision should make the remaining question and its owner clear.

We kept the ₹46 lakh cost and four-week schedule in the account. A quality improvement can be worthwhile while still requiring a commercial choice about time and resources. Concealing that cost would make future planning less informed. I would rather a programme budget recognise the possibility of finding and correcting a meaningful issue than create a schedule whose success depends on every review being reassuring. The purpose of the review is to learn something consequential before the user relies on the output. If the organisation treats every adverse finding as a failure to protect the launch, it can discourage the evidence it needs.

## Give management a decision rather than a decorative percentage

Commercial colleagues wanted one figure for the finished collection. I understood why: a concise number is easier to put in a delivery summary than an explanation of selection and scope. I declined to infer it from our diagnostic set. We could provide a clear account of the observed issue, the correction and the additional examination required for a broader claim. That gave the client a basis for deciding how to use the searchable text. A percentage whose meaning could not survive a question about the sample would have made the summary shorter while making the decision less reliable.

This discipline also matters in internal reporting. A dashboard can combine figures from different exercises and make them appear comparable even when their samples, definitions or purposes differ. The board or executive team may then see an improving trend that reflects a change in measurement rather than a change in the service. I want the owner of a material measure to explain what it counts and when the basis has changed. The explanation need not overwhelm the report. It should be sufficient for a decision-maker to know whether the number answers the question being asked or only resembles the kind of answer expected.

Directors can ask what a quality claim is based on, which use it supports and what remains outside its scope. They should understand whether management has funded a response to a known fault or established a broader assurance about the service. They do not need to design the page sample themselves. Their responsibility is to connect the evidence with the company's material promise and ensure that an attractive summary does not hide a limit that changes the decision. Where the consequence is specialised, the board should expect appropriate expertise rather than rely on the apparent precision of the number alone.

I would make the next quality review begin with the intended claim. If the business needs to diagnose a failure, choose cases that help explain it. If it needs an estimate about a wider collection, design an examination capable of supporting that estimate. If it needs to verify a correction, retain the conditions that make the result interpretable. Our fifty-two problem pages were valuable because they changed the work we did. Their value did not depend on becoming a statement about every page we had not reviewed. Good data leadership keeps that distinction intact as the evidence moves towards the people authorised to act on it.