Why a High-Scoring Index May Still Need Correction

You receive an index with a strong evaluation score. The headings read well, most page references work, and the formatting looks finished. Before sending it to production, inspect the defects. One missing central conclusion or misleading heading can matter more to a reader than dozens of successful lookups.

In Can AI Evaluate a Book Index?, we separated a weighted quality score from publication readiness. Here is how to use that distinction at the editorial desk: identify the reader's failed lookup, choose the repair, and verify that the repair actually restores access.

Begin with the source

Imagine a 180-page book about rental assistance. Pages 22–28 discuss assistance and tenant protections; 61–64 explain eligibility; 90–94 describe applications. The central conclusion, on 142–48, is that vouchers alone cannot guarantee affordable rents when housing supply is constrained. That conclusion qualifies the earlier discussion and deserves findable access.

Index A: useful destinations, missing argument

• Application process, 90–94
• Eligibility, 61–64
• Rent assistance, 22–28, 90–94
• Tenant protections, 22–28
• Vouchers. See Rent assistance

Assume these are the only routes to those subjects anywhere in Index A. Every listed page leads to useful treatment, and the see reference resolves. But a reader looking for the book's conclusion about vouchers finds only the earlier discussion. Nothing reaches 142–48 or names the supply constraint.

Checking the listed pages cannot reveal this omission. The reviewer must work in the other direction: take the source's central conclusion and try to retrieve it through the index. Strong locator precision says little about a passage for which no locator was supplied.

Repair A by adding source-supported access, such as “Vouchers: limits under constrained supply, 142–48,” with appropriate access under housing supply as well. Reconsider the existing see reference: vouchers now have their own substantive entry, so a simple substitution to Rent assistance no longer serves the same purpose. Check the surrounding structure rather than attaching a page number to an unchanged trail.

Index B: more coverage, false promises

• Application process, 90–94
• Eligibility, 61–64
• Rent assistance, 22–28, 90–94
• Tenant protections, 22–28
• Voucher eligibility. See Housing eligibility
• Vouchers: guaranteed affordable rents, 142–48; supply constraints, 204

Index B shares A's useful destinations and apparently includes the conclusion. Yet “guaranteed affordable rents” reverses the author's position. Page 204 does not exist in this book. Assume Housing eligibility is absent throughout the index, so that reference ends at a missing target. Extra entries have introduced three separate failures.

Repair B by replacing the reversed heading with wording that preserves the limitation, verifying the supply-constraint locator against 142–48, and directing Voucher eligibility to the existing Eligibility entry if the passages justify that equivalence. Do not merely replace 204 with a plausible number: the complete heading must fit the destination. Inspect other locators and references for the same failure pattern.

Both excerpts have recognizable subjects, compact wording and several correct destinations. Their consequential defects require different work. A needs access reconstructed from the source; B needs claims and navigation corrected. Neither excerpt is large enough to justify a whole-book quality score.

Three answers that must stay separate

A weighted score combines dimensions such as coverage, fidelity, locator reliability and navigation. A result of 90/100 describes performance under the declared rubric. It does not mean that 90% of entries are correct, nor does it tell you which repairs remain. Different profiles of strengths and weaknesses can produce similar totals.

A measured rate answers a narrower question. In a hypothetical audit, 19 substantively supported locators out of 20 inspected means 95% supported within that inspected set. It says nothing by itself about omitted conclusions, untested locators or broken cross-references. Report the denominator and audit scope beside the rate.

Publication readiness asks whether consequential defects remain under the agreed rules. The Subject Index Evaluation Standard 0.2 treats central omissions, serious stance distortions, fabricated or nonexistent locators and invalid cross-references as independent gates. Strong results elsewhere cannot cancel a failed gate. State the method and version; different versions must not be treated as interchangeable.

These checks also have a professional basis: the American Society for Indexing's award criteria ask whether significant concepts are covered, alternative access works, and references avoid circular routes or missing targets. Those criteria support editorial inspection; they do not prescribe our numerical score or readiness gates.

Before approving the repaired index

• ✓ Read the defect inventory and audit scope alongside the score. Uninspected evidence leaves a question open.
• ✓ Test central subjects and conclusions from the source toward the index, including their likely reader terms.
• ✓ Check corrected headings and locators together against the final paginated source.
• ✓ Follow every changed reference to a useful destination; check whether related routes also need repair.
• ✓ Verify each blocking defect after correction and assess whether its cause affects other entries.
• ✓ Record the resulting readiness assessment separately from the responsible editor's release approval.

A strong score helps describe an index. An evidence-linked repair record tells the editor what still needs attention. Approve the corrected version only when the required checks support that decision, rather than carrying forward the assurance of an earlier total.