Building an Institutional Item Bank With a Review and Approval Cycle

Three years into operation, a department head opens the item bank and finds fourteen thousand items. Then tries to answer a simple question: how many of these are fit to use in a final exam tomorrow?

Nobody knows. Some were written by faculty who have left. Some are duplicates in different wording. Some reference material removed from the syllabus. Some have never been administered, so there is no data on how they perform. The bank is not an institutional asset. It is an item graveyard.

That outcome is not an accident. It is the natural consequence of a bank with no lifecycle. When entry is open to everyone, deletion is forbidden, and review is optional, bloat is only a matter of time.

This article is a practical guide to building a bank that gains value over time instead of accumulating mass: the item lifecycle, who reviews and against what criteria, how versioning works, how the bank is organized and legacy content migrated — with a four-phase rollout plan.

Key Takeaways

A thousand approved items beat ten thousand unreviewed ones. Size is not a quality indicator.
A mandatory four-state lifecycle — draft, pending review, approved, archived — is the backbone of the bank.
Editing an approved item creates a new version rather than overwriting the old one; otherwise historical reports are corrupted.
Review needs written criteria, not personal taste. Without a checklist it becomes a bottleneck that adds nothing.
An institutional bank needs declared ownership: who authors, who reviews, who approves, who retires.

Why Most Item Banks Fail

Growth with no retirement. Items go in and never come out. And because deletion is prohibited to protect records, the bank grows without limit. The remedy is not deletion but archiving: removing an item from circulation while keeping it tied to the assessments that used it.

Absent review. When any instructor can push an item straight into use, keying errors, ambiguous stems, and dead distractors reach live examinations.

Invisible duplication. The same item in two wordings appears twice in one exam. This happens because search relies on text matching rather than on outcome and topic.

Lost ownership. A departing faculty member leaves hundreds of items with no custodian, so nobody edits them and nobody retires them.

A bank with no data. Items never administered, or administered without psychometric analysis, leave nobody able to say which ones work.

The Item Lifecycle: Four States

The foundation is a strict state machine, not an optional "ready" flag:

1. Draft. The item is being authored. Visible only to the author and not insertable into any assessment, however complete it looks.

2. Pending review. The author has submitted it, so it locks to them and becomes visible to the reviewer. It returns to draft only by the reviewer's decision, with a written reason.

3. Approved. It has passed review and can be used in assessments. From this point, any edit creates a new version that starts its own cycle at pending review.

4. Archived. Removed from circulation because it has aged, leaked, or been shown to be faulty by psychometric analysis. It stays linked to prior assessments and does not appear in default search.

The governing rule: state transitions happen only through a declared action recorded in an audit trail — who moved the item, when, and why.

Who Reviews, and Against What Criteria?

The minimum four roles:

Role	Authority
Author	Creates and edits own drafts, submits for review
Reviewer	Approves or returns with a written reason
Bank custodian	Archives, merges duplicates, manages organization
Quality manager	Views coverage and accreditation reports

The critical constraint: an author does not review their own item. In small departments where that is hard, use reciprocal review between two colleagues or review from an adjacent department.

The Review Checklist

Without written criteria, review becomes personal taste and a bottleneck that adds nothing. A reasonable minimum:

Does the item actually measure the outcome it is mapped to, rather than an adjacent one?
Does its cognitive level match the outcome's level without exceeding it?
Is the stem clear and self-contained, with no missing information and no misleading surplus?
Is the key definitively correct, and is it the only correct answer?
Is every distractor plausible and built on an anticipated error, rather than filler?
Are the options consistent in length and phrasing? (The longest option is often the key — a cue students learn to read.)
Is the item free of linguistic giveaways, such as absolutes like always and never in the wrong options?
Is it free of cultural or contextual bias that penalizes a group for reasons unrelated to the content?
Is the language correct, and do symbols and equations render properly?
Does an equivalent item already exist in the bank, measuring the same outcome the same way?

Item ten is the most neglected and the most effective at preventing bloat.

Versioning: Why an Approved Item Is Never Overwritten

Picture an item used in a final exam, after which a keying error is discovered and corrected in place. Every report issued for that exam now references an item that no longer exists in its original form, and any subsequent review will compare student scores against a key that was never applied to them.

Hence: a version is cloned, not edited. Editing an approved item creates v2, which starts at pending review, while v1 locks and remains attached to every assessment that used it.

There is an essential distinction here:

Correcting a wrong key on an exam already administered → not merely a new version; it requires rescoring the whole cohort in a single reliable transaction, with the reason documented.
Improving the wording for future use → a new version only, with no retroactive effect on any grade.

Conflating the two is a common source of lost confidence in a system.

Organization: Hierarchy and Tags Together

The bank needs two organizing axes, and neither substitutes for the other:

A tree structure (course → topic → outcome) for browsing and reporting. This is the formal structure alignment is built on.
Free tags for operational use: final exam, image needs review, temporarily withdrawn, suitable for practice. Fast, and they do not corrupt the structure.

The bank also needs a distinction between a personal bank (an author's drafts and experiments) and an institutional bank (shared approved items). Blurring the two floods the shared bank with immature material.

Migrating Legacy Items

Most institutions start with thousands of items in Word files, spreadsheets, and legacy systems. The common error is importing all of them at once in an approved state.

The safer path:

Import to draft without exception, whatever the source.
Automated triage first: detect textual duplicates, items with no key, incomplete option sets, and items with no linked outcome.
Human triage by priority: start with high-enrolment courses, not alphabetical order. The return on review is highest where the student numbers are.
Gradual approval: no item moves to approved before passing the checklist, even if it has been in use for years — longevity is not evidence of quality.

Expect a high rejection rate. It is normal for a substantial share of a legacy bank to be set aside in the first serious triage. That is not a loss; it is overdue cleaning.

Continuous Improvement After Approval

Approval is the start of an item's cycle, not the end. After each administration, data accumulates and informs decisions:

Facility and discrimination indices identify which items work and which are faulty.
Negative discrimination calls for immediate suspension and a key review.
A dead distractor calls for replacing that option, not the whole item.
An item with high exposure — used across many assessments — becomes a candidate for precautionary archiving.

Bank health indicators worth reviewing periodically: the ratio of approved to total items, the proportion actually used within a year, coverage of every outcome by a sufficient number of approved items, and average review turnaround.

The most dangerous indicator: a learning outcome covered by only two approved items. It means attainment for that outcome rests on two items, and one faulty item corrupts the entire report.

A Four-Phase Rollout

Phase one — foundation (one month). Build the curriculum structure (course → topic → outcome), define roles and permissions, and write and formally adopt the review checklist.

Phase two — pilot (one term). One or two courses only. Build an approved, outcome-mapped bank with no fewer than four approved items per outcome, and run it in live assessments.

Phase three — expansion (two terms). Extend to the full department and begin legacy migration by priority. Turn on bank health reports and review them monthly.

Phase four — maturity (ongoing). A data-driven improvement cycle: review faulty items each term, archive precautionarily, and feed accreditation reports directly from the bank.

Do not start with expansion. The item bank projects that fail most often are the ones that began by trying to migrate everything in a single term.

How EvaliX Addresses This

The item bank in EvaliX runs a mandatory four-state lifecycle with an approval gateway preventing any item from entering an assessment before review. Any edit to an approved item creates a new version and locks the previous one to the assessments that used it, so historical reports remain exactly as issued.

The bank combines a tree structure tied to the curriculum with free tagging, and separates personal from institutional banks. Psychometric analysis runs continuously across every administration and surfaces faulty items to the bank custodian without anyone going looking.

And when an answer key is corrected, the platform runs an atomic rescore across the full cohort, documented in an audit trail recording who performed it, when, and why.

Request a demo to review a phased migration plan for your existing item bank.

FAQs

How many items does a course need?

Tie the number to outcomes rather than to the course. A reasonable floor is four approved items per learning outcome per targeted cognitive level. If you want randomized draws giving each student a different form, multiply by at least three.

Who owns an item — the instructor or the institution?

This is a policy question rather than a technical one, and it should be settled in writing before the bank is built. Common institutional practice is that items authored as part of official duties are institutionally owned, with authorship attributed to the individual in the system record. The absence of that provision creates disputes when a faculty member leaves.

Should weak items be deleted?

No. Archive rather than delete, because deletion severs the link to prior assessments and corrupts student records. Archiving removes the item from circulation while leaving the history intact.