National Archives Repository Health

The AITS Content Tracking System (CTS) gives an executive one number: a composite Repository Health score. At 98.4%, you know your holdings are soundly described and findable, and you can say so with confidence in any room. At 64%, there is work to do, and the breakdown behind the score shows exactly what it is.

BI charts and reports show you the data. CTS uses AI and machine learning to tell you what it means: how healthy the repository is, what is driving the score, and what to do to improve it.

You choose what healthy means. The score is built from the KPIs that you and your managers select and weight. AITS recommends a proven starting set, and you add to it, drop from it or change the weights as your priorities change.

Behind the score is AI trained on your own holdings. It begins with a health check that establishes where the repository stands today, then keeps the score current around the clock, for every record and description as it is added.

For example: a file unit whose title is only a number, a digitized record with no searchable text, or an item whose dates fall outside the dates of its own series. CTS flags it, scores its effect on repository health, and recommends action, before a researcher or an AI assistant fails to find the record.

What you receive: the score, the breakdown behind it, and a prioritized list of what to fix first. It is evidence you can show a minister, a board or an auditor.

You do not have to imagine it. The live demo shows CTS measuring a real public collection.
Open the dashboard and click on the health score to see the composite makeup.

Most archives cannot say. That is the problem: not that the holdings are known to be poorly described, but that the condition of the catalogue as a whole is not known at all.

An archive holds records transferred over generations, described at different times, to different standards, by different hands. Some series are described down to the item. Others carry a title and little else. Digitization adds images faster than descriptions can keep up. Problems usually surface one at a time: a researcher who cannot find a file, a request that takes weeks, a digitized record nobody can search. Until then, nobody could have known how many more there are.

An archive's holdings are the permanent memory of a government and its people. As you know, three things follow from that:

• Permanent: nothing ages out. A record described poorly today stays hard to find for as long as it is kept, which is forever.
• Public: the catalogue is the only way in. A record that cannot be found through its description might as well not be held.
• Trusted: courts, historians, journalists and citizens rely on the record being authentic, in context, and correctly identified.

Each of those rests on the repository: the descriptions, the hierarchy that gives every record its context, the digital copies and the access decisions recorded against them. A record is only as usable as its description, and a gap in the catalogue is invisible to everyone searching it.

This matters more as archives adopt AI. AI search and AI assistants are only as reliable as the descriptions they work from, and the gains come from finding issues early: each one corrected improves the next result, and the improvements compound.

AITS pointed CTS at catalogue descriptions the US National Archives publishes as open data, checked every one against a set of KPIs, and rolled the results up into one composite Repository Health score. Two pages show the result:

• The CTS Dashboard: the score at a glance, with the breakdown behind it and the recommended actions. This is what an executive sees.
• The Health Report: the dimensions, every KPI with how many records passed, the actionable recommendations ranked by the gain each would bring, and where the risk sits series by series.
• Every record, not a sample: each description was checked as it was loaded, the same way CTS assesses each item as it arrives in an engagement.
• Seven of the eight dimensions were measured. Security was left out, because everything in the collection is already public.

Open the CTS Dashboard Read the Health Report

What this is and what this is not. The score is measured from the published catalogue descriptions, not from the documents themselves, and against KPIs AITS chose for this presentation. The National Archives would choose its own. Nothing was edited, and the National Archives' own systems were not involved. The demo is not endorsed by, or connected with, the US National Archives.

CTS is not a fixed set of rules applied to every repository. Its AI is trained on your own records, so it learns what a sound record looks like here. The measuring is automated. The standard being measured against is yours.

It learns how you solve problems. When staff confirm, correct or dismiss a finding, that decision trains the model. CTS learns not only to recognize an issue but how your people resolve it, so the alerts get sharper and the number needing review falls.

You decide how much it does. Every type of issue starts with CTS recommending and a person acting. When you are satisfied it handles a type correctly, you can let it prepare the fix for approval, and later give it standing permission to resolve that type itself. Staff check a sample, every action is logged and reversible, and you can withdraw the permission at any time.

The description stays with the archivist. CTS recommends corrections to repository records: titles, dates, links, duplicates, missing fields. It never alters a record or decides what it means. Appraisal, arrangement and description remain the archivist's.

People stay in charge. CTS learns from your archivists and records staff, and their judgement is what it cannot replace. It takes the repetitive work off their desks and leaves them the review, the exceptions and the decisions.

The repository improves with it. Every issue corrected raises the score, and the trend shows whether the repository is getting healthier or slipping. Better descriptions make the AI more accurate, and a more accurate AI finds the next issue sooner. The two improvements compound.

CTS does not run against the official repository. The archive's environment is replicated in a secure cloud environment hosted in-country, and the AI is trained and run on that replica. It works alongside the systems you already have, with no platform replacement.

Every item is assessed, from the first to the last. The replica starts empty. Your content is loaded into it in bulk, and CTS assesses each item as it arrives. Existing records, newly digitized records and new accessions all go through the same check, so the score covers the whole repository and not a sample.

The official repository remains untouched. Nothing CTS does is destructive, and any result or correction can be compared against the original at any time.

Corrections are made and checked in the replica first. Applying them to the official catalogue remains the archive's decision, through its own change controls.

Health is not one property. CTS scores the repository across eight dimensions, each built from KPIs the archive defines and weights. These eight are the starting set AITS recommends, and the ones used in the live demo; each rolls up into the composite score, and each can be opened to see what is pulling the score down.