# A protocol for knowledge freshness.

A proposed way to check whether learning still reflects the approved source, including a controlled outdated-source test.

Updated October 20, 2026

## Define the unit

Use a source-linked learning artifact as the unit of review. Each record needs a source owner, applicable version, audience, review status, and last verification date. The presence of a URL alone cannot establish that a source is authoritative or current.

## Introduce a safe change

In a disposable test set, alter an approved policy and preserve its prior version. Identify which lessons, answers, and assessments should be affected. Ask the workflow to surface the change without automatically releasing revised material.

## Score the result

Measure correctly identified affected artifacts divided by all artifacts known to be affected. Separately count false alarms and missed critical changes. Record time until reviewer-approved material becomes available in Web, and in Chat once it ships. Use a known unaffected control to check excessive alerts.

## Repeat the test on a schedule agreed in advance

Decide how often the test runs before anyone sees a result: after each policy release, after each product release, or at a fixed interval the source owners accept. Record which sources were in scope, who changed the test policy and who scored the outcome. Keep the test set apart from live learning, and delete the altered policy when the run ends so it never reaches a learner.

## Publish limitations

For Elvina, automatic detection of source changes is planned, not available today, and this protocol does not establish that Elvina currently implements a freshness monitor. An actual report would require an environment, release, dataset, observed outcomes, and reviewer. Until then it is a reproducible acceptance design rather than a performance claim.

No launch date is committed for planned features.