How to read recommendations

A recommendation is a short statement in our own words about what to do with one table or one group of tables, followed by the sources that back it. We give the link and a summary; we do not copy the source. Where a menu path or a field name matters, it is written exactly as D365 shows it.

There is no single best answer for every system. Microsoft's own guidance says a cleanup should be run only after the business has confirmed the data is not needed and after it has been tried in a test environment; we say the same on every page.

The top of each recommendation answers three different questions, each with its own label. They are kept apart because a well-sourced action can still be dangerous, and a risky one can come from an excellent source.

A source that only retells another one does not count as independent, so a page copied from Microsoft's documentation never raises a recommendation above its original.

“Useful” records that a signed-in reader with a confirmed email address found the recommendation helpful. It does not mean the advice is safe or correct: a simple, risky recipe can collect more clicks than a careful one, which is exactly why trust, risk and reversibility are shown separately and cannot be raised by clicks.

A reader can mark a recommendation once. If the meaning of its text later changes, the earlier marks stop counting, because they endorsed different words. Correcting a typo does not reset them. Site administrators do not vote.

You can switch the order on any table or group page:

• Recommended (the default): those with a warning status last, then by source trust (A before B before C), then by how many readers found it useful, then by the date of the newest source.
• Most useful: by the number of readers who marked it useful, then by source trust.
• Newest: by the date of the newest source.

The default puts an official source ahead of a popular weaker one. When nothing has been marked useful yet, the order is simply by trust.

Each recommendation lists its sources with their trust letter, publisher and date. Links open in a new tab and are marked nofollow. A person's name and role appear next to a source only after we have checked the attribution; until then you see the publisher and the link.

A recommendation is published only after a person has compared our summary with the source and found them consistent. The badge “Checked by a person” shows this. If you ever see “Not yet checked by a person”, treat the recommendation as a draft.

A group needs the same check, because a recommendation for a group applies to every table in it at once.

A coloured strip at the top of a card means the recommendation has a status you should read first:

• Preview: the feature is not generally available, and its behaviour or availability can change.
• Not available yet: announced but not released.
• Deprecated or replaced: an outdated approach, or one a newer recommendation supersedes.
• Status not verified: we could not confirm from a source whether the feature is generally available.

These strips stay visible whatever the number of readers who found the card useful.

Dynamics 365 finance and operations keeps its live data in its own SQL database. Dataverse is not where it lives. Microsoft's archive feature is optional and is switched on by hand: it copies inactive data to Dataverse long-term retention, which acts as external cold storage with its own capacity, and moves the rows into history tables in the same SQL database, removing them from the live tables.

This is why an archive is not the same as a cleanup, and why we show two things for the size:

• the live table shrinks;
• the history table grows, and the database as a whole shrinks only after the history is purged.

When we last read Microsoft's page on purging archived history (September 2026), it was a preview feature and was described as temporarily unavailable in production. Check the source on the recommendation, because this can change.

A table can have both a delete recommendation and an archive recommendation. They pursue opposite goals: one removes data for good, the other keeps it for audit. Decide first whether you must keep the data; the card marks such pairs with “Also see”.

More about how the size numbers are measured is in the D365 setup guide.

Clients can report what happened after they applied a recommendation to their project: it worked, worked partly, had no effect, or broke something, and how much space it freed. A card shows the total only when at least three different projects have reported, only counts projects that share their statistics, and never names a project.

On a Microsoft-managed production environment you have no direct access to the SQL database. A recommendation that relies on your own SQL or X++ is therefore marked as unofficial, and it can only be tried where you do have that access, such as a developer environment. Recommendations may also say they apply to a sandbox, to cloud production, or to on-premises systems only.