Datasets

What it does

Behaviours, not fields. Choosing between two modules is choosing between these.

Freshness per dataset, because a stale file is how a wrong answer gets given confidently.

Row counts and sources, so an import that half-failed is visible.

Named sources other modules read from, rather than a grid somebody works in.

How it works

1
Datasets are named sources other modules read from, rather than grids somebody works in.
2
Row counts and refresh times sit together, so a half-failed sync is visible.
3
A stale dataset is obvious rather than discovered through a wrong answer.
4
Modules pointing at a dataset follow it when it refreshes.

Best for

Budgets, quantities, anything people would otherwise export to Excel.

Not the right choice for

Records with long text — the grid fights you.

What a row holds

Every field, and what the agent listens for to fill it — which is the difference between a page you type into and one that fills itself.

Field	Type	Filled from
Datasetrequired	text	What the data is, named for what it holds.
Rowsrequired	number	How many, so a half-failed import is visible.
Sourcerequired	text	Uploaded or synced, and from where.
Refreshedrequired	text	When it last updated. Freshness is the whole point of this module.

What it does not do

A module chosen for something it cannot do costs more than one that was never offered.

No schema validation between refreshes, so a changed column breaks the modules downstream.
No diffing — you cannot see what changed in the last refresh.
Sync failures show as a stale date rather than an error.
Large datasets are not paginated here; this is a manifest, not a browser.

Before it works

What you have to provide, if anything.

Name datasets for what they hold, not where they came from. The source changes; the content does not.
Pair with Data table when somebody needs to look inside one.

In practice

A shop whose trade prices came from a monthly CSV.

“Are we quoting the right trade prices?”

What was said to the agent

What the agent built

A dataset list with row counts and last-refreshed dates.

What changed

The postcode zones were from March. The price list was current, but the delivery zones had been wrong for five months.

Better with

These do a different job on the same data. An agent can have all of them.

Data tableLive

The data is genuinely tabular, arbitrary, and being worked with directly rather than presented.

FilesLive

Files need somewhere to live and the only questions are who uploaded it, how big it is, and can I download it.

AutomationsLive

Repetitive reactions are being done by hand, and the trigger is something the system already knows about.

Questions

What data does the Datasets page use?

It reads dataset records the agent has collected. Elastic uploaded data. Nothing has to be imported first — the page appears once there is something to show.

Do I have to build or configure it?

No. You describe what you need in a sentence and the agent chooses the view and fills it. Custom pages exist for the cases no built-in view covers, and those are reviewed before they go live.

When is a spreadsheet the wrong choice?

Records with long text — the grid fights you. It is best for: budgets, quantities, anything people would otherwise export to Excel.

Does it stay up to date?

Yes. The page is a view over the data rather than a copy of it, so anything the agent learns afterwards shows up without a rebuild or a re-import.