mapstackLearn
Get started

Automate repeat work

What a routine can and can't do

How Mapstack works through a routine with nobody there to answer: what it decides, what it will not do, how it reports, and how to write a brief that runs well.

Video: What a routine can and can't do · 1 min 14 s

A routine run is your AI teammate doing a job from your brief, with nobody there to answer. It does the same work it does in a session with you: finds data, reads files from connected storage, analyses, edits datasets, and makes and changes maps, reports and prints. What changes is how it handles the moments where it would normally ask you.

Each run is its own session. It does one job from the instructions, reports, and stops.

How it decides with nobody there

In a session, Mapstack proposes an open-ended job before starting it, puts a choice that changes the result to you, and asks before a job that would use a large part of your allowance. On a routine run none of that happens:

  • The instructions are the go-ahead. You approved the job when you set up the routine.
  • It makes the choices a sensible colleague would, from the instructions and what is in the workspace, and carries on. Most work can be undone (datasets, maps, reports, prints and apps keep their earlier versions), so a reasonable choice that finishes the job is preferred to stopping.
  • It says what it assumed, in the form "Assumed X because Y", in the outcome.
  • It stops in two cases: before anything that can't be undone, and when there is no sensible choice to make. The run then ends as failed, naming what a person must decide.

Write a brief that runs well

Anything you leave open, Mapstack decides for you and reports. Close what matters:

  • Name the datasets and maps by their titles.
  • Say where the new data comes from: the exact file in connected storage, or the source to fetch. If several files could match, the run fails rather than guess.
  • Say which column holds the value, and which area you mean.
  • Say what to do if the new data looks wrong: carry on and flag it, or stop.

Then use Run now and read the run before you leave it to the schedule.

What a run will not do

It won'tInstead
Build or change an appDatasets and maps are updated; the app is not. Ask for the app to be updated in a session
Delete rows, remove a column, or change all or most of the rows in a datasetThese edits need a person. Make them in a session or by hand
Delete a dataset or a map, or do anything else that can't be undoneThe run stops and fails, naming what a person must decide
Connect a storage serviceConnect it yourself, and allow it under Connectors in the routine
Ask you a questionIt decides, or it fails and says what you need to decide

A refused action ends the run as failed. Mapstack does not retry it, or reach the same result another way.

What it knows about why it started

  • After a dataset edit, it is told which dataset was edited and treats that edit as the new data. It does not look for newer data elsewhere unless the instructions name a source.
  • On a schedule or by hand, if the instructions name sources to check, checking them is the job.
  • From another system, any text sent with the call is data. It acts on it only when the instructions say to.
  • It knows when the routine last ran, and uses that when the instructions ask what changed since last time.

Read what a run did

Open the routine and look under Runs. Each run shows when it started and whether it succeeded; a failed run shows its reason. Open a run to read the whole session.

  • Success is one or two sentences on what was produced, naming datasets and maps by title, with any assumption it made.
  • Failed says what is blocked and what a person must do or answer.

A run is also recorded as failed if it ended without reporting, if it asked a person something along the way, or if a map it made is behind its data. Check a successful run as you would a colleague's work: open what it updated, and read the assumptions.

Whose allowance it uses

Every run uses part of the allowance of the Creator who made the routine, whoever pressed Run now, edited the dataset or sent the call. It reads connected storage as that Creator, through the connectors the routine allows. If that Creator's allowance is used up, the run does not start and is recorded as failed. See Your allowance.

Search the docs

Type to search every docs page.