Skip to main content

How an answer was made

Every answer Qry gives keeps a record of how it was made. Anyone who can open the conversation can read it: its owner, the members of its workspace, and administrators.

The row of actions​

Under each answer there is one row of actions: copy, 👍 / 👎, generate again, how this answer was made, and when it was given. On the last answer the row is always in view; on earlier ones it appears when you point at the answer or move to it with the keyboard (on touch screens it is always there). A thumb you have pressed stays visible.

The row of actions under an answer: copy, good answer, bad answer, how this answer was made, and when.

Generate again is offered on the last answer only. It asks the same question again from that point; earlier answers stay as they were.

What the record shows​

Click How this answer was made (the scroll icon).

The record of an answer: the model, where it looked, the query it ran and the skills at their versions.

SectionWhat it says
ModelThe model that answered, as it is named in the model picker, with its exact identifier beneath, and when.
Where it lookedThe datasource, catalog and schema the answer worked in — its own, even if the conversation's selection changed later.
Chosen in the questionTables or skills the person pinned with @ and /.
QueriesEach statement sent to a datasource, whether it ran or failed, and how many rows it returned.
Metric viewsDatabricks metric views the answer read.
SkillsEach skill that reached the answer, how it reached it — in every answer, read for this question, documents it could search or chosen by you — and the version in force when it answered.
Documents citedThe passages the answer quoted from a workspace's files or a knowledge skill.

Two limits:

  • A query on a datasource you cannot open is shown without its SQL, columns or result. You see that it ran, not what it read. A workspace can be shared by people with different grants.
  • Answers given before the record existed have none. The card says so.

Run again​

Run again under a query repeats that statement against today's data and compares the result with the first run: same result, how many rows are new and how many are gone, or — when the first result is no longer kept — how many rows there were then and now.

Run again: the same result as when it was answered, with today's rows.

  • The model is not involved. The statement is the one recorded, run through the same permission checks as any query, as you.
  • Only the person the answer was given to sees the button. The SQL was written by the model; Qry runs it only for the person who asked for it, never as someone else. Workspace members see the record without it.
  • Only statements that read. Qry checks the statement before running it and refuses anything that could change data: an answer that once wrote something is never repeated by a click.

Rating an answer​

👍 counts at once. 👎 counts at once too, and then asks — beside the button — what went wrong: the wrong data, a wrong figure, it misread the question, or it left something out, plus an optional note. All of it is optional.

After a thumbs down: four reasons and a note, all optional.

Ratings are one per person and answer: pressing the other thumb changes yours, pressing yours again takes it back. Administrators see them under Admin › System › Answer ratings, and any rated answer can become an answer check.

Adding an answer to a dashboard​

Add to dashboard uses the datasource, catalog and schema the query actually ran in, taken from the record — not whatever is selected in the chat at the time you click.