Work answer feedback
Read the Not satisfied comments people leave under answers, mark them In review, keep a team note, and turn each into a convention or correction card.
Under every answer people can click Satisfied or Not satisfied. People who ask from Manas give the same feedback there: by saying so in Manas web chat or Slack, or with the 👍 / 👎 buttons in Slack and Manas web chat (see Use Tattva from Manas). A Not satisfied comment comes to your team in Ontology › Maintain › Feedback. This page explains how you are told about a comment, how to work it, and what the person who left it sees.
What people leave
Satisfied is saved with the answer. Nothing else happens, and it does not appear in the list. The top of the Feedback page counts how many answers were marked Satisfied.
Not satisfied opens a dialog with six quick reasons, and the person can pick any of them:
- Figure looks wrong
- Definition is not what we use
- Wrong period
- Breakdown missing
- Wording unclear
- Other
They can also say what was wrong and what is correct. They must pick a reason or say what was wrong. The comment is saved with the answer, so you see the question, an excerpt of the answer, who left it and when. Open the answer and its Trail opens the conversation, where the Trail shows how the answer was worked out.
A comment changes no answer, not even its author's. The fix you make in the ontology is what changes answers, for everyone.
A question answered with a registered dashboard's link records which dashboard it matched on the run (phase_outputs.dashboard_match). See Dashboards for repeated questions.
Where it came from
Each comment shows where it was left:
| Chip | Left |
|---|---|
| From Tattva | With the buttons under an answer in Tattva |
| From Manas web | With the 👍 / 👎 buttons in Manas web chat |
| From Slack | In a Slack channel by telling Manas, or with Slack's 👍 / 👎 under an answer |
| From Manas | By telling Manas in web chat, a Slack DM or /manas |
From Manas and Slack there is no snapshot of the answer as the person saw it, so the answer excerpt is taken from the stored answer. 👎 reasons are mapped to the six above: Slack's Wrong or made up and Manas web's Inaccurate or incorrect answer → Figure looks wrong, Slack's Incomplete → Breakdown missing, Slack's Hard to read and Manas web's Poor style or tone → Wording unclear, the rest → Other. What was wrong shows what the person typed, then the reasons in their own wording, for example "June is missing — Picked in Slack: Wrong or made up, Too slow".
A 👍 / 👎 is saved under the person when their email belongs to someone who can use the project. Otherwise it is saved under whoever asked the question (for a question asked in a channel, the channel), and the card names the person next to it, for example "Sales channel (ravi@kapiva.in in Slack)". A comment left by telling Manas in a channel is the channel's. Each 👍 / 👎 is one comment: a comment added later, other reasons or a changed rating update it and put it back to New; Manas re-sending it unchanged leaves it, and your triage, alone.
How your team is told
Each new Not satisfied comment sends a bell notice, "New feedback: Not satisfied", to every owner, admin and ontology admin of the project. Each person gets their own notice, so one person reading it does not mark it read for the others. The person who left the comment is not notified. A workspace admin who is a plain member of this project is not notified either; the project role counts first. Clicking the notice opens the comment. A 👎 from Slack or Manas web chat is announced when it arrives and again when its first typed comment arrives; reasons picked afterwards update it without another notice.
To post each comment to a Slack channel too, a project admin adds a Slack incoming webhook under Settings › Project › Answer feedback. Only https://hooks.slack.com/… addresses are accepted. The Slack message shows who left it (and where, when it came from Manas or Slack), the reasons, the question, what was wrong and what is correct, and links to the comment. Leave the field empty to keep the bell only.
The list
Open Ontology › Maintain › Feedback. Filter by status, and with From anywhere by where it came from (the counts follow that choice):
| Status | Meaning |
|---|---|
| New | Nobody has picked it up yet |
| In review | Someone on the team is working on it |
| Fixed | Applied as a fix, or already covered by an existing rule |
| Not changed | Nothing changes; the person was told why |
Anyone in the project can open the list. Only owners, admins and ontology admins see the team's side and the buttons.
Asked but not answered
The Asked but not answered filter lists questions Tattva did not answer because of the answer scope, grouped by what was refused, with how many times each was asked and when it was last asked:
| Group | Why it was not answered |
|---|---|
| Not answered yet | The question was about a topic your team marked Not answered yet, or fell outside the Scope line |
| Question kind switched off | The kind of question (for example forecasts) is unticked in Settings › Project › Answer scope |
| Metric not defined yet | Figures come from metric definitions only, and the metric asked for has none |
| Before the metric's data starts | The period started before the metric's Available from date |
Click a group to see each question, its date and Open to open the conversation. These are records of refusals, not comments: there is nothing to decide. Use them to choose what to define or switch on next.
Asked often
The Asked often filter groups recent questions by what they asked, not by their wording: the metrics, the breakdown (by product, by month…), the kind of period (a month, a quarter, the last N days, a date range…) and the filters the person stated. "RTO rate last 15 days" and "What's the RTO rate for the past 15 days?" are one group. A filter Tattva guessed is not counted, and a filter on a date field counts as the period. Only groups asked at least twice are listed, most asked first. Pick Last 7, 30 or 90 days (30 by default).
Each group shows:
| Asked | How many times, by how many people, and when it was last asked |
| Model cost | What the answers to it cost in model calls, in total |
| Asked as | Up to five different wordings, newest first |
| Each question | Its date and Open to open the conversation |
The line at the top says how many of the period's questions asked something asked at least twice. These groups are the list of dashboards worth building. Owners, admins and ontology admins see Register a dashboard and Create in Looker Studio under each group: both open Make › Dashboards with the form filled in from the group (its metrics, breakdowns, filtered fields and period). See Dashboards for repeated questions.
Work a comment
- In review: marks the comment as picked up and records who did it and when. Back to New undoes it while no card has been made from it.
- Team note: a note for your team. The person who left the comment never sees it, and it is never put in their notification. Click Save note to save it.
- Create convention: opens a form prefilled from the comment, with a slug, a category and the rule (taken from what the person said is correct). Edit it and click Create convention card. The card goes to Review › Pending, linked to the comment, and the comment moves to In review. When the card is accepted in Review, the convention is live, the comment moves to Fixed ("Applied as convention …"), and the person is notified. If the card is rejected, the link is removed and the comment stays In review, so you can make another card or decide it.
- Create correction: opens the comment's reasons and text, prefilled, for you to edit into the fix to make. Click Send to Review. The comment goes to Review › Pending as a
correctioncard with your text under Team's correction, and moves to In review. Deciding that card in Review settles the comment. See Review queue. - Applied, Already covered, Not applied: decide the comment here directly, as on a correction card. Applied and Already covered need the type and reference of the rule; Not applied needs a reason, which the person sees.
- Delete: removes the comment outright, for duplicates and tests. The person keeps no record of it.
What the person who left it sees
Under the answer, a line shows where their comment stands, for example "You flagged this answer — In review by the team", "— Applied as convention “cancelled_orders_excluded”" or "— Not applied: …". They get a bell notice when it is decided. Review › My corrections lists all of their comments with the outcome. The team note is never shown to them.