Skip to content
Pondara Pondara docs

Triage the bandeja

The Bandeja is your inbox. Every closed engagement deposits one or more AI signals there — one per sub-competency the dialogue exercised — and each row waits on your verdict before it can move the learner’s record. This guide walks you through reading the queue and clearing it.

Before you begin

  • You are signed in as a trainer.
  • Your cohort has run at least one engagement that produced AI signals. If the Bandeja is empty, that is the steady state — there is no work to do.

What’s in the Bandeja

Each row is one pending AI signal. The columns:

  • Learner — the person who ran the engagement.
  • Scenario — the situation they practised.
  • Dimension — which of the eight sub-competencies the AI observed.
  • AI proposal — the level the AI suggests, with a confidence band (Low / Medium / High).
  • Status — pending, or already in disagreement with a previous verdict on the same dimension. See the status glossary.
  • Age — how long the row has been waiting.

The triage rhythm

Pick whichever fits the day:

  • Oldest first. Sort by age and clear the top of the queue. This keeps the 7-day SLA safe and is the right default if you are catching up.
  • By learner. Group rows by learner and clear each person in one pass. This is the right default if you batch coaching by person — you read all their signals together and the picture is more coherent.
  • By scenario. Useful right after you assign a new scenario to the cohort. You see how the same situation reads across different learners.

Click any row to open the verdict drawer, where you render the call. See render a verdict for the three actions and the undo window.

The 7-day rule

A row pending more than 7 days escalates to your org admin’s overview. Nothing destructive happens — the verdict is still yours to render — but the admin sees the count and may follow up. The Bandeja highlights rows over the threshold so you can clear them first.

When something goes wrong

  • The Bandeja is empty but you expect rows. Check that recent engagements actually closed — interpretation-only or abandoned engagements do not produce AI signals.
  • Items appear out of order. Sort by age to force the oldest to the top; the default sort can vary if you switched it last session.
  • A row refuses to open. The underlying dialogue is being re-extracted by the AI; wait a minute and retry. If it persists, capture the row id and open a ticket.
  • A row shows “disagreement” before you’ve touched it. A previous verdict on the same dimension contradicts this new signal. Open it, read both, and decide. See disagree with the AI.