Skip to content
Help Center Back to AgenticObjects

Reading a brief

A brief is an agent's written report — assumptions first, then the story, then the findings, each one traceable to its evidence and its run

App 5 min read Updated

A brief is the report a scheduled agent writes after it studies your data: what it found, told in a few paragraphs, with every finding one click away from its evidence. It is designed to be read from the top down, because the assumptions come before the numbers.

What this screen is for

A brief turns a run into something you can read in two minutes and quote with confidence. The agent asks its own questions of your data, turns the answers into findings — one card per thing worth knowing — and writes a short narrative that ties them together.

The page puts the assumptions first on purpose. Before you read a single figure, you know which period it covers, whose rows it was computed on, how far it can be trusted and how current the data is. A number read without those facts is easy to misquote; a number read with them is easy to defend.

What you see

  1. The header. A Back to the archive link, the brief's title, a Brief badge, the agent that wrote it and when it was produced.
  2. The declarations strip — Declarations — the assumptions behind the numbers below. It always stays open. Windows read has a line of its own: the period or periods the numbers cover. Beneath it, one quiet row holds Scope (the row-level scope the brief was produced under), Provenance tier and Freshness.
  3. The narrative. A few paragraphs in the agent's words. Any figure the engine could not verify is marked with † and explained at the foot of the text. The Evidence button under it shows the narrative's figures and its run's query in a drawer from the right, as in the chat; your technical-review request and the reviewer's reply live here too (the reply appears under the button).
  4. Findings covered. The heading counts them, for example Findings covered (4). Each card shows:
    • a severity badge — Low, Medium, High or Critical;
    • the trust tier — Governed or Derived;
    • a trend word, such as up — better or down — worse;
    • the finding's title, which links to its full evidence;
    • a Measured line naming the period the finding covers;
    • a chart, where the finding's data has a clear shape, with The numbers underneath;
    • Finding detail, which expands to the agent's full reasoning;
    • an Evidence button that opens that finding's figures and query in its own drawer.
  5. The run link. At the foot, The run that produced this brief, with how the run was started and when.

How to use a brief

Open a brief

  1. In the App, open Briefs from the navigation, or pick the newest brief on your Dashboard.
  2. Select a brief to open it.
  3. Read the declarations strip first, then the narrative.

Check the evidence behind a finding

  1. Press Evidence on the finding's card, or press a verified figure in its body and choose More → on the card. The evidence opens in a drawer without leaving the brief; Esc closes it.
  2. For the finding's own page, select its title; you can leave feedback there too. See Finding detail.
  3. Use your browser's back button to return to the brief.

Trace the brief to its run

  1. Scroll to the foot of the brief.
  2. Select The run that produced this brief.
  3. The run's record opens in Briefs & runs. See Briefs & runs.

How to read it

  • Assumptions come first. If a number surprises you, check Windows read and Scope before anything else — most surprises are a different period or a different slice of rows.
  • Every declaration is spelled out. When the record holds no value for a line, the strip says so in words, so an empty space never passes for a real value.
  • The tier is the engine's call. Hover the tier badge: the engine stamps it from the evidence, and the model can never write it. See Trust tiers.
  • Trend words follow a fixed rule. Where a metric declares which direction is good, you read up — better, up — worse, down — better or down — worse. Otherwise the brief says up or down and passes no judgment.
  • Charts follow fixed rules. A series over time becomes a line; two to twelve categories become columns; more than twelve show the ten largest, with a note on how many more there are; a comparison becomes a card with This period, Compared with and Difference; a share becomes a horizontal bar. Every chart is drawn from the finding's own figures, all listed under The numbers.
  • Interpretive language is labelled. A finding marked Interpretive contains a judgment in the agent's own words. The engine labels it so you can read it as interpretation rather than measurement.

Good to know

  • A brief carries only governed and derived figures. Exploratory figures stay in chat, where you can see their badge and weigh them yourself.
  • Your access is respected inside a brief. Findings outside your access are left out, and the brief states how many. A finding with no page of its own for you can still be read in full inside the brief.
  • A brief is a fixed record. Its declarations and figures are frozen when it is produced, so the brief you read today reads the same next quarter.
  • Every brief traces back to one run. The run link at the foot takes you to the exact run that wrote it.
  • Recommendations live on your Dashboard. Each recommendation names the findings it is based on, so you can move from action to evidence in one step. See Home.