Finding detail

Every finding, recommendation and brief your agents publish has a page of its own. It holds the statement itself, the evidence behind each figure in it, the items it is connected to, and the actions you can take from there. When you want to know whether a finding is worth acting on, this is where you check.

What this screen is for

A finding is only useful if you can see what it stands on. This page keeps the statement and its proof together: each figure the agent cited is listed with the metric it measures and the period it covers, and a computed figure shows how it was computed. You do not have to trust the sentence — you can read the record under it.

The page is also where a finding connects to the rest of your work. A recommendation shows the findings it rests on; a finding shows the recommendations built on it and the briefs that cover it. From here you can take the finding into a conversation, open the run that produced it, or tell the team whether it helped.

What you see

1. Back to the findings feed — returns you to the findings on Home.
2. The title and identity line — the type (Finding, Recommendation or Brief), the severity (Low, Medium, High or Critical), how new it is (New, Ongoing or Repeat), an Experiment badge when it came from an experiment run, an Interpretive badge when the text carries interpretive language, then the agent, the trust tier and the time it was produced.
3. The body — the statement as the agent wrote it. Figures the engine wrote from the evidence are drawn with a dotted underline; hover over one to see its source. A figure the model stated without verification carries a † and the legend † model statement — not verified by the engine.
4. Evidence — a button under the body; as in the chat it opens a drawer from the right (from the bottom on a phone). Inside is one row per cited figure: the figure, the metric, and the period and slice it was read from. Rows with a computation open a derivation panel showing the operands, the operation (This figure came from:) and the engine's own spelling: of it. If you hold that permission, Show the query opens the executed statement in the same drawer; if not, Request Technical Review asks a reviewer to check it, and the reply appears under the Evidence button. Esc closes the drawer. At the top of the drawer, when it applies, Ground says the frame the statement was read over: for example that a "highest" was read over a set of a given size, that a share was taken over a frozen breakdown, or that a list was cut at a row limit.
5. Held values — a collapsed section, visible to people who can read the agent's runs, listing any number that could not be tied to evidence, with its reason.
6. Linked items — groups such as Rests on 2 findings, Covers 4 findings, 1 recommendation rests on this and Covered by 1 brief, each row a link.
7. Actions — Deepen in chat, Evidence: run detail, the 👍 and 👎 feedback buttons with their counts, and Remove from my feed.

How to …

Check a figure before you quote it

1. Press a figure the engine verified: a small one-sentence card says what it is — the metric, the slice, the period and its dates, and how far the data is complete. More → on the card opens the drawer on that figure's row. A figure the engine computed from other figures says the calculation instead: each figure it was computed from with its value, the operation, the period, and the engine's own result — for example "63.0% = Patient revenue (248,660,570) ÷ Revenue excl. VAT (394,608,578) · year to date · recomputed by the engine: 63.01%". A figure the record does not hold is named without a value.
2. Read the metric and the period beside it — that is exactly what was measured.
3. If the row offers derivation, open it to see the operands and the operation the engine re-ran.
4. For the full step-by-step record, press Evidence: run detail.

Take a finding into a conversation

1. Press Deepen in chat.
2. A conversation with the same agent opens, with your question box already started from the finding's title.
3. Finish the question and send it. Nothing is sent until you do.
4. Pressing Deepen in chat again later brings you back to that same conversation.

Give feedback

1. Press 👍 if the finding was useful to you, 👎 if it was not.
2. The counts update and your own choice is highlighted. You can change your vote at any time; only your latest vote counts.

Clear it from your own feed

1. Press Remove from my feed.
2. You return to the findings feed, without this item. Your colleagues' feeds are untouched.

How to read it

• Severity and novelty are separate claims. Severity says how much the finding matters; New, Ongoing and Repeat say whether you have seen it before.
• A dotted figure was written by the engine. It was taken from the evidence, not typed by the model, and its source is one hover away.
• A † figure is a model statement. Treat it as context, not as a verified number.
• "operand missing" is stated, not hidden. When a computation's second operand is not addressed in the record, the row says so and shows only what was recorded.
• Linked items show the chain one level up. A recommendation always rests on at least one finding; that is how you trace an action back to its observation.

Good to know

• Every label in Evidence is read from the record frozen at the time of the run, so renaming a metric later never changes what this finding meant.
• A number that cannot be tied to evidence never enters the body. It waits under Held values with its reason and is kept out of export, distribution, memory and calculation.
• Every linked item is checked against your own access before it is listed, so the page never shows you more than you are entitled to see.
• Experiment output keeps its Experiment badge permanently. Experiment runs record what they produce but never deliver it.

Related

• Home — the feed that leads to this page
• Trust tiers — how to weigh the tier and the † mark
• Chat — where Deepen in chat takes you
• Reading a brief — the delivered document that covers findings
• Runs and Activity — the run record behind Evidence: run detail