How reports work

One report layout for every agent: six tabs, a result status, notices and blocks that always link back to evidence.

Every AGENTEX report, from any agent, renders in the same shell. You learn to read one report and you can read them all. This page explains the parts of that shell and the rules every report follows.

What a report is#

A report is the saved output of one run. It is immutable: once the run finishes, the report, its evidence and its status never change. It records:

the agent, its version and the hash of that version's manifest;
your inputs and the network researched;
the block or slot the reads were pinned to;
the findings, grouped into sections;
every read the run made (the evidence);
the limitations of this agent and this run.

The six tabs#

Tab	Contents
Overview	The archetype badge, result notices, the headline, a summary and the key findings.
Findings	The report blocks: sections, tables, charts, timelines and flows, each linked to its evidence.
Evidence	Every read the run made, with what it proves, the observation, its source and its block or slot. See Evidence and provenance.
Sources	The providers and methods behind the evidence. See Sources and execution.
Execution	Your inputs, the steps the agent ran, calls used against the cap, and timing.
Export	Downloads (JSON, CSV, Markdown, readable HTML) and share links. See Export and sharing.

Agent pages show a saved example through the same blocks, so what you see before you buy is what you get after.

Result status#

Status	Meaning	Charged with the default policy
Complete	Every required finding was observed and backed by evidence captured in this run.	Yes
Partial	At least one required finding could not be established. Each one states why.	No
Failed	The run could not produce a report.	No

See Partial and unavailable data for how a report explains what it could not see.

Notices#

Notices appear above the findings when something affects how you should read them:

the result was partial or failed;
bounded coverage: fewer blocks were read than you asked for (for example because a log limit was reached);
conflicting sources: two sources disagreed about the same fact;
datasets that were unavailable.

Blocks#

Findings are shown as blocks chosen by the agent's archetype and its declared presentation:

Block	Shows
Metrics	A grid of key findings.
Section	One report section, with a status per finding.
Table	A dataset as rows; it becomes cards on a phone.
Bars	Values drawn to scale, with the exact value beside each bar.
Timeline	Rows ordered by block or time.
Flow	Movements from one address to another, with amounts.
Graph	The subject and related addresses (roles, spenders, signers).
Signals	The tools' own observations: risk, conflict or information.
Coverage	The block range actually read against the range requested.
Limitations	What the report cannot establish.

Every block that shows data links to the evidence of the step that produced it. Sections the layout does not place are collapsed under All findings in this report, so no finding is ever hidden.

Formatting rules#

Formatting never changes a value:

addresses are shortened and linked, with the full address a click away;
times are shown in UTC;
amounts keep their exact value; small values show four significant digits, and the exact figure is in the evidence;
untrusted text (token names, metadata, revert reasons) is shown as data. Control and invisible characters are removed, and a notice says when that happened.

Reports are records, not advice#

A report states what public data showed at the block or slot it was read, from the sources listed. It is not investment advice, a rating, an audit or a safety guarantee.