Reports, AI Log, and Traceability

Home · Structured Outputs, Evidence, and Freshness · Portfolio and Technical Administration

Objective

This page explains the difference between artifact, artifact version, and PM Doc, then details the role of the AI Log for audit and support.

Three Concepts to Distinguish

Term	Role
Artifact	Governed object resulting from a run or review work
Artifact Version	Precise state of an artifact, with history, diff, and lineage
PM Doc	Reviewed, edited, downloaded, published, or reinjected project document into the Knowledge

Observed Lifecycle

The most common flow is:

1. an agent produces a structured output;
2. depending on the flow and available rights, that output can then feed an artifact or a PM Document;
3. the user opens Reports & Artifacts when a governed document exists or must be prepared;
4. they review the content, consult the diff and lineage;
5. they approve, publish, download, or add the document to the Knowledge.

In other words, an agent answer does not automatically become a final document: the observed chain remains structured output → artifact / PM Doc → review → downstream decision.

Visible Statuses

Artifact

The confirmed statuses at the artifact/version level are:

• draft
• approved
• published
• superseded

PM Doc

PM Docs expose at least the following compatibility states:

• draft
• final

If Reports & Artifacts looks empty

A project can display an empty surface for several normal reasons:

• no artifact or PM Doc has been created yet;
• the run stayed at the structured output stage without downstream transformation;
• the wrong project or a filter hides the expected rows;
• the environment does not contain the same seeded demo data as other screenshots or tests.

In that case, start again from Agents or AI Log, then return to Reports & Artifacts when a governed object actually exists.

Actions Available in Reports & Artifacts

The hub exposes actions such as:

• review / preview;
• diff;
• lineage;
• edit;
• download;
• add to knowledge;
• publish;
• delete.

Diff, Lineage, and Versioning

Diff

The diff compares artifact versions. It is most useful for:

• reviewing differences before publication;
• distinguishing an agent's proposal from an actually validated version;
• understanding why a final document differs from a previous draft.

Lineage

The lineage connects:

• the source run ID;
• the context snapshot ID;
• the structured output ID;
• the artifact and its version;
• linked publications and evidence.

Versioning

The interface also allows selecting a version by number and identifying the previous version, the base for comparison, or a superseded version.

Approval and Publication

Observed behaviors show that:

• Approve validates an artifact version still in draft;
• Publish corresponds to governed distribution to a target destination or output format when the flow allows it;
• publication sends governed formats and destinations;
• Add to knowledge reinjects a reviewed document into project search;
• Download remains separate from both publication and add to knowledge.

Available Editors

The product includes several editing surfaces:

• a rich / markdown editor for document-oriented outputs;
• a spreadsheet-like grid editor for table-oriented outputs;
• a review / source read-only mode if the format is not editable in the same way.

Confirmed downloads: DOCX and XLSX.

Practical Limits to Keep in Mind

Case	Useful Reading
actually editable content	use the editor for review and controlled adjustments before approval
content mainly in review mode	treat the screen as a validation surface, not as the primary place for rewriting
need for a reference rendering to share	the governed download remains the most reliable output to share outside the application

AI Log: What the Page Covers

The AI Log has two distinct tabs:

• Runs;
• Activity.

Runs vs Activity

Tab	What it's for
Runs	Review an agent execution, its traceability metadata, and its linked artifacts
Activity	Review a project event timeline and inspect the raw payload of an event

When to Open Runs

Open Runs to:

• understand why a result was produced;
• find the provider actually used;
• review freshness, confidence, and citations;
• link a run to an artifact or a PM Doc.

When to Open Activity

Open Activity to:

• reconstruct a timeline;
• verify that a draft, approval, or publication has indeed left an event;
• inspect the raw payload when doing support or audit.

Examples of Useful Events to Confirm in Activity

Depending on the flow, Activity can help you confirm that a downstream event was properly recorded, for example:

• creation of a draft;
• approval of an artifact;
• publication;
• preparation of a notification;
• decision for approval, rejection, or governed execution.

Concrete Examples of Type / Kind

The values below are examples observed in seeded flows and demonstrations. They provide a good reading reference, without claiming to cover all future events.

Type	Kind	Practical Reading
agent_execution	completed / failed	an agent run ended successfully or failed
document_upload	started / completed / failed	a document started ingestion, completed it, or failed
report_generated	completed	a PM Doc or report was generated and logged
marketplace_update	completed	a notification or system update was recorded in the history

Visible Filters and Columns

Runs

Observed filters:

• search on agent / status / run ID;
• agent filter;
• status filter.

Visible columns: Created, Agent, Status, Cost, Run ID, Actions.

The Cost field exposes at least tokens and the number of calls when this information is available.

How to read Trace ID, Cost, Confidence, and Source freshness

Keep the following interpretation simple:

• Run ID is the most practical handle for finding a visible row again inside Runs;
• Trace ID is mainly for cross-service diagnostics, support, and audit correlation;
• Cost (tokens / calls) shows runtime usage for transparency and investigation, not a business approval signal by itself;
• Confidence is a review cue, not proof on its own;
• Source freshness tells you how current the cited evidence still is over time.

If you need a portfolio-level signal about budget or usage pressure, the relevant indicator is cost_pressure in Portfolio, not the raw Cost card of an individual run.

Activity

Observed filters:

• search on type / kind / actor;
• type filter;
• kind filter.

Visible columns: Created, Type, Kind, Actor, ID, Actions.

Technical Details Visible in a Run

The details of a run can expose:

• Trace ID;
• Context snapshot ID;
• Structured output ID;
• citations;
• Effective AI Provider;
• Deployment-selected AI Provider;
• Model family;
• Confidence;
• Source freshness;
• linked artifacts.

AI Runtime Transparency

The following distinction is important:

• AI provider selected at deployment: initial environment choice;
• effective AI provider: provider actually used for this run.

These two values can differ. In case of doubt, the AI Log is authoritative for the observed run.

Simple reading of Effective AI Provider

Effective AI Provider answers the question: “Which provider actually served this exact run?”

For investigation, this value is more reliable than an assumption based only on the administrative configuration, because the observed runtime value can differ from the provider selected at deployment.

Surface to Open First Based on Symptom

Symptom	Surface to Open First	Why
content divergence between draft and final version	Diff	to see what actually changed
doubt about evidence or provenance	Lineage	to link run, structured output, artifact, and sources
doubt about runtime, provider, or AI execution	Runs	to review Trace ID, effective provider, freshness, and confidence
doubt about publication, approval, or downstream flow	Activity	to confirm the timeline and raw payload

Recommended Investigation Workflow

When a deliverable, publication, or notification seems questionable, start from the most visible to the most technical:

1. open the concerned PM Doc or artifact;
2. consult the diff to see what actually changed;
3. open the lineage to note source run ID, Context snapshot ID, and Structured output ID;
4. switch to Runs to review freshness, confidence, citations, effective AI provider, and linked artifacts;
5. use Activity to confirm the downstream flow: draft, approval, publication, notification, or governed action;
6. finally, keep the Trace ID if the investigation needs to be resumed by support or audit.

Case 1 — A final PM Doc seems incorrect

1. open the final PM Doc and confirm the version actually published;
2. use Diff to identify what was added, removed, or rephrased;
3. open Lineage to find the source run ID and Structured output ID;
4. in Runs, check the evidence, freshness, and effective provider;
5. if the final content does not reflect the expected flow, finish in Activity to verify the approval, publication, and the actor who pushed the version.

Case 2 — A prepared action or notification did not complete

1. start from the visible draft, action queue, or notification;
2. note relatedArtifactId or relatedNotificationId when they exist;
3. use Activity to confirm if an approval, rejection, or execution was logged;
4. if the flow remains unclear, return to Runs to check the source run and the logic that produced the recommendation;
5. finally document the Trace ID and related identifiers before escalating.

Which IDs to Prioritize for Retention

Incident Type	IDs to Prioritize for Retention
questionable run result	Trace ID, Context snapshot ID, Structured output ID
suspicious artifact or final PM Doc	source run ID, Structured output ID, concerned artifact version
action or notification with unclear follow-up	Trace ID, relatedArtifactId, relatedNotificationId
cross-support investigation	all IDs above, plus approximate time and concerned project

Traceability Best Practices

• keep the Trace ID when a result is important;
• before publishing, verify the lineage between the deliverable and the evidence;
• in case of discrepancy between two versions, use the diff before arbitration;
• add to the Knowledge only deliverables that have actually been reviewed;
• use Activity to complete the investigation when Runs is not enough.

Next

• Structured Outputs, Evidence, and Freshness
• Portfolio and Technical Administration
• Maintenance, Support, and FAQ
• Glossary