Reviewing Feedback

The Feedback page reports how the agents and workflows in your deployment are performing in the judgment of the people using them. End users rate agent responses as they chat, and Solace Agent Mesh rolls those ratings up for every agent and workflow, so you can see which ones draw complaints, whether a problem is growing or receding, and how much activity sits behind each number.

This page is written for administrators who own a deployment's agents and want to improve the experience those agents give. For the end-user side of feedback, that is, rating a response in a conversation, see Giving Feedback. For the operator configuration that turns collection on, see Collecting and Publishing User Feedback.

Opening the Feedback Page

On the navigation bar, select Signals & Usage.
Select Feedback.

The page opens at /feedback on the Dashboard tab. The Feedback History tab lists the individual ratings behind the numbers the Dashboard aggregates.

Everything the Dashboard reports covers only the agents and workflows your role reaches. Two administrators with different grants can read different totals for the same date range, so compare numbers between administrators only when you know both hold the same grants.

Access and Visibility

Access to the Feedback page comes from role-based access control (RBAC). A user needs at least one of the following scopes, or a wider grant that covers one of them, such as agent:*:* or agent:<id>:*.

Scope	Grants
`agent:*:read_feedback`	Feedback for every agent in the deployment
`agent:<id>:read_feedback`	Feedback for one agent
`workflow:*:read_feedback`	Feedback for every workflow in the deployment
`workflow:<id>:read_feedback`	Feedback for one workflow

Agent grants and workflow grants resolve independently. An agent wildcard reaches no workflows, and a workflow wildcard reaches no agents, so an administrator who reviews both needs a grant of each kind. A grant whose first segment is a bare asterisk, such as *, *:*:*, or *:*:read_feedback, reaches both kinds at once.

A grant that puts a wildcard inside the first or second segment, such as agent*:*:read_feedback, grants no feedback access at all. Agent Mesh drops that grant rather than treating the partial wildcard as a match. A user whose only feedback grant takes that form is denied the page, and a user who holds it alongside valid grants sees only what those other grants reach. Grant the exact forms listed in the preceding table.

Without a feedback read grant, the Feedback page is unavailable. The navigation bar does not list it, and a direct link to it is denied.

For more information about the scope grammar and how you assign scopes to a role, see RBAC Reference.

Narrowing What the Numbers Cover

The Dashboard reports on one slice of your deployment at a time, and the filter bar across the top of the page decides which slice. The cards, the trend chart, and the Agents & Workflows table all read the same window, so you never compare a number from one window against a number from another.

Date Range sets the window that the rest of this page calls the selected date range. It offers four preset windows: the last 24 hours, 7 days, 30 days, and 90 days. A Custom range option lets you name a start day and an end day instead. Both tabs open on the last 90 days. A custom range covers whole calendar days on your own clock, the end day included, and can span at most 366 days.

Move the range rather than trusting the default. A negative ratio that looks alarming over 90 days is often a single bad week that has already passed, and narrowing to the last 7 days tells you whether the complaints are still arriving.

Agent / Workflow narrows every number on the page to a single agent or workflow. The list holds the agents and workflows that handled activity in the current date range, ordered by name, up to the first 100. Narrowing the range can therefore drop one of them out of the list, so widen the range if the one you want is missing.

Reset Filters appears only while a filter is active, and it clears more than the subject: it also returns Date Range to its default window, and on Feedback History it clears Rating and User as well. Clear Filters, on the empty states, is the same action under a different label. If you reset to get back to every agent, set your window again afterwards.

Both tabs keep their filters in the page address. You can bookmark a narrowed view or send it to the person who owns the agent, and the browser's Back button steps back through the filters you set.

How Far Back a Range Can Reach

Retention caps the window. Agent Mesh deletes stored feedback and task history on a schedule, and a range that reaches past the retention period mixes surviving data with days that have been swept, so its numbers describe neither period honestly. The Date Range filter therefore offers only the presets retention can still answer for, does not accept a start day older than the retention period allows, and shortens a range that arrives from an older link or bookmark. A deployment with the retention sweep turned off has no such cap.

The two tabs use different retention periods. The Dashboard uses the shorter of the task and feedback retention periods, because every ratio on it needs both a count of ratings and a count of activities. Feedback History uses the feedback retention period alone, because it lists ratings and nothing else. For more information about both periods, see Automatic Data-Retention Sweep in Managing Backups and Data Retention.

The Two Filters on the Feedback History Tab

The Feedback History tab carries the same Agent / Workflow and Date Range filters, and a tab switch carries both selections with it. Its Agent / Workflow list, however, is built from the ratings it has already loaded rather than from a full list of your agents, so the surest way to narrow it to one agent is to arrive from that agent's row on the Dashboard. For more information, see Reading the Ratings Behind a Row.

Feedback History adds two filters of its own.

Rating limits the list to Positive or Negative ratings, and opens on Any Rating. Setting it to Negative is the shortest path from a bad row on the Dashboard to what people wrote. A tab switch does not carry Rating the way it carries the subject and the range. Feedback History restores your last rating selection when you come back to it, until you reload the page.

User narrows the list to the feedback one person submitted, and opens on All Users, which is also how you clear it again without resetting every other filter. It finds that person by searching your organization's directory rather than by searching the ratings on screen, so you can name someone whose ratings are not on the page in front of you. The directory search runs through the profile provider, which connects Agent Mesh to a toolset that looks up user profiles, and it requires that toolset to include a search_users tool. When your deployment has no profile provider, or the provider's toolset does not include search_users, Feedback History does not show User. If you expect the filter and it does not appear, ask your administrator to configure a profile provider whose toolset includes search_users. For more information about configuring a profile provider, see The Platform Block.

Reading the Headline Numbers

Three cards across the top of the Dashboard summarize the selected date range across every agent and workflow your role reaches, unless you have narrowed the Dashboard to one of them. The cards follow the same subject filter as the chart and the table, so a Dashboard narrowed to a single agent reports that agent's numbers here.

Card	Reports
Total Activities	Every agent or workflow run that started in the range, whatever its outcome. When an agent runs as a step inside a workflow, the run counts toward the workflow, not the agent. This count is not positive plus negative feedback. Most runs are never rated, so this number is far larger than the two rated counts beside it.
Positive Feedback	The number of runs rated positive.
Negative Feedback	The number of runs rated negative, with the negative ratio for the range beside it in parentheses.

The negative ratio divides negative feedback by Total Activities, not by the number of rated activities. A ratio of 0.400% against 50,000 activities means 200 negative ratings out of everything that ran, not out of the handful of runs anyone rated. Reading the ratio as a share of rated activity overstates the problem by a wide margin.

When nothing at all happened in the range, all three cards show a dash and the text No data yet. The cards fall back to that state both when no data arrives and when every count is zero, because the two mean the same thing to a reviewer. A range that carries real activity and no negative ratings keeps its numbers, so you can tell an idle deployment apart from a busy one that nobody complained about.

Read the cards as scale, not as a verdict. Ratings are voluntary, so they tell you where to look rather than how good an agent is. The counts tell you how much evidence stands behind the ratio, because the same ratio means something different over 40 activities than it does over 40,000.

Reading the Negative Trend Chart

The Negative Trend (Top 5) chart plots the five agents or workflows with the highest negative ratio in the selected date range, one line each, so you can see whether a problem is new, steady, or receding.

The five lines are ranked by negative ratio rather than by volume. The chart therefore does not necessarily show your busiest agents, and a low-traffic agent with a handful of negative ratings can outrank a heavily used one.

Two controls change what the chart plots:

Ratio plots each agent's negative ratio over time. Ratio is the default, and it is what compares agents against each other regardless of how much traffic each one handles.
Count plots the absolute number of negative ratings over time. Use it to judge how many negative ratings a rising ratio actually represents.

The spacing of the points follows the range you selected. A range of two days or less plots one point per hour, a range up to 31 days plots one point per day, and anything longer plots one point per week. The default range of 90 days therefore plots weekly points.

Hover over a line to read that point's ratio together with the counts behind it, in the form "7 negative of 1,412 activities". Select a line, or its entry in the legend, to narrow the whole Dashboard to that agent or workflow. Selecting a line sets the Dashboard's Agent / Workflow filter. To return to the full view, select All Agents & Workflows in that filter, or click Reset Filters.

After the Dashboard is narrowed to a single agent or workflow, the chart takes that subject's name as its title and the lines are no longer selectable, because nothing is left to narrow to.

A line that climbs across the range means a growing share of that agent's activity is being rated negative. Before you act on a climbing line, select Count. A ratio can rise because negative ratings rose, or because total activity fell while the complaints stayed where they were, and only the count tells the two apart.

An agent or workflow that ran in the range without drawing any ratings appears as a flat line at zero, which is a real reading rather than an empty chart. When nothing ran in the range at all, the chart reports that no feedback has been recorded yet and that it populates after end users start rating agent responses. When the Dashboard is narrowed to a single agent that had no activity in the range, the chart reports that no feedback was recorded for that agent in the selected window.

Finding the Agents and Workflows That Draw the Most Negative Feedback

The Agents & Workflows table answers the question the rest of the Dashboard raises: which agents are performing worst, and on what evidence. It lists every agent and workflow with activity in the selected date range, including the ones nobody rated.

Column	Reports
Agent / Workflow	The display name of the agent or workflow. An agent with no display name shows its identifier instead.
Type	Whether the row is an agent or a workflow.
Total Activities	Runs that started in the range, whatever their outcome.
Positive Feedback	Runs rated positive.
Negative Feedback	Runs rated negative.
Negative Ratio (%)	Negative feedback divided by total activities, on the same denominator as the headline card.
Trend	A small line chart of that row's negative ratio across the range.

The table opens sorted by Negative Ratio (%), highest first, so the worst-rated rows are already at the top. Select any other heading to sort by that column. The Trend column has no sort control, because a line has no single value to order by. The table shows 20 rows at a time, and the summary under it reports how many rows the range holds in total.

When nothing ran in the range, the Dashboard replaces the table with No Agents & Workflows Found and notes that agents and workflows appear there after they handle activity in the selected window. When a filter excludes every row, it shows No Agents & Workflows Match Your Filters instead, where Clear Filters restores the full list.

Reading a Row

Take the rows in the order the table gives you, and read across each one:

Read Negative Ratio (%) first. It is the column that compares agents to each other regardless of traffic.
Read Total Activities on the same row. A high ratio over a dozen activities is a lead worth checking. The same ratio over tens of thousands is an incident.
Read Negative Feedback for the number of runs that drew a negative rating. That count, not the ratio, is the size of the evidence.
Read Trend to see which direction the row is moving.
Compare the row's ratio against the negative ratio on the Negative Feedback card, with no single subject selected. A row close to that figure is ordinary. A row several times higher is the outlier you came for. With the Dashboard narrowed to one agent, the card reports that same agent, so the comparison has nothing left to say.

A row that is high on ratio, high on activities, and rising on trend is the one to act on first. A row that is high on ratio with almost no activities usually needs more data before it means anything.

Reading the Trend Column

Each row's trend line compares the second half of the date range against the first half. Agent Mesh takes the mean negative ratio of each half and reports the direction of the difference:

A rising line means the negative rate grew. Rising is the unwelcome direction here, because the line tracks negative feedback.
A falling line means the negative rate shrank.
A level line means the two halves differ by less than 10 percent of the first half, which is ordinary movement rather than a change worth chasing.

An agent with no negative feedback in the first half of the range shows as rising on any negative feedback at all, however small. The 10 percent band does not apply to it, because there is no first-half figure to measure the change against. Agent Mesh reports that case deliberately, because a problem that has just started is the one most worth knowing about.

Color alone does not carry the direction. Hover over any trend line to read it in words, as "Negative rate rising", "Negative rate falling", or "Negative rate steady". A screen reader announces the same wording. A row with only one point of data in the range draws no line, because a single point has no direction.

Use the trend to sort a long table into work. A high ratio that is falling is a problem already being solved, and a middling ratio that is rising is the one that becomes next month's complaint.

Reading the Ratings Behind a Row

The Dashboard tells you which agent to worry about. It does not tell you what anyone said. Select a row in the Agents & Workflows table to open the Feedback History tab narrowed to that agent or workflow, with your date range intact.

Feedback History lists one row for each rating, newest first, 20 rows to a page.

Column	Reports
Timestamp	When the rater submitted the rating. Hover over it to read the year as well.
Agent / Workflow	The agent or workflow that produced the rated response.
User	The person who submitted the rating.
Rating	Positive or negative.
Comment	What the rater wrote, or a dash where they wrote nothing. Feedback History cuts off a comment too long for the column and shows it in full in the detail panel.

The Dashboard lists every agent and workflow that handled activity, including the ones nobody rated, so a row you select can open an empty list. An empty list is an answer rather than a fault: nobody rated that agent in that window. Feedback History reports No Feedback Match Your Filters in that case, where Clear Filters restores the full list. An unfiltered tab with no feedback at all reports No Feedback Found instead, and notes that feedback appears there after users rate agent or workflow responses in chat.

Reading One Rating in Full

Select a row to open the detail panel beside the list. It holds everything recorded for that one rating:

Feedback Details carries the rating itself, under a heading that names the agent or workflow.
User and Time name who submitted the rating and when.
Entrypoint names where the rating came from: the Web UI, Slack, Microsoft Teams, or WhatsApp.
Comment carries what the rater wrote, in full. The panel shows the first six lines of a long comment, and Read More opens the rest.
Ask Activity Monitor opens a chat with the Activity Monitor agent, which answers questions about the rated conversation. For more information, see Reaching the Conversation Behind a Rating.
Recent Feedback (7 Days) lists up to five other ratings for the same agent or workflow, no matter how you have filtered or paged the table. Select one to read it in the panel. The list reports when there are none.

The Entrypoint value is worth reading on a run of complaints. When a single agent draws them all through one entrypoint and none through the others, examine that integration before you examine the agent.

Recent Feedback answers the question a single complaint always raises: whether you are reading one person's bad day or a pattern. Under that list, Filter by Agent or Filter by Workflow narrows the table to the subject you are reading, so the rest of its history is on screen behind the panel. Agent Mesh hides the action when the table already filters to that subject.

A rating you open from Recent Feedback can fall outside the table's own filters. The panel says so when it does, and the list highlights no row, because your filters leave that rating no page to sit on.

The open rating is part of the page address, so a reload or a Back navigation reopens the same panel, and a link you send to the person who owns the agent opens on the rating you were reading.

Reaching the Conversation Behind a Rating

A rating tells you how someone judged a response. It does not tell you what the agent did to produce it. Ask Activity Monitor, in the Feedback Details section of the detail panel, carries the rating over to the Activity Monitor, a built-in agent that answers questions about tasks that have already run.

Selecting it opens the chat page with the Activity Monitor already selected and a question already written for you. The question states whether the rating was positive or negative, names the session and the task that produced the rated response, and quotes the rater's comment when they left one. Send the question as it stands or edit it first, and then ask follow-up questions the way you would with any other agent. Each rating you open this way starts its own chat, so two ratings you examine in quick succession do not collapse into one. For more information about what the Activity Monitor can answer, see Chatting with the Activity Monitor (Experimental).

The action appears only on a rating that has a task recorded against it, and even then the task itself can be gone. Stored feedback and task history expire under separate retention periods, and deleting a task leaves the ratings given on it in place, so a rating can outlive the rated conversation. The action stays available on such a rating, and the Activity Monitor reports the task as not found after you ask. That answer does not prove the conversation never happened, because the Activity Monitor gives the same answer for a task you cannot reach as for one that never existed. Expect it on the oldest ratings in a wide date range. For more information about the retention periods, see Automatic Data-Retention Sweep in Managing Backups and Data Retention.

When the Action Is Unavailable

Ask Activity Monitor appears but is unavailable unless both of the following conditions are true:

The Activity Monitor is available to your account. It is on by default, an administrator can turn it off for the deployment, and your role needs the same permission to invoke it that any other agent requires.
The rating is one you submitted yourself, or your role holds the activity:*:read scope. Most of the ratings an administrator reviews were submitted by other people, so reviewing those this way means holding that scope. For more information, see Access and Visibility in Chatting with the Activity Monitor (Experimental).

When one of those conditions is not met, the line under the action names what is missing: "Needs the Activity Monitor agent, and either your own conversation or tenant-wide activity access."

Agent Mesh treats a rating you submitted as a conversation of your own, which it almost always is. A rating you left on someone else's conversation is the exception: the action is available, and the Activity Monitor then reports the task as not found, because the task itself answers only to its owner.

Asking About the Current Dashboard View

Ask AI, at the right of the Dashboard filter bar, carries the view in front of you over to the Activity Monitor, the built-in agent that answers questions about tasks that have already run. It opens the chat page with a question already written, naming the selected date range and the agent or workflow you have narrowed to, so you can ask about that slice without describing it again. Send the question as it stands or edit it first. The action appears only when the Activity Monitor is available to your account, and Agent Mesh withholds it rather than offering the question to a different agent.

Ask AI and Ask Activity Monitor cover different ground. Ask AI asks about a whole filtered view: every agent and workflow in the window, or the single one you selected. Ask Activity Monitor, in the detail panel, asks about one rating and the conversation that produced it. For more information, see Reaching the Conversation Behind a Rating.

Exporting the Dashboard as a Report

Export, on the Dashboard filter bar, downloads the current view as a single self-contained HTML file. The file opens in any browser and needs no connection to Agent Mesh, so you can attach it to a ticket, send it to the team that owns the agent, or keep it as a record of the window before you act on it.

The report follows the date range and the Agent / Workflow filter you have set, and nothing else about the screen. Sorting the table or paging through it does not change what you get.

Inside, the report carries the three headline totals, a negative trend chart, and one row for each agent and workflow, ordered by negative ratio, highest first. The chart plots the same five highest ratios the Dashboard draws, and says so whenever the window holds more subjects than it can plot. The table carries up to 100 subjects, and says so when the window holds more. A heading names the window and the time Agent Mesh generated the report, in UTC. On a deployment where the retention sweep runs, a footer names the retention period and the oldest day the data reaches back to.

A report covers exactly what your role reaches, like the Dashboard it came from. Two administrators with different grants export different documents for the same range.

Export stays available for a window with no activity in it, because an empty report is a real measurement and worth handing to someone who expects otherwise. When an export fails, Agent Mesh reports the failure on screen rather than saving a broken file.

Next Steps

You have just learned how to read agent health out of end-user ratings. Next, you might want to follow one of those agents through a single task to see where its answers come from. For more information, see Monitoring Activity.

Provide feedback