AI Explanations and Recommendations

When AgentQ is configured for your workspace, the Quality Score side panel is augmented with AI-written explanations and actionable recommendations. This page describes what the AI adds to the panel, where it appears, and how to read each indicator.

Availability signal: the sparkle icon on Quality Score chips

Every Quality Score chip in the platform (on the Source Datastores listing, on datastore and container overviews, on the Explore fields view, and in inline metadata cards) is joined by a sparkle when AgentQ is configured. Hovering the chip shows a supplemental line that reads AI-augmented explanations. Absence of the sparkle means the tenant has no LLM Integration and the panel will render the static content only.

Clicking any Quality Score chip opens the same right-side slide-in panel, whether the chip is on a datastore, a container (table or file), or a field. The panel's anatomy is identical at all three levels; only the underlying signals the AI reasons about differ (a field-level panel reasons about that field's profile, a container-level panel about the container's checks and fields, a datastore-level panel about its worst-scoring containers).

For the step-by-step tutorial on reading the panel, see View AI Explanations and Recommendations.

Panel anatomy

The panel is organized top-to-bottom into these sections:

1. Header with the asset avatar, name, metadata pills, and (for containers and fields) a breadcrumb back to the parent datastore or container. On datastore and container panels opened from the asset's own page, users with the required permission also see a Score button that opens the weight and decay settings; the button is not shown on field-level panels or when the panel is opened from a listing or Explore.
2. Overall score card with a large circular gauge showing the numeric score alongside the AI summary of the score's main drivers, plus a delta line showing how much the score moved over the recent period.
3. Decay-period bar visualizing the window the score covers, with start, today, and end labels, a marker for when the score was measured, and a note of how many days have passed since the score was last updated.
4. Dimension Scores, one row per dimension, each showing:
   • the dimension name and icon
   • a weight badge with the weight at the time the score was measured
   • a short AI-written explanation of the dimension's current effect on the total score, with an info icon exposing the static definition on hover (without AgentQ, the static definition appears inline instead)
   • a small sparkline of the dimension's recent history
   • the dimension's own circular score gauge
   • a chevron-stack indicator encoding the direction and magnitude of the dimension's effect on the total score
5. Recommendations, an amber-bordered section with one actionable entry per dimension scoring below 50. Hidden entirely when every dimension meets the threshold.
6. AgentQ configuration banner at the bottom, shown only when the integration is missing and the current user has permission to configure integrations.

The AI summary at the top

A single sentence sits next to the overall score gauge, written by AgentQ from the current dimension scores, dimension weights, recent trend, and (for containers) top failing checks. While the summary is streaming, skeleton bars occupy the slot until the first characters arrive; the full sentence then replaces them. When AgentQ is not configured, this slot renders the platform's static description of the score.

The summary is deliberately short (a single sentence, capped in length) so it fits at the top of the panel and reads at a glance.

Per-dimension AI explanation

Under each dimension name, a short AI-written line describes whether the dimension is currently boosting or dragging the total score, or having no meaningful effect. While the explanation is streaming, a small skeleton line sits in its place; once it lands, it replaces the static definition. The static definition remains reachable via the info icon on the dimension row.

Direction and magnitude: the chevron-stack indicator

The right edge of every dimension row carries a stack of chevrons that encodes both direction and magnitude:

The direction and magnitude are computed by the backend from each dimension's contribution to the total score, not by the AI. This makes the indicator stable across renders (the AI text may rephrase itself, but the chevron stack for the same score always looks the same).

Hover tooltips on the indicator carry a title with the impact level (Low Impact, Medium Impact, High Impact, or No Impact) and a body matching the direction (Boosting the overall score, Dragging the overall score down, This dimension is not moving the overall score right now). Dimensions that have not been measured get their own tooltip wording explaining how the missing measurement is being treated.

The weight badge

The small badge next to each dimension label shows the weight that was in effect at the time the score was measured, not the current setting. If you edit a dimension's weight later, historical scores keep their original weight badges: they still show the number that produced the score you see.

A weight of 0 on the badge marks a dimension as excluded from the score. The row still renders (so you can see the dimension's raw score); the badge tooltip explicitly notes that a weight of 0 means the dimension is not currently being scored.

The Recommendations section

Only dimensions scoring strictly below 50 trigger a recommendation. Each qualifying dimension gets exactly one entry in the amber-bordered section that names the dimension, restates its icon, and carries an AgentQ-written paragraph of advice. Each entry can carry inline action buttons: most jump to a pre-filtered anomalies or checks view for the asset, and an Add a check action opens the create-check form directly (this action is not offered at the datastore level), so acting on a recommendation is a single click.

If no dimension is below 50, the entire Recommendations section is hidden; there is no empty state. A score of exactly 50 does not qualify.

Level differences

The panel behaves the same at datastore, container, and field level:

• Datastore panel reasons about worst-scoring containers and containers-without-checks in the decay window.
• Container panel additionally reasons about top failing checks and fields-without-checks.
• Field panel reasons about the specific field's profile and its check coverage.

Some level-specific patterns to expect:

• A field-level panel with only a few measured dimensions will more often show no Recommendations section, simply because it has fewer opportunities to score below 50.
• The Recommendations CTAs at the datastore level jump to the datastore's anomalies or checks view; at the container level, they jump to the container's view; at the field level, they jump to the field's view.

When AgentQ is not configured

Without an LLM Integration configured for the workspace, the Quality Score panel still opens and renders the following:

• The overall score gauge and static explanation.
• The dimension rows with their static definitions, sparklines, chevron stacks, weight badges, and per-dimension circular gauges (all of these come from the score itself, not from AI).

Absent when the integration is missing:

• The one-sentence AI summary of the main drivers (replaced by the static description).
• The per-dimension AI-written explanation lines (replaced by the static definitions inline).
• The AI-written advice on Recommendation cards (replaced by static per-dimension recommendation copy). The Recommendations section still renders whenever a dimension scores below 50; only the wording of each card changes.
• The sparkle icon on the Quality Score chips (users see no availability signal).

Users with permission to manage integrations see a small banner at the bottom of the panel with a shortcut to Integration settings so they can enable AgentQ and unlock the AI surfaces.