Emerging

Source freshness and coverage

State when sources were updated and how much of the corpus was actually searched.

MAsk MilkyNorthfield Dental

Has Northfield Dental raised any concerns about the renewal?

Milky 4 documents

I didn't find any. Their recent tickets are routine: a login reset ticket · 3 days ago and a request for an invoice copy billing · today.

Based on 2 of 4 sources

Call notes weren't searched: that index is 41 days old. Renewal conversations usually happen on calls.

What was searched 2 of 4 sources

Help desk tickets
Synced 2 hours ago · 3 matches
Billing
Synced today 08:10 · 1 match
Call notes
Not searched · last synced 21 Aug, 41 days ago
Leo's email
Not connected

Try it: Only 2 of 4 sources were searched. Refresh the stale call notes and watch the confident “no” change.

In the wild

How one product ships this pattern. Click a screenshot to enlarge it.

Perplexity answer about EU AI Act enforcement with a citation hover card open: '1/3' paging, '3 sources', the cited digital-strategy.ec.europa page, and a 'Government' label explaining what kind of source the domain is. — Perplexity
Hovering a citation shows how many sources back that sentence and a source-type label, here 'Government', explaining why the domain can be relied on.
Source: Perplexity (captured Sep 2026) ↗

When to use

Retrieval products where staleness or thin coverage is the dominant failure mode.

Seen in

Perplexity date filters, Glean coverage notices, enterprise RAG copilots surfacing thin retrieval.

For engineers

Surface retrieval metadata already returned by the index; flag below-threshold coverage explicitly.

Build prompt

A neutral, component-only prompt for a coding agent. Edit the stack block or anything else to fit your project, then copy it.

# Build the "Source freshness and coverage" pattern: `SourceCoverage`

## Stack (edit this block to match your project)
- React 18+ with TypeScript: one function component, typed props, no `any`.
- Tailwind CSS, using your design system's tokens if you have them.
- Presentational only: data comes in through props, decisions go out through callbacks. No fetching or global state inside.
- If your codebase already has conventions (naming, shared components, tokens), follow them over this spec.

## What the pattern is for
State when sources were updated and how much of the corpus was actually searched.

Use it when: Retrieval products where staleness or thin coverage is the dominant failure mode.

## What to build
A compact coverage line attached to an AI answer that says how many sources were searched out of how many are available, and how fresh each one is, with an expandable list. When coverage is below a threshold or a key source is stale, it says so next to the answer and offers to refresh and search again. It renders under a retrieval answer.

### Props
- `sources`: Array<{ id: string; name: string; state: "searched" | "stale" | "not-searched" | "not-connected" | "syncing"; lastSyncedAt?: Date; matches?: number; reason?: string }>, from the index's retrieval metadata.
- `documentCount`: number, documents actually used in the answer.
- `threshold`: { minSources?: number; maxAgeDays?: number }, below which coverage is flagged.
- `warning?`: { title: string; detail: string }, why the gap matters here, e.g. "Renewal conversations usually happen on calls".
- `onRefresh?(sourceId: string)`: asks the parent to re-sync and re-run. The component never syncs itself.
- `onConnect?(sourceId: string)`: for sources that aren't connected.
- `expanded`: boolean and `onExpandedChange(v: boolean)`.

### States it must render
- Full coverage: a quiet line "Based on 4 of 4 sources, all synced today".
- Thin coverage: an amber notice by the answer, "Based on 2 of 4 sources", with the warning detail and "Refresh call notes and search again".
- Expanded list: each source with an icon and text state, e.g. "Help desk tickets: synced 2 hours ago, 3 matches", "Call notes: not searched, last synced 21 Aug, 41 days ago", "Email: not connected".
- Refreshing: the source row shows "Syncing 6 calls since 21 Aug" and the answer dims until the parent passes the new one.
- Updated: the count rises ("3 of 4 sources"), changed rows highlight briefly, and any remaining gap is still stated.

### Interactions
- The coverage line toggles the source list.
- Refresh calls `onRefresh(sourceId)`; Connect calls `onConnect(sourceId)`.
- Hovering a source highlights the parts of the answer drawn from it, when the parent provides that mapping.

### Edge cases
- A "none found" answer on thin coverage: the warning is required, since absence from two sources is not absence.
- Sync timestamps in the future or missing: show "Sync time unknown" rather than "just now".
- Many sources (more than 8): show the flagged ones first, then a count of the rest.
- Refresh finds nothing new: say "Refreshed, no new matches" so the person knows it ran.

## Must avoid
Presenting a confident answer built on two documents as if the whole corpus was consulted.

## Engineering constraint
Surface retrieval metadata already returned by the index; flag below-threshold coverage explicitly.

## Accessibility
- Source states are words, not only icons; each row's accessible name includes name, state and age.
- Announce when a refresh finishes and the new coverage count.

## House rules
- Real `<button>`/`<input>` elements, full keyboard use, visible focus; announce status changes with `aria-live="polite"`; respect `prefers-reduced-motion`.
- All user-facing text comes from props (or a `labels` prop) so it can be translated.
- One consistent accent marks anything the AI made or is doing; amber means "needs a person", red means risky, green means done. Never rely on colour alone.
- Fits a 360 px wide container without horizontal scrolling.

## Acceptance checks
- [ ] Every state above can be rendered by passing props, with no internal fake data.
- [ ] Every interaction and edge case above behaves as described.
- [ ] Nothing in "Must avoid" can happen with this component.

## Deliver
1. The component file with its exported prop types.
2. A short usage example that renders each state.

Download this prompt as Markdown, or every prompt as JSON for an agent.

Critique checklist for trust indicators

0 of 4 checked

These questions apply to every pattern in this family. Ticks are saved in this browser.

Can a claim be traced to a source in one click?
Does uncertainty change the interface, not just the wording?
Is confidence calibrated, or decorative?
Does the design make verification cheaper than re-doing the work?