UXS-006: Dashboard

Status: Active
Authors: Podcast Scraper Team
Parent UXS: UXS-001: GI/KG Viewer -- shared tokens, typography, layout, states
Normative layout & charts: see ## Dashboard implementation specification below (full product spec merged into this UXS).
Related PRDs:
- PRD-025: Corpus Intelligence Dashboard
Related RFCs:
- RFC-071: Corpus Intelligence Dashboard
- RFC-062: GI/KG viewer v2
Implementation paths:
- web/gi-kg-viewer/src/components/dashboard/DashboardView.vue
- web/gi-kg-viewer/src/components/dashboard/BriefingCard.vue
- web/gi-kg-viewer/src/components/dashboard/IndexStatusCard.vue
- web/gi-kg-viewer/src/components/dashboard/TopicClustersStatusBlock.vue
- web/gi-kg-viewer/src/utils/chartRegister.ts
- web/gi-kg-viewer/src/stores/indexStats.ts, web/gi-kg-viewer/src/stores/dashboardNav.ts
Shell IA: VIEWER_IA.md — canonical shell layout, navigation axes, subject rail, status bar, first-run behavior

Summary

For shell layout, the three navigation axes, subject rail persistence and clearing, status bar, and first-run empty corpus behavior, see VIEWER_IA.md. This document specifies the Dashboard main tab only (briefing card, Coverage / Intelligence / Pipeline sub-tabs, charts, and layout below).

The Dashboard main tab is briefing + three tabs only: Coverage (default), Intelligence, Pipeline. Corpus artifact picking (List, All / None, Load into graph) lives on the status bar (List opens data-testid="artifact-list-dialog"). Deep links to Library, Digest, and Graph use dashboardNav handoffs consumed when those tabs activate. Chart.js uses shared Tufte-style defaults from chartRegister.ts.

Layout

Briefing card (data-testid="briefing-card") — last run / health / short actions; always above tabs.
Tablist aria-label="Dashboard tabs" — Coverage | Intelligence | Pipeline.
Coverage — coverage by month, feed coverage table, artifact activity (from listed artifacts), Index status (data-testid="index-status-card") — status-only; index facts + rebuild actions moved to the Configuration dialog's Index section, reached via Manage in Configuration (index-status-manage).
Intelligence — digest snapshot, Topic briefings (topic-briefing-cards, PRD-033 FR6.1 #888 — retrieval-grounded topic-briefing-card per top topic, see §5), Topic clusters status (topic-clusters-status-block), topic landscape, top voices (when API available). Topic momentum / emerging connections omitted until RFC-073 data ships (no placeholder UI). Search activity (query-activity-chart, PRD-033 FR6.2 — daily search-volume bar chart, see §5). Topic momentum stays omitted.
Pipeline — run history strip, duration trend, stage timings, numeric outcomes, episodes per run; optional per-feed run heatmap only when server exposes stable per-feed fields.

The legacy CorpusDataWorkspace / Pipeline | Content intelligence split on Dashboard is removed; those components are not part of this surface.

Corpus artifacts (status bar)

List (when health is ok and a corpus path is set) fetches GET /api/artifacts and opens the Corpus artifacts dialog. Load into graph switches the main tab to Graph when load succeeds (same behavior as the former workspace).

E2E contract

E2E surface map — Dashboard tab, briefing-card, Dashboard tabs tablist, Index status card, artifact-list-dialog; openCorpusDataWorkspace only switches to Dashboard and waits for the briefing card.

Revision history

Date	Change
2026-04-06	Initial content (in UXS-001)
2026-04-13	Extracted from UXS-001 into standalone UXS-006
2026-04-19	Corpus workspace on Dashboard; left rail query-only IA
2026-04-19	Dashboard: briefing + tabs; artifacts via status bar dialog
2026-04-20	§6.0 jobs: operator `--config` + optional `profile:` merge (#593)
2026-04-20	§6.0 jobs (cont.): feeds via `--feeds-spec` when `feeds.spec.yaml` exists
2026-04-21	§6.0: Operator YAML = Config; shallow PUT validation (feed + secret keys)
2026-06-05	PRD-033 FR6.1 briefing cards + FR6.2 search-activity chart (#888 + follow-up)

Dashboard implementation specification

Status: Ready for implementation Author: Design session (Marko + Claude), April 2026 Repo: chipi/podcast_scraper Target area: web/gi-kg-viewer/src/components/dashboard/ Related docs: UXS-006, RFC-071, TUFTE_CHART_CRITIQUE.md, UXS-001 Replaces: Current UXS-006 content (rewrite, not amendment)

1. Overview

The Dashboard is restructured around two principles:

5-second answers. The primary question — "is everything OK, and what should I do?" — is answered by a permanent briefing card above the tabs, before any chart is read.

Actionability. Every data point either has an attached action or is explicitly decorative context. Observations without a "so what?" belong in depth charts, not on the primary surface.

Structure:

├── BRIEFING CARD  (permanent, always visible, above tabs)
│   Last run · Corpus health · Action items
│
└── TABS
    ├── Coverage      — what fraction of my corpus has intelligence coverage?
    ├── Intelligence  — what is my corpus telling me?
    └── Pipeline      — how is my pipeline performing?
```yaml

Three tabs instead of the current two. The old "Status" concept becomes
the briefing card. The old "Content Intelligence" becomes Coverage +
Intelligence split by purpose.

---

### 2. Global Chart Rules (Tufte)

These apply to every chart in the dashboard without exception. They
are not per-chart suggestions — they are design system constraints.

#### 2.1 Non-negotiable rules

**No legends.** Every series is labelled directly — end-of-line for
line charts, end-of-bar for bar charts, positioned annotation for
everything else. No legend box anywhere in the dashboard.

**Title = insight.** Every chart title states the finding, not the
variable name. "Coverage by month" is a label. "N months below average
GI coverage" is a title. Mandatory, not optional.

**No gridlines.** Remove all gridlines from all charts. Reference
lines (average, target) are a single `border`-token hairline annotated
directly — not a grid.

**No dual y-axes.** If two series have different scales, they get
separate charts. No exceptions.

**Remove top and right spines.** Only left and bottom axis lines on
all Chart.js charts.

**Y-axis starts at 0 on all bar charts.** No truncated axes.

**Mandatory insight line.** Remove the word "optional" from UXS-001
and UXS-006. Every chart has an insight line below it. If the data
does not support a clear takeaway, the chart does not belong on the
dashboard.

**Charts earn their place.** Before adding a chart, ask: does this
need visual encoding, or is a number or table sufficient? Three numbers
are not a chart. If the answer is "a table," build a table.

**No chartjunk.** No 3D effects, no gradient fills, no drop shadows
on chart elements, no decorative borders around chart areas.

#### 2.2 Chart.js global config

Apply these defaults in `chartRegister.ts` so every chart component
inherits them without per-component overrides:

```typescript
Chart.defaults.plugins.legend.display = false
Chart.defaults.plugins.tooltip.enabled = true

Chart.defaults.scales.linear = {
  ...Chart.defaults.scales.linear,
  grid: { display: false },
  border: { display: false },
  ticks: { maxTicksLimit: 5, color: 'var(--ps-muted)' }
}

Chart.defaults.scales.category = {
  ...Chart.defaults.scales.category,
  grid: { display: false },
  border: { display: false },
  ticks: { color: 'var(--ps-muted)' }
}

2.3 Direct end-label plugin

A shared Chart.js afterDraw plugin for direct series labelling. Register once in chartRegister.ts. Used by all multi-series line charts:

const endLabelPlugin = {
  id: 'endLabel',
  afterDraw(chart: Chart) {
    const ctx = chart.ctx
    chart.data.datasets.forEach((dataset, i) => {
      const meta = chart.getDatasetMeta(i)
      if (!meta.hidden && meta.data.length > 0) {
        const lastPoint = meta.data[meta.data.length - 1]
        ctx.fillStyle = dataset.borderColor as string
        ctx.font = '10px Inter, system-ui'
        ctx.textAlign = 'left'
        ctx.fillText(
          dataset.label ?? '',
          lastPoint.x + 6,
          lastPoint.y + 3
        )
      }
    })
  }
}
Chart.register(endLabelPlugin)
```yaml

---

### 3. Briefing Card

#### 3.1 Purpose

Permanent section at the top of the Dashboard page, above the tab bar.
Always visible regardless of which tab is active. Answers the three
primary questions in under 5 seconds:

- "Did the last run succeed?" (Last run section)
- "Is my corpus healthy?" (Corpus health section)
- "What should I do?" (Action items section)

#### 3.2 Layout

```text
│  CORPUS      213 episodes · 67% GI · 82% indexed · 3 feeds      │
│  ──────────────────────────────────────────────────────────────  │
│  → 71 episodes have no GI artifacts          [View in Library]   │
│  → Index last rebuilt 14 days ago            [Rebuild now]       │
│  → 2 episodes failed in last run             [View failures]     │
└──────────────────────────────────────────────────────────────────┘

Background: elevated token (lifts the card above the page surface)
Border: border token, rounded-sm
Padding: p-4
Divider between health and action items: border-t border-border mt-3 pt-3
Section labels ("LAST RUN", "CORPUS"): text-[10px] font-semibold tracking-wider muted small caps
Data values: text-sm surface-foreground

3.3 Last run section

Sources: GET /api/corpus/runs/summary most recent item.

Format: ● [status] · [N] episodes · [duration] · [age] [Details →]

Status badge:

● Success — success token dot
● Partial — warning token dot (some feeds failed in multi-feed run)
● Failed — danger token dot

Age: relative time — "2 hours ago", "yesterday", "3 days ago". Not a raw timestamp. Computed client-side from run timestamp.

Duration: human-readable — "3m 24s", not "204s".

Multi-feed: "3 of 4 feeds succeeded" replaces episode count when multi_feed_summary is present and any feed failed.

"Details →" link: navigates to Pipeline tab.

Empty state (no runs found): muted text — "No pipeline runs found. Run podcast scrape to begin."

data-testid="briefing-last-run"

3.4 Corpus health section

Sources: GET /api/corpus/coverage (NEW endpoint — see Section 8) + GET /api/index/stats.

Format: [total] episodes · [gi]% GI · [indexed]% indexed · [feeds] feeds

Each metric is a link:

Episode count → Library tab
GI% → Coverage tab (scrolls to coverage chart)
Indexed% → Coverage tab (scrolls to index section)
Feed count → Library tab (feed filter)

Warning threshold: if GI% < 50% or indexed% < 60%, that metric renders in warning token colour. Immediate visual signal without reading action items.

Thresholds are tunable parameters (UXS-001 table):

Parameter	Default	Status
GI coverage warning threshold	50%	Open
Index coverage warning threshold	60%	Open

data-testid="briefing-corpus-health"

3.5 Action items section

Max 3 items. Triage order: pipeline failures first (blocking), coverage gaps second, operational staleness third.

Sources: assembled client-side from coverage data + index stats + run summary.

Item format: → [plain-language finding] [primary action link] [secondary action link]

Action item rules:

Plain language, present tense: "71 episodes have no GI artifacts" not "GI coverage: 67%"
Primary action link: the most direct next step
Max one secondary action link
Never says "run enrich" or references RFC-073 enrichers — out of scope

Possible items (ordered by priority):

Condition	Text	Actions
Last run has failures	"N episodes failed in last run"	[View failures] → Library
Last run failed entirely	"Last run failed — no episodes processed"	[View in Pipeline]
GI coverage < 50%	"N episodes have no GI artifacts"	[View in Library]
Index not built	"Vector index has not been built"	[Open index controls]
Index stale > 7 days	"Index last rebuilt N days ago"	[Open index controls]
Feed not indexed	"Feed X is not in the vector index"	[View coverage]
No runs in > 7 days	"No pipeline runs in N days"	[View in Pipeline]
Topic clusters missing	"Topic clusters not built"	[View in Coverage]

"Open index controls" routes to the Configuration dialog's Index section (sources-dialog-index-panel), where the rebuild is triggered — the rebuild action no longer fires inline from the Briefing card (config consolidation). All other actions are navigation links.

All-clear state: When no items qualify: a single line — ● Everything looks good — using success token dot. This positive confirmation is as important as finding items — it tells you to stop looking.

data-testid="briefing-action-items" data-testid="briefing-action-item" (per item) data-testid="briefing-all-clear" (empty state)

4. Coverage Tab

Answers: "What fraction of my corpus has intelligence coverage, and where are the gaps?"

Four sections, top to bottom.

4.1 GI coverage by month

Chart type: Single-series bar chart.

What it shows: GI coverage percentage per publish month (0–100%). One bar per month. Bars are the primary story — what fraction of episodes published that month have GI artifacts.

Not a stacked chart. Not episode counts. Coverage rate is the signal, not volume.

Tufte compliance:

Single series, no legend needed
Y-axis: 0–100%, labelled "Coverage %", max 5 tick labels
X-axis: month labels, text-[10px]
Reference line: horizontal hairline at corpus average coverage %, annotated directly — "avg 67%" placed at the right end of the line
Bars below average: warning token fill
Bars at or above average: gi token fill
No gridlines, no top/right spines
Direct annotation on the lowest bar: "Lowest: [month] — N%"
Insight line below chart (mandatory): computed from data, e.g. "3 months below average — [month], [month], [month] need attention"

or "Coverage improving: up [X]pp since [earliest month]"

Interaction: clicking any bar navigates to Library filtered to episodes from that month without GI artifacts (since + end of month + topic_cluster_only=false + filter for no-GI).

Source: GET /api/corpus/coverage → by_month array (NEW).

Component: CoverageByMonthChart.vue (new, replaces VerticalBarChart.vue for this use case)

Empty state: "No episode metadata found. Run the pipeline to generate corpus data."

data-testid="coverage-by-month-chart"

4.2 Feed breakdown

Not a chart — a table with inline data bars.

What it shows: Per-feed coverage summary. Each row = one feed. Columns: Feed name | Episodes | GI coverage | KG coverage | Indexed.

GI coverage and KG coverage columns show a 40px inline progress bar (CSS, not Chart.js) plus the percentage number. This gives visual comparison without a separate chart.

Tufte note: 40px inline bars are proportional (0–100% always), direct labelled (number next to bar), and scannable. No legend, no axis, no gridlines.

Sort order: ascending by GI coverage — worst-covered feeds at top, since those are the action targets.

Row interaction: clicking a row navigates to Library filtered to that feed.

Insight line: "Feed [name] has lowest GI coverage at [N]% — [M] episodes without GI artifacts."

Source: GET /api/corpus/coverage → by_feed array (NEW).

Component: FeedCoverageTable.vue (new)

data-testid="feed-coverage-table" data-testid="feed-coverage-row" (per row)

4.3 Artifact activity

Chart type: Grouped bar chart — new artifacts per day, last 30 days.

Replaces: The current cumulative GI+KG line chart.

Why: Cumulative lines always trend upward and look healthy even when work has stopped. A recency chart makes silence visually obvious.

What it shows: For each day in the last 30 days: count of new GI artifacts + count of new KG artifacts written that day.

Tufte compliance:

Two series: gi token bars and kg token bars, grouped per day
Series labelled directly: "GI" and "KG" as text-[10px] text positioned above the first bar of each series (not a legend)
X-axis: day labels, only show every 7th label to avoid clutter
Y-axis: starts at 0, labelled "New artifacts"
No gridlines, no spines except left + bottom
Silence is visible: days with no bars are visually obvious gaps
If last 14 consecutive days have no bars: mandatory insight reads "No new artifacts in 14 days — pipeline may not be running"
Otherwise insight: "Last GI: [date] · Last KG: [date]"
Annotate the most recent active day: small dot + date label

Source: GET /api/artifacts → mtime_utc values per artifact, bucketed by day client-side. Existing artifactMtimeBuckets.ts — extend to produce daily counts rather than cumulative.

Component: ArtifactActivityChart.vue (new — replaces CategoryLineChart.vue for this purpose)

data-testid="artifact-activity-chart"

4.4 Index status

Not a chart — a card.

The question "is my index current?" doesn't need a chart. It needs three facts:

Feeds in index: 3 of 3
```python

- "⚠ Rebuild recommended" appears when index staleness heuristic
  from `index/stats` is true. Uses `warning` token.

- "Last rebuild error: [message]" in `danger` token when
  `rebuild_last_error` is present.

- **Status-only.** The index facts + rebuild *action* live in the
  Configuration dialog's **Index** section (`sources-dialog-index-panel`) —
  the single canonical place per concern (config consolidation). The card has a
  **Manage in Configuration →** button (`data-testid="index-status-manage"`)
  that opens Configuration → Index rather than triggering a rebuild inline. The
  Briefing **Open index controls** action and the status-bar rebuild
  bolt route to the same dialog.

No Chart.js involved. Pure key-value status card.

**Source:** `GET /api/index/stats` (existing).

`data-testid="index-status-card"`

---

### 5. Intelligence Tab

Answers: "What is my corpus telling me?"

Five sections. Some require enricher data and degrade gracefully.
Enricher triggers ("run enrich") are out of scope — degraded states
show what's missing, not how to generate it.

#### 5.1 Corpus snapshot

**Always available.**

An expanded version of the digest compact view. Shows the rolling
window summary (default 7d) with a little more breathing room than
the Digest tab's compact layout.

Content:

- Window line: "Last 7 days — 12 new episodes across 3 feeds"
- Top 3 topic bands from the digest, one line each: topic label +
  episode count ("AI policy — 4 episodes")

- Each topic band line clickable → Digest tab (which loads that topic)
- "Open Digest →" link at bottom

Not a chart. A structured text summary using the digest API data.

**Source:** `GET /api/corpus/digest` with `window=7d`, `compact=false`,
`max_rows=3`.

`data-testid="intelligence-snapshot"`

#### 5.1a Topic briefings (PRD-033 FR6.1, #888)

**Always available when the digest returns scored topic bands.**

A compact grid of **retrieval-grounded** briefing cards — one per top topic, ranked by
retrieval signal (best hit score lifted by insight density + distinct-show breadth).
Each card is grounded in a *direct query*, **not** a briefing pack (the MCP briefing-pack
form is orthogonal, RFC-093): the digest topic bands are already built by running a
semantic search per topic, so each card surfaces:

- Topic label (a `topic-briefing-card-link` button when the topic maps to a KG node →
  opens the **Topic Entity View** rail via `focusTopic`; plain text otherwise) + episode count.
- **Top segment**: the band's best-scored hit (`summary_preview`) with its retrieval
  score badge — the "grounded in retrieval" evidence.
- **Cross-show** (`topic-briefing-card-cross-show`, mapped topics only): the top insight
  from `cross_show_synthesis` plus the number of distinct shows — the corpus
  differentiator, lazily fetched from `GET /api/relational/cross-show`.

**Source:** the already-fetched `GET /api/corpus/digest` bands + `GET /api/relational/cross-show`.
`data-testid="topic-briefing-cards"` (container), `topic-briefing-card` (each).

#### 5.1b Search activity (PRD-033 FR6.2)

**Available once at least one search has been logged.**

A Tufte-simple single-series bar chart of **daily search volume** over the last 30 days
(`query-activity-chart`). The data is the append-only search-activity log
(`<corpus>/search/query_log.jsonl`) that `/api/search` writes — **timestamp + detected
intent only, never the raw query text** (privacy-light), read via
`GET /api/corpus/query-activity`.

**Honest scope:** the PRD's literal "query volume *by topic* over time" is not
achievable — queries are not topic-tagged, and a corpus-coverage-by-topic series would
mislabel corpus volume as query volume. This ships the supported subset: *search volume
over time*. The chart is hidden until the log has entries (`total > 0`).

#### 5.2 Topic landscape

**Available when topic clusters have been built.**

A compact grid of topic clusters. Each cluster card:

- Cluster canonical label (`text-sm font-semibold`)
- Member topic count badge ("4 topics")
- Episode count (summed from `members[].episode_ids`, deduplicated)
- Clicking → Graph tab, focused on that TopicCluster compound node

Grid: `repeat(auto-fit, minmax(min(100%, 12rem), 1fr))` — same card-width logic as Digest topic bands
but cards are smaller and denser — no hit rows, just cluster identity.

**Insight line:** "N topic clusters covering M distinct topics."

**Degraded state** (404 from topic-clusters endpoint): `muted` text —
"Topic clusters not yet built for this corpus." No action prompt.

**Source:** `GET /api/corpus/topic-clusters` (existing).

`data-testid="intelligence-topic-landscape"`

#### 5.3 Top voices

**Available when `GET /api/corpus/persons/top` exists (NEW endpoint —
see Section 8).**

Top 5 persons by insight count. Each person card:

- Display name (`text-sm font-semibold`)
- Episode count badge: "23 episodes"
- Insight count badge: "67 insights" (`gi` token)
- Top 3 topic chips (`kg` token border) — topics from the endpoint
- "View profile →" link → Person Landing (UXS-010) in subject rail

Cards in a horizontal scrollable row on narrow viewports, 2-column
grid on wide viewports.

**Insight line:** "N speakers with grounded insights — [top person]
leads with N insights across M episodes."

**Degraded state** (endpoint not available or 0 persons): `muted` text —
"No speaker intelligence found for this corpus."

**Source:** `GET /api/corpus/persons/top?limit=5` (NEW).

`data-testid="intelligence-top-voices"`

#### 5.4 Topic momentum

**Enricher-gated — degrades gracefully.**

Available when `temporal_velocity` enricher data is present (RFC-073).

Top 5 topics by recent activity. Each row:

- Topic name (`text-sm`)
- Sparkline: 8-point mini line chart, last 8 weeks of mention
  frequency. `series-1` token stroke, no fill. No axes, no labels —

  the shape is the signal.

- Trend badge: accelerating (`success`) / stable (`muted`) /
  declining (`warning`) — same as UXS-007

- Episode count: "N episodes" in `muted`
- Clicking → Topic Entity View (UXS-007) in subject rail

**Tufte note on sparklines:** y-scale normalises to each topic's own
range (0 to its max). This shows *relative* trend shape, not absolute
volume. Volume is shown as the episode count number. This is acceptable
for a "relative momentum" display — the spec must note this so
implementers don't try to share y-scales (which would make
low-volume topics invisible).

**Insight line:** "N topics accelerating in the last 30 days."

**Degraded state:** `muted` text — "Topic momentum data not available
for this corpus." No enricher prompt.

**Source:** RFC-073 enricher output (not yet available).

`data-testid="intelligence-topic-momentum"`

#### 5.5 Emerging connections

**Enricher-gated — degrades gracefully.**

Available when `topic_cooccurrence` enricher data is present (RFC-073).

Top 4 topic pairs by unexpected co-occurrence strength. Each row:

- "Topic A ↔ Topic B" (`text-sm`)
- Episode count: "14 episodes" (`muted`)
- Clicking → Graph tab, both topic nodes highlighted + focused,
  subgraph showing their shared connections

**Insight line:** "N topic pairs co-occur more than expected given
their individual frequency."

**Degraded state:** `muted` text — "Topic co-occurrence data not
available for this corpus." No enricher prompt.

**Source:** RFC-073 enricher output (not yet available).

`data-testid="intelligence-emerging-connections"`

---

### 6. Pipeline Tab

Answers: "How is my pipeline performing?"

Now freed from being the default tab (briefing card handles the quick
answer), the Pipeline tab can be dense and detailed.

#### 6.0 HTTP pipeline jobs (RFC-077)

When **`jobs_api`** is true on health, the tab includes **`PipelineJobsCard`**
(`data-testid="pipeline-jobs-card"`): list **HTTP-triggered** pipeline jobs for
the current corpus, **Run** (enqueue), **Reconcile**, **Refresh**, per-row **Cancel**
when status is `queued` or `running`, and inline help when the API is off. This
surface is **in addition to** run-history charts fed from **`GET /api/corpus/runs/summary`**
(see below). Spec: [RFC-077](../rfc/RFC-077-viewer-feeds-and-serve-pipeline-jobs.md).

**Config a job uses:** the subprocess passes **`--config`** to the server-resolved operator YAML path (same file as status bar **Operator YAML** / **Config**). That file may contain **`profile: <preset>`** — effective pipeline settings are **packaged preset defaults merged with explicit keys** in the file ([GitHub #593](https://github.com/chipi/podcast_scraper/issues/593)). **Feeds** for the job come from corpus **`feeds.spec.yaml`** via **`--feeds-spec`** when that file exists, not from duplicate root keys in operator YAML (those keys are rejected on **`PUT /api/operator-config`** — **top-level** denylist only, same as secrets). Any **Run** / help copy in this tab should say **operator file + optional profile**, not “edit feeds here.”

#### 6.1 Run history strip

**Not a chart — a row of status dots.**

Last 10 runs rendered as coloured dots in a horizontal row. No axes,
no labels — maximum information density.

- `●` success token = run succeeded
- `●` warning token = partial success (some feeds failed)
- `●` danger token = run failed

Dots are ordered left-to-right oldest to newest. Most recent dot is
slightly larger (12px vs 8px) to draw the eye to current state.

Native `title` tooltip per dot: "Run [date] · [N] episodes ·
[duration] · [status]"

Clicking a dot: expands that run's detail inline below the strip,
replacing the current "latest run detail" section with the selected
run. Clicking again collapses. Only one run expanded at a time.

**Insight line:** "Last 10 runs: [N] success, [M] partial, [P] failed"
— or simply "All 10 runs succeeded" in `success` token if clean.

**Source:** `GET /api/corpus/runs/summary` (existing).

`data-testid="pipeline-run-history-strip"`
`data-testid="pipeline-run-dot"` (per dot)

#### 6.2 Duration trend

**Chart type:** Bar chart, last 5 runs' durations.

**What it shows:** Run-to-run duration variability. Is the pipeline
getting slower?

**Tufte compliance:**

- 5 bars — simple, no legend needed
- Y-axis: starts at 0, labelled in minutes, max 4 ticks
- X-axis: run dates, `text-[10px]`
- Reference line: horizontal hairline at 5-run average, directly
  labelled "avg [X]m [Y]s"

- Latest bar: `primary` fill. Others: `surface-foreground` at 60%
  opacity. Draws eye to the current run vs history.

- If latest bar > average + 20%: `warning` fill on latest bar
- No gridlines, no top/right spines
- Direct bar labels: duration value above each bar ("3m 24s") in
  `text-[10px] muted`

- **Title = insight** (mandatory): "Latest run [X]% [faster/slower]
  than 5-run average" — computed. If within 10% either way: "Run

  duration stable — avg [X]m [Y]s"

**Source:** `GET /api/corpus/runs/summary` last 5 items sorted by
timestamp.

**Component:** Reuse/update existing chart component.

`data-testid="pipeline-duration-trend"`

#### 6.3 Latest run breakdown

Two sections side-by-side on wide viewports, stacked on narrow.

**Stage timings (left):**

**Chart type:** Horizontal bar chart, one bar per pipeline stage,
sorted by duration descending.

**What it shows:** Where time was spent in the most recently selected
run (from the history strip — defaults to latest).

**Tufte compliance:**

- Horizontal bars — stage names are text, need space to breathe
- Sorted descending: longest bar at top, shortest at bottom
- Direct value labels: duration ("1m 23s") printed at the right end
  of each bar — **no x-axis needed** once values are labelled

- Bar fill: `primary` token for the longest bar (the bottleneck),
  `surface-foreground` at 60% opacity for all others. The bottleneck

  is visually distinct without any annotation.

- No gridlines at all — bar length communicates value
- No spines — labels make axes redundant
- **Title = insight** (mandatory): "[Stage name]: [X]% of total run
  time" — always names the bottleneck

**Episode outcomes (right):**

**Not a chart.** Three numbers are not a chart.

```text
✗  2  failed  [View failures →]

text-sm numbers, intent token colours for each count. "View failures →" link appears only when failures > 0, navigates to Library filtered for that run's failed episodes.

Source: GET /api/corpus/runs/summary for the selected run (or run detail endpoint if stage timings need fuller data — see open questions in Section 9).

data-testid="pipeline-stage-timings" data-testid="pipeline-episode-outcomes"

6.4 Episodes per run

Chart type: Bar chart, all runs in summary (up to 150, capped).

What it shows: Episode count processed per run over time. Shows pipeline activity and cadence.

Tufte compliance:

Single series, no legend needed
Y-axis: starts at 0, "Episodes" label, max 5 ticks
X-axis: run dates (sparse labelling — every Nth label)
No gridlines, no top/right spines
Total corpus as text annotation below chart: "Total: [N] episodes across [M] runs" — not a cumulative overlay on the same axis

(that was the dual-axis lie)
Title = insight: "Average [N] episodes per run" or "Processing volume declining — last 3 runs below average" if trend is negative

Source: GET /api/corpus/runs/summary (existing).

Component: Reuse VerticalBarChart.vue with updated config.

data-testid="pipeline-episodes-per-run"

6.5 Feed processing history

Chart type: Heatmap grid (small multiples).

Only shown for multi-feed corpora. Hidden for single-feed.

What it shows: Per-feed run success/partial/failure over the last 5 runs. Rows = feeds, columns = runs (left-to-right oldest to newest).

Tufte compliance:

Cell colour: success fill / warning fill / danger fill
Cell glyph: ✓ / ⚠ / ✗ inside each cell — colour alone is not sufficient (accessibility + Tufte "don't rely on colour only")
Cell size: ~28px × 28px. Gap spacing between cells instead of borders (whitespace is cleaner than lines)
Feed names: left-aligned row headers, text-xs
Run dates: top-aligned column headers, text-[10px]
No legend — ✓/⚠/✗ glyphs are self-explanatory
Insight line (mandatory): "Feed [name] failed [N] of last 5 runs" — triggered when any feed has ≥ 2 failures in the window.

If all clean: "All feeds succeeded in last 5 runs" (success token).

Source: GET /api/corpus/runs/summary per-feed breakdown (verify field availability in CorpusRunSummaryItem — see Section 9).

Component: FeedRunHistoryGrid.vue (new)

data-testid="pipeline-feed-history-grid"

7. Tab Order and Defaults

Coverage is the default tab (not Pipeline as today) because it answers the most frequent ongoing question — "how complete is my intelligence coverage?" — whereas Pipeline is consulted reactively after running.

The briefing card's "Details →" link on the Last run section jumps to the Pipeline tab.

8. New API Endpoints Required

8.1 `GET /api/corpus/coverage`

Priority: CRITICAL. Unblocks briefing card scorecard + Coverage tab.

Single catalog scan checking GI/KG artifact presence per episode, grouped by month and feed. Same implementation pattern as /api/corpus/stats.

Response schema:

class CoverageByMonthItem(BaseModel):
    month: str           # "YYYY-MM"
    total: int
    with_gi: int
    with_kg: int
    with_both: int

class CoverageFeedItem(BaseModel):
    feed_id: str
    display_title: str
    total: int
    with_gi: int
    with_kg: int

class CorpusCoverageResponse(BaseModel):
    total_episodes: int
    with_gi: int
    with_kg: int
    with_both: int
    with_neither: int
    by_month: List[CoverageByMonthItem]
    by_feed: List[CoverageFeedItem]

Implementation: One filesystem scan. For each episode metadata file: check sibling *.gi.json and *.kg.json existence, group by publish_date month and feed_id. No GI/KG file reading — just existence checks. Fast.

Test: tests/unit/podcast_scraper/server/test_viewer_corpus_coverage.py

8.2 `GET /api/corpus/persons/top`

Priority: MEDIUM. Enables Intelligence tab Top Voices section. Intelligence tab degrades gracefully without it.

Partial GI artifact scan: reads Person node counts from loaded GI artifacts, aggregates by person canonical id, returns top N by insight count.

Response schema:

class TopPersonItem(BaseModel):
    person_id: str            # "person:{slug}"
    display_name: str
    episode_count: int
    insight_count: int
    top_topics: List[str]     # canonical topic ids, max 3

class CorpusTopPersonsResponse(BaseModel):
    persons: List[TopPersonItem]
    total_persons: int        # total distinct persons in corpus

```yaml

**Implementation:** Requires reading GI artifacts. Two options:
(a) scan all `*.gi.json` files, extract Person node counts — full scan,
slower; (b) use the loaded graph in memory if available — faster but
state-dependent. Prefer (a) for correctness — this is a background
fetch, latency is acceptable.

**Test:** `tests/unit/podcast_scraper/server/test_viewer_corpus_persons.py`

---

### 9. Open Questions

These are not blocking Phase 1–2 but need resolution before the
implementation is complete.

**Stage timings granularity:** The current `CorpusRunSummaryItem`
schema provides "compact metrics per run." Verify whether it includes
per-stage timing breakdown or only total duration. If stage timings
are only in the full `run.json` file and not in the compact summary,
a `GET /api/corpus/runs/{run_id}` detail endpoint may be needed for
the stage timings chart. Check the existing `corpus_run_summary.json`
schema before adding a new endpoint.

**Feed per-run breakdown:** `FeedRunHistoryGrid` requires per-feed
success/failure per run. Verify whether `CorpusRunSummaryItem` includes
per-feed breakdown. If not, this chart silently hides (it is conditional
on multi-feed corpus anyway) and is tracked as a follow-up.

**`GET /api/corpus/coverage` performance:** If the corpus has thousands
of episodes, the existence-check scan may be slow. Add a
`Cache-Control: max-age=60` header and consider a lightweight cache
in the route handler (invalidated on corpus path change). Profile with
a real large corpus during implementation.

---

### 10. Files to Touch

#### New components

```text
web/gi-kg-viewer/src/components/dashboard/ArtifactActivityChart.vue
web/gi-kg-viewer/src/components/dashboard/FeedRunHistoryGrid.vue
web/gi-kg-viewer/src/components/dashboard/IntelligenceSnapshot.vue
web/gi-kg-viewer/src/components/dashboard/TopicLandscape.vue
web/gi-kg-viewer/src/components/dashboard/TopVoices.vue
web/gi-kg-viewer/src/components/dashboard/TopicMomentum.vue
web/gi-kg-viewer/src/components/dashboard/EmergingConnections.vue

Modified components

web/gi-kg-viewer/src/components/dashboard/DashboardView.vue
  — briefing card always rendered above tabs

web/gi-kg-viewer/src/utils/chartRegister.ts
  — global Chart.js defaults (no legends, no gridlines)
  — endLabelPlugin registered globally

web/gi-kg-viewer/src/utils/artifactMtimeBuckets.ts
  — extend to produce daily new-artifact counts (not just cumulative)

New API modules

web/gi-kg-viewer/src/api/corpusPersonsApi.ts
  — fetchTopPersons(limit)

New server routes

src/podcast_scraper/server/schemas.py
  — CorpusCoverageResponse, CoverageByMonthItem, CoverageFeedItem
  — CorpusTopPersonsResponse, TopPersonItem

Add coverage and persons/top routes to app.py.

New server tests

Add or extend integration tests for the new coverage and persons routes (mirror patterns used for existing corpus API tests).

UXS amendments (after implementation)

docs/uxs/UXS-006-dashboard.md
  — Full rewrite (current file is thin; this spec replaces it)
  — Remove API/Data left panel section (moved to status bar)
  — Add briefing card spec
  — Add three-tab structure
  — Add Tufte compliance rules (cross-reference TUFTE_CHART_CRITIQUE.md)

docs/uxs/UXS-001-gi-kg-viewer.md
  — Remove "optional" from insight line rule — mandatory
  — Add to tunable parameters:
      GI coverage warning threshold    50%    Open
      Index coverage warning threshold 60%    Open
      Action items max                 3      Open
      Top voices limit                 5      Open

docs/guides/SERVER_GUIDE.md
  — Add GET /api/corpus/coverage and GET /api/corpus/persons/top
    to the API reference table

E2E surface map

  — coverage-by-month-chart, feed-coverage-table, artifact-activity-chart
  — index-status-card
  — intelligence-snapshot, intelligence-topic-landscape
  — intelligence-top-voices, intelligence-topic-momentum
  — intelligence-emerging-connections
  — pipeline-run-history-strip, pipeline-run-dot
  — pipeline-duration-trend, pipeline-stage-timings
  — pipeline-episode-outcomes, pipeline-episodes-per-run
  — pipeline-feed-history-grid
```yaml

---

### 11. Implementation Phases

#### Phase 1 — Briefing card + Coverage tab

**Goal:** The two most impactful surfaces. Requires `GET
/api/corpus/coverage` new endpoint.

Steps:

1. Implement `GET /api/corpus/coverage` server route + tests
2. `BriefingCard.vue` — all three sections wired to coverage + runs/summary
3. Coverage tab: `CoverageByMonthChart.vue`, `FeedCoverageTable.vue`
4. Apply global Chart.js Tufte defaults in `chartRegister.ts`
5. `ArtifactActivityChart.vue` (replaces cumulative line)
6. `IndexStatusCard` section (from existing index/stats data)

**Checkpoint:** Briefing card shows last run + corpus health + action
items. Coverage tab shows month chart, feed table, activity chart,
index card. No regressions in existing Pipeline tab.

#### Phase 2 — Pipeline tab restructure

**Goal:** Improved Pipeline tab with Tufte-compliant charts.

Steps:

1. `RunHistoryStrip.vue` with click-to-expand per run
2. `DurationTrendChart.vue` with insight title + reference line
3. Stage timings chart with direct value labels + bottleneck highlight
4. Episode outcomes → inline numbers in run card (remove chart)
5. `EpisodesPerRunChart.vue` (remove cumulative overlay)
6. `FeedRunHistoryGrid.vue` (multi-feed only)

**Checkpoint:** Pipeline tab fully restructured. All Tufte rules
applied. Episode outcomes is no longer a chart.

#### Phase 3 — Intelligence tab

**Goal:** Intelligence tab with available data + graceful degradation.

Steps:

1. `IntelligenceSnapshot.vue` (digest API, existing data)
2. `TopicLandscape.vue` (topic-clusters API, existing data)
3. Implement `GET /api/corpus/persons/top` + `TopVoices.vue`
4. `TopicMomentum.vue` stub with degraded state (enricher not yet
   available)

5. `EmergingConnections.vue` stub with degraded state

**Checkpoint:** Intelligence tab shows snapshot + topic landscape
immediately. Top voices shows when persons/top endpoint ships. Momentum
and connections show degraded states (no enricher prompts).

#### Phase 4 — UXS + E2E updates

1. Rewrite `UXS-006-dashboard.md` to match implementation
2. Update `UXS-001-gi-kg-viewer.md` tunable params + mandatory insight
   line rule

3. Update `SERVER_GUIDE.md` API table
4. Update `E2E_SURFACE_MAP.md` with all new testids
5. Run `make test-ui-e2e`

---

### 12. What This Does Not Change

- Chart.js library itself — same version, same registration pattern
- `indexStats.ts` store — unchanged
- Digest and Library tabs — unchanged
- Subject rail — unchanged
- Token system — all chart colours use existing `--ps-*` variables
- `DashboardOverviewSection.vue` is absorbed into `BriefingCard.vue`
  and can be removed after migration

FilesExpand file tree

UXS-006-dashboard.md

Latest commit

History

UXS-006-dashboard.md

File metadata and controls

UXS-006: Dashboard

Summary

Layout

Corpus artifacts (status bar)

E2E contract

Revision history

Dashboard implementation specification

1. Overview

2.3 Direct end-label plugin

3.3 Last run section

3.4 Corpus health section

3.5 Action items section

4. Coverage Tab

4.1 GI coverage by month

4.2 Feed breakdown

4.3 Artifact activity

4.4 Index status

6.4 Episodes per run

6.5 Feed processing history

7. Tab Order and Defaults

8. New API Endpoints Required

8.1 GET /api/corpus/coverage

8.2 GET /api/corpus/persons/top

Modified components

New API modules

New server routes

New server tests

UXS amendments (after implementation)

E2E surface map

8.1 `GET /api/corpus/coverage`

8.2 `GET /api/corpus/persons/top`