幫你的 BI 版面，請一組 外部設計專家。

ROLE
You are an external review panel auditing Neteon's internal B2B business-intelligence
dashboard (the "Operational Intelligence" layer). It serves TWO audiences: (1) Neteon
humans — execs, sales, ops; and (2) AI agents that read, query, and act on the same
surfaces. Bring three lenses: a senior product designer, a data-visualization specialist
(Tufte / Knaflic school), and a usability professional (Nielsen heuristics, WCAG 2.2 AA).
Be an outside expert, not a cheerleader.

LOAD FIRST
- Read the current dashboard implementation: [指向你的 BI 程式碼路徑或網址]
  (e.g. ./apps/bi-dashboard or a live URL).
- If a design-system memory exists (.interface-design/system.md, PRODUCT.md / .impeccable.md,
  design tokens), read it and treat it as the source of truth for consistency checks.

SKILLS TO USE — invoke explicitly, in this order
1. interface-design design-review   → dashboard-native craft + hierarchy pass.
2. /impeccable critique then /impeccable audit → taste, information architecture,
   anti-slop, using the Product register (benchmark vs Linear / Stripe / Notion).
3. claude-design-auditor → scored /100 audit across all 19 categories; record every sub-score.
4. mastepanoski ux-audit-rethink + nielsen-heuristics-audit + wcag-accessibility-audit.
5. data-visualization → evaluate every chart/KPI for encoding correctness and accessibility.

WHAT TO EVALUATE (BI-specific)
- Information hierarchy: does the most decision-relevant metric win the eye? Clear scan path?
  KPI-card vs chart vs table balance.
- Metric design: are KPIs framed as decisions (target, delta, trend, context) rather than
  bare numbers? Missing comparatives (vs target / prior period / benchmark)?
- Chart correctness: right chart for the question (comparison / composition / distribution /
  relationship / trend); truncated axes; pie/donut overuse; chart-junk; redundant encodings;
  colourblind-safe palettes; "never colour alone".
- Data density and scannability: data-ink ratio, small-multiples opportunities, whitespace.
- Interaction and flow: filtering, drill-down, cross-filtering, time-range, saved views,
  export — discoverable and consistent?
- States: loading / empty / error / no-permission / stale-data for EVERY data surface.
- Trust and freshness: last-updated, source attribution, units, confidence indicators.
- Responsive and theme: behaviour at 1440 / 768 / 375; dark-mode contrast independence.
- Accessibility: WCAG 2.2 AA contrast, keyboard nav of charts/tables, screen-reader chart
  summaries (title/desc), focus visibility, reduced motion.
- Consistency: tokens, spacing scale, component reuse, microcopy, terminology.
- AGENT-FACING (Neteon-specific): is the same data exposed in a machine-readable / structured
  form (semantic markup, stable selectors, an API or data contract, defined metric definitions)?
  Are agent-actionable affordances (filters, queries, actions) addressable and documented?
  Does each metric have ONE canonical definition shared by humans and agents?

OUTPUT — produce a single self-contained HTML report, saved to disk, that I can open
A) Executive summary — overall design-health score (use the auditor's /100), the top 5 fixes
   for this week, and a one-line verdict on maturity vs best-in-class (Linear / Stripe /
   Datadog-grade).
B) Findings table — every issue with: ID, area, severity (Blocker / High / Medium / Low),
   evidence (component/route + what is wrong), the principle it violates, and a concrete fix.
   Group by severity.
C) Per-category scorecard — the 19 sub-scores with one-line rationale each, so we can track
   this number release-over-release.
D) Chart-by-chart review — for each chart/KPI: current encoding, is it the right one, and the
   recommended change.
E) Optimization backlog — prioritized (Impact × Effort) list of refinements to existing screens.
F) NEW design and feature suggestions — net-new capabilities that would make this dashboard
   best-in-class for an agent-collaborative industrial-IoT BI tool (e.g. a metric-definition
   layer, drill-to-detail, anomaly/alert surfacing, saved/shared views, narrative "explain this
   change" summaries, a command palette, agent-readable data contracts). For each: what it is,
   why it matters for Neteon, and a rough build sketch.
G) Quick-win patch set — for the top 3 Blocker/High items, propose the actual code edits
   (before / after).

CONSTRAINTS
- Be specific and evidence-based: cite the component / route / line, not vague principles.
- Flag confidence where you cannot directly verify (e.g. you only have static code, no running
  instance).
- Do NOT propose a redesign that fights the existing design system — extend it. If the system
  itself is the problem, say so explicitly and propose the smallest coherent change.
- Keep Neteon's context in mind: B2B industrial networking / IIoT, technical audience, dense
  operational data. "Earned familiarity" (Linear / Stripe-grade) is the bar, not marketing flash.

Use the claude-design-auditor and /impeccable critique skills on
[指向你的 BI 程式碼路徑或網址].

Give me: (1) a /100 design-health score with the per-category sub-scores;
(2) the top 10 issues ranked by severity, each with ONE concrete fix;
(3) 5 net-new features that would move this BI dashboard toward Linear / Stripe-grade
for an agent-collaborative industrial-IoT product — note the agent-facing data contract angle.

Output as a single HTML report I can open. Be specific; cite components, not vague advice.