Amendment 2026-05-29 (v1.1): This decision pivoted from Metabase Cloud → Hex Apps. The v1.0 body below stands as the original analysis; the v1.1 amendment in the Change log section at the bottom is authoritative for current state. Two production dashboards (Daily BC Dashboard, Strategic Weekly Review) are now live in Hex per the v1.1 decision.

WF-6.9 — Dashboard Tool Decision

Type: Vendor / architecture decision Value Driver: VD6 Financial Operations (data infrastructure) → consumers across VD1–VD5 Version: 1.0 Date: 2026-05-12 Decision owner: Roger Thompson Status: Approved — Roger Thompson, 2026-05-13 Parallel to: WF-6.7 Warehouse Engine Decision (same lens applied to the BI layer)

Summary

Recommendation: Adopt Metabase Cloud (Pro tier) as the primary dashboard/BI tool for Cedarfell. Reject Hex, Sigma, Tableau, Power BI, and custom Streamlit on a mix of cost, operator-profile, and operational-burden grounds. Hold Evidence.dev as a “static report” complement if a specific reporting need emerges (PDF-style monthly packets, embeddable narratives) — not in v1.0 scope.

One-line rationale: Metabase Cloud is right-sized for Cedarfell’s audience (BCs as primary consumers, Roger as power user, Jana as occasional viewer), MotherDuck-compatible out of the box, $1K/year at 1 CSA scaling to ~$6K/year at 10-CSA portfolio (within range of WF-6.7’s right-sized cost framework), low-ops (managed, no self-host burden), replicable per CSA, and lock-in-free because every dashboard is a SQL query that any BI tool can re-implement.

Three-year TCO advantage over Tableau Cloud: approximately $15K–$30K at 1 CSA growing to $60K–$120K at 10-CSA scale.

Background

Trigger. WF-6.4 Phase 3 (Dashboard Layer) was originally scheduled for Q4 2026 but is landing earlier because Phase 2 sprint 1 has progressed faster than planned. As of 2026-05-12, 47 gold/silver views are live in MotherDuck across main and main_main schemas — 15+ views beyond what WF-6.8 v1.0 specified. The analytical layer is well ahead of the consumer layer; this decision picks how those views surface to the team.

Use case. Build the dashboard layer that consumes Cedarfell’s MotherDuck warehouse. Six scorecards designed in /Templates/Scorecard-Templates-v1.0.md are the target consumer surfaces (Daily BC, Daily Owner, Weekly, Monthly, Quarterly, Annual). Daily BC Dashboard is the first build (per Roger 2026-05-12).

Operator profile (unchanged from WF-6.7). Sole owner-operator (Roger) with corporate management background, two BCs running daily ops (Vincent + Brandon), one bookkeeper at Tensor (Jana), one offshore bookkeeper (Sheryl), one HR/Fleet coordinator (Aaliyah migrating to Swan Island). No internal data engineer, BI engineer, or full-stack developer — and the operating model isn’t going to support hiring one. Tool choice must serve consumers who aren’t dashboard builders.

Consumer-vs-builder split. This is the critical distinction the tool must handle:

• Builders (1–2 people): Roger, occasionally Claude in the dev-repo session. Comfortable with SQL, dbt, technical config.
• Daily consumers (3–5 people): BCs (Vincent, Brandon) every morning. Roger every morning. Aaliyah occasionally. Drivers monthly (one-page scorecard print, not a dashboard interface).
• Weekly/Monthly consumers (2–3 people): Jana monthly P&L; Roger + BCs Friday weekly review.

The tool must let one technical builder produce dashboards that non-technical consumers use without training. That’s the defining requirement.

Strategic constraints.

• Replicability — KPI Taxonomy STR.EXIT.04 (Playbook replicability score) and STR.EXIT.06 (Data warehouse maturity) include the BI layer. New CSA owners must clone the dashboard pack without specialized skills.
• MBO 2035 exit — the chosen tool will be in place ~9 years. Switching cost matters less than year-1 fit because the SQL queries port across tools easily.
• Lock-in tolerance — same as WF-6.7. As long as queries live in dbt or SQL files (not vendor-proprietary scripting), migration risk is bounded.

Data scale (carried from WF-6.7).

DuckDB on MotherDuck handles each of these scales with margin. The dashboard tool’s job is presentation, not aggregation — heavy lifting stays in dbt gold views.

Options evaluated

Option A — Metabase Cloud (Pro tier) (recommended)

Architecture: Managed SaaS at metabase.com. Native MotherDuck/DuckDB connector via JDBC. Question + Dashboard primitives; SQL editor for power users; click-to-build for casual users. Email + Slack alerts on metric thresholds. Embedded dashboards possible if needed.

Pricing as of 2026:

• Starter: ~$85/mo annual (5 users) — suitable for 1 CSA
• Pro: ~$500/mo annual (10 users, 50 dashboards) — suitable for 5+ CSAs
• Enterprise: custom

Operator fit: Excellent for non-technical consumers; SQL editor good for Roger and dev-repo workflows.

Option B — Evidence.dev (held as complement, not primary)

Architecture: Open source, SQL + Markdown templating that compiles to a static site. dbt-native. Deploy to Vercel/Netlify for ~$20/mo. Every dashboard is a git-versioned .md file with embedded SQL. Charts via templated components.

Pricing: $0 (self-host) + ~$20/mo hosting.

Operator fit: Excellent for Roger and dev-repo workflows (git-versioned, dbt-native, CI/CD-friendly). Poor for interactive BC consumption — static site, no filters, no drill-down without rebuilds. Best for narrative reports and PDF-style monthly packets.

Option C — Hex

Architecture: Notebook + dashboard hybrid. Strong MotherDuck partnership (Hex was an early integration partner). SQL + Python in cells. Strong collaboration features.

Pricing as of 2026:

• Team: ~$30/user/mo (annual) — 5 users = $1,800/yr
• Professional: ~$50/user/mo — 5 users = $3,000/yr

Operator fit: Better for analyst-led exploratory work than operator-consumed dashboards. Notebook-paradigm doesn’t match a BC’s “show me yesterday’s number” need.

Option D — Looker Studio (Google, formerly Data Studio)

Architecture: Free, Google-hosted. Connects to MotherDuck via community JDBC connector (limited maturity) or via BigQuery pass-through (extra hop, extra cost).

Pricing: $0.

Operator fit: Acceptable for casual consumption. Limited charting palette. MotherDuck connector is community-maintained, not vendor-supported — reliability risk. Google account dependency.

Option E — MotherDuck native (Pulse + Notebook)

Architecture: Native MotherDuck dashboarding. Pulse (released 2025) for monitoring/alerts; notebook-style query result sharing.

Pricing: Bundled with MotherDuck subscription; no incremental cost.

Operator fit: Lightweight; not a full BI suite. Good for ad-hoc result sharing among technical users. Not yet rich enough for a Daily BC Dashboard. Worth re-evaluating at the WF-6.9 v2.0 review (annual) — MotherDuck may close the gap.

Option F — Power BI Pro

Architecture: Microsoft cloud, integrated with M365. DuckDB connector via ODBC.

Pricing: $10/user/mo. 5 users = $600/yr. Requires Microsoft 365 Business Standard ($12.50/user/mo) for some features.

Operator fit: Strong if Cedarfell already runs M365 (unknown — needs Roger confirmation). Otherwise the M365 dependency adds friction. Heavier learning curve than Metabase for casual consumers.

Option G — Tableau Cloud

Architecture: Salesforce-owned managed BI. Premium charting. DuckDB via JDBC.

Pricing:

• Creator: $75/user/mo
• Explorer: $42/user/mo
• Viewer: $15/user/mo
• 1 Creator + 2 Explorer + 5 Viewer = ~$3,000/yr at minimum

Operator fit: Over-built for Cedarfell. Best for organizations with dedicated BI teams.

Option H — Sigma Computing

Architecture: Spreadsheet-paradigm BI; cloud-hosted; strong for Excel-trained operators. DuckDB via JDBC.

Pricing: Custom; typical entry $3–5K/year for small teams.

Operator fit: Good for Jana’s Excel-trained mindset; over-built for BC daily consumption. Cost not justified at 1 CSA.

Option I — Apache Superset (self-hosted)

Architecture: Open source BI. Self-host on Docker/VM. Free.

Pricing: $0 + ~$30/mo VM hosting + Roger’s time.

Operator fit: Reasonable feature set but adds DevOps burden Cedarfell doesn’t have. Same operator-burden penalty as Redshift in WF-6.7.

Option J — Custom Python (Streamlit / Plotly Dash)

Architecture: Code-driven dashboards in Python. Deploy to Streamlit Cloud, Hugging Face Spaces, or Railway. ~$20/mo hosting.

Pricing: $0–$20/mo hosting + Roger/Claude build time.

Operator fit: Maximum control. Maintenance burden Cedarfell can’t carry alone — every dashboard change requires code changes. Rejected on same operator-profile grounds as custom warehouse engines.

Cost analysis (Total Cost of Ownership)

All figures USD/year unless noted. Mid-range vendor pricing as of 2026-05.

TCO at 1 CSA (today)

TCO at 5 CSAs (portfolio, ~2030)

TCO at 10 CSAs (~2031)

TCO at 25 CSAs (MBO 2035 exit)

Cumulative 2026–2035 (delta retained as margin vs. Tableau)

Mid-case projection weighted by likely CSA-count trajectory:

• Option A (Metabase Cloud) cumulative: ~$80K – $150K
• Option B (Evidence.dev) cumulative: ~$40K – $60K
• Option G (Tableau Cloud) cumulative: ~$200K – $400K
• Margin retained by going Option A vs Tableau: ~$100K – $250K over ten years
• Margin retained by going Option B vs Tableau: ~$140K – $340K over ten years

Evidence.dev is the lowest-cost option but pays for it in consumer fit. Metabase Cloud is the cost-vs-consumer-fit sweet spot.

Risk assessment

Aggregate risk: Option A presents lower aggregate risk than Hex, Tableau, Sigma, or custom Python; comparable risk to Evidence.dev (different risk shape — adoption risk lower with Metabase, build-effort risk lower with Evidence).

Performance considerations

Query latency. At Cedarfell’s scale, Metabase Cloud over MotherDuck returns dashboard pages in 1–3 seconds for typical Daily BC views. Heavy monthly P&L roll-ups may want materialization to keep page loads under 5s — that’s a dbt config change, not a tool change.

Concurrent users. Metabase Cloud Pro tier supports 10 concurrent users; Cedarfell + 10-CSA portfolio stays well within design point.

Mobile. Metabase has responsive dashboards out of the box. BCs checking dashboards from phones at the belt (likely scenario) — supported.

Embedding. Metabase supports interactive embedding (Pro tier) and static embedding (Starter tier). If we want dashboards embedded in Slack or in a custom portal, the path is clear.

Alerts. Metabase supports email + Slack alerts on threshold breaches. Useful for: contract-standing composite turning YELLOW, OT exceeding budget, attendance call-out pattern hitting threshold.

Strengths and concerns per option

Option A — Metabase Cloud (Pro)

Strengths

• Right-sized cost at every projected scale ($1K → $50K range over 10 years)
• Best-in-class consumer experience for non-technical users — click-to-build, intuitive filters
• Managed SaaS — zero ops burden
• Native MotherDuck integration — Metabase ↔ MotherDuck partnership documented
• Open-source escape hatch — if Cloud pricing changes, self-host the OSS version
• dbt-aware — can read dbt model docs, surface column descriptions in dashboard authoring
• Alerts + scheduled email digests built in — supports the Daily Owner morning ritual
• Mobile-responsive — BCs can check from phones

Concerns

• Per-user pricing past Pro tier could compound at portfolio scale (manageable; Pro covers 10 users at flat $500/mo)
• Not as visually polished as Tableau or Sigma — acceptable trade
• Some chart types limited (no Sankey, limited custom viz) — not needed for operational scorecards

Option B — Evidence.dev

Strengths

• Lowest cost — $20/mo hosting, no per-user fees
• dbt-native, git-versioned dashboards — perfect for dev-repo workflow
• Replicable per CSA in literally seconds (git clone + config repoint)
• Static site = bulletproof reliability + sub-second page loads
• No vendor lock-in — your dashboards are just markdown + SQL

Concerns

• Static site = no interactive filtering or drill-down without rebuilding
• BCs can’t change date ranges or filter to one driver without a code change
• No alerts (would need a separate alerting layer)
• Limited charting palette compared to Metabase
• Younger ecosystem (founded 2022) — smaller talent pool

Verdict on B: Hold as a complement for static report deliverables (e.g., monthly PDF packet). Not v1.0 primary.

Option C — Hex

Strengths

• Premier MotherDuck partnership
• Excellent analyst-led workflow (SQL + Python notebooks + dashboards in one place)
• Strong collaboration features (comments, sharing)

Concerns

• Notebook paradigm doesn’t fit operator daily consumption — BCs don’t want a notebook
• Per-user pricing scales poorly at portfolio
• Over-built for Cedarfell’s primary use case (operational dashboards)

Option D — Looker Studio

Strengths

• Free
• Familiar to anyone who’s used Google Workspace

Concerns

• MotherDuck connector is community-maintained — reliability risk
• Limited charting + interaction relative to Metabase
• Google account dependency adds friction
• No alerts in the free tier

Option E — MotherDuck native (Pulse + Notebook)

Strengths

• Bundled with existing MotherDuck subscription
• Sub-second performance (no network hop)

Concerns

• Not yet a full BI suite
• Pulse is monitoring-focused, not dashboard-focused
• Worth revisiting in 12 months; not v1.0 ready

Option F — Power BI Pro

Strengths

• Low per-user cost if M365 is already paid
• Mature, enterprise-grade

Concerns

• M365 dependency unknown for Cedarfell
• Heavier learning curve than Metabase
• Microsoft ecosystem lock-in
• DuckDB connector via ODBC (not first-class)

Option G — Tableau Cloud

Strengths

• Premium charting and visualization
• Strong enterprise feature set

Concerns

• 5–10× cost of Metabase at every projected scale
• Designed for organizations with dedicated BI teams
• Over-built for Cedarfell’s operational use case

Option H — Sigma

Strengths

• Excel-paradigm fits Jana’s mental model
• Strong row-level security and governance

Concerns

• Cost not justified at 1 CSA
• Cloud-only, not as flexible for embedded scenarios
• Over-built for daily BC use case

Option I — Apache Superset

Strengths

• Free, open source
• Comparable feature set to Metabase

Concerns

• Self-hosting required — same DevOps burden Cedarfell can’t carry
• Smaller community than Metabase
• Steeper learning curve than Metabase

Option J — Custom Streamlit / Plotly Dash

Strengths

• Maximum control
• Strong Python/SQL integration

Concerns

• Every dashboard change requires code change
• No non-technical authoring path
• Rejected on operator-burden grounds (same logic as custom warehouse engine in WF-6.7)

Recommendation

Proceed with Option A — Metabase Cloud (Starter tier at 1 CSA, upgrade to Pro at 5+ CSAs).

This decision optimizes for what matters at Cedarfell’s scale and trajectory: consumer experience for non-technical users (BCs, Jana), managed operations (no DevOps burden), MotherDuck-native integration, replicable per CSA without specialized skills, and cost discipline that compounds in our favor as the portfolio grows. Open-source escape hatch limits lock-in risk to near-zero.

Hold Option B (Evidence.dev) as a complement for any future static-report deliverable that needs to be a PDF, embedded in a portal, or shared as a versioned artifact. Not v1.0 scope.

Reject Options C, F, G, H, I, J on a mix of cost, operator-profile, and operational-burden grounds.

Reject Options D, E as too immature for v1.0 — revisit Option E (MotherDuck native) at annual review.

Negotiation points (Metabase specifically)

When subscribing to Metabase Cloud:

1. Annual prepay discount — Metabase Cloud annual prepay typically saves 15–20% vs monthly. Don’t commit annually until 60–90 days of validation on Starter tier.
2. Free 14-day trial — use it. Build the Daily BC Dashboard during trial; only convert to paid after a real morning cycle proves the value.
3. Starter → Pro transition — confirm pricing-by-user-count vs flat-tier. At 5 CSAs we want a predictable cliff edge, not unbounded scaling.
4. SOC 2 attestation report — request before any future buyer or contract review (MBO 2035 narrative).
5. Data export guarantees — confirm in contract: dashboards exportable as JSON, queries exportable as SQL, at no charge.
6. Acquisition / shutdown clause — include 12-month data-export guarantee post-event.
7. Open-source compatibility — confirm that JSON dashboard exports from Cloud are importable into self-hosted OSS Metabase. (This is the lock-in escape hatch.)

Implementation impact

The following artifacts will be created or updated based on this decision:

The dev repo (cedarfell-warehouse) doesn’t need code changes from this decision — Metabase Cloud connects directly to MotherDuck via JDBC; gold views serve as the data layer unchanged.

Decision and sign-off

Cross-references

• WF-6.4 Scorecard Data Acquisition Plan — /VD6 - Financial Operations/Workflows/
• WF-6.7 Warehouse Engine Decision — /VD6 - Financial Operations/Workflows/ (sibling decision; same evaluation lens)
• WF-6.8 Gold View Map — /VD6 - Financial Operations/Workflows/ (data sources for every dashboard)
• KPI Taxonomy v1.1 — /Strategic/ (metric definitions surfaced in dashboards)
• Scorecard Templates v1.0 — /Templates/ (target layouts for dashboards)
• Decision Catalogue — /VD6 - Financial Operations/Data/ (decisions every dashboard supports)
• Decisions-Log.md — entry dated 2026-05-12 for this decision

Change log

v1.1 amendment — Hex Apps replaces Metabase Cloud (2026-05-29)

Decision. Cedarfell adopted Hex Apps (Team plan) as the production BI tool instead of Metabase Cloud.

Trigger. Discovered 2026-05-13 while setting up the Daily BC Dashboard: Metabase Cloud does not include the MotherDuck/DuckDB connector. The MotherDuck-compatible Metabase driver (motherduckdb/metabase_duckdb_driver) is a community-maintained plugin and requires a self-hosted Metabase instance to load. This broke v1.0’s “Metabase Cloud is MotherDuck-compatible out of the box” premise.

Options considered.

1. Self-host Metabase — preserves v1.0 tool choice but absorbs the operational burden v1.0 explicitly chose to avoid (Docker container maintenance, Metabase upgrades, app-DB backups, ~$20/mo hosting). Rejected.
2. Hex Apps (chosen) — Hex’s premier MotherDuck partnership provides native connectivity out of the box. v1.0 had rejected Hex as “notebook paradigm doesn’t fit BC consumption,” but the Hex Apps mode (notebook is authoring; consumers see polished dashboards) addresses that concern. v1.0’s analysis was based on a narrower view of Hex than warranted.
3. Evidence.dev / Looker Studio / others — all carry their own MotherDuck integration trade-offs. None as clean as Hex Apps.

Cost impact. Hex Team plan (~$30/user/mo, currently 1 active user = $360/yr) replaces Metabase Cloud Starter ($85/mo = ~$1K/yr). Lower cost at 1-CSA scale; scales similarly within WF-6.9 v1.0’s $6K-at-10-CSA ceiling. Three-year TCO advantage over Tableau still holds.

Operating impact.

• No ops burden change — both managed SaaS, no hosting to maintain.
• Hex’s “Apps” mode renders the notebook as a polished dashboard. BCs never see notebook cells.
• Scheduled refresh on Hex matches v1.0’s expectations for daily/weekly dashboard cadence.
• Per-CSA replication: Hex collections clone similarly to Metabase collections.

Production state as of 2026-05-29.

• Daily BC Dashboard — live at portal.cedarfell-logistics.com/dashboards/daily-bc (shipped 2026-05-20)
• Strategic Weekly Review — live at portal.cedarfell-logistics.com/dashboards/strategic-weekly (shipped 2026-05-29)
• Both behind Cloudflare Access; iframe-based embed; URLs stored in Cloudflare Pages env vars.

Downstream document impact.

• WF-6.10 Dashboard Layer Plan — all “Metabase” references should be read as “Hex”. Build-effort estimates (originally in Metabase days) are unchanged — Hex builds at similar speed for the same scope.
• WF-6.12 Internal Portal Build Plan — Phase C “Metabase Cloud trial” step replaced with “Hex Team trial”. Embedding mechanic (iframe with env-var URL) is unchanged.
• Substantive rewrites of WF-6.10 and WF-6.12 deferred; this amendment is the authoritative pointer for now.

Re-evaluate at. WF-6.9 annual review (per the original document cadence). If Metabase Cloud adds native MotherDuck support, reconsider — but switching cost from Hex is low because every Hex query is portable SQL.

Appendix A — Reference architecture (target state)

            ┌───────────────────────────────────────────────────────┐
            │   MotherDuck cloud database `cedarfell`               │
            │   ──────────────────────────────────                  │
            │   • 30 bronze tables (~16K rows)                      │
            │   • 47 silver + gold views                            │
            │   • 1 materialized gold table (monthly_driver_scorecard)│
            └────────────────────────┬──────────────────────────────┘
                                     │ JDBC connector (Metabase ↔ MotherDuck)
                                     ▼
            ┌───────────────────────────────────────────────────────┐
            │   METABASE CLOUD (Starter → Pro)                      │
            │   ──────────────────────────                          │
            │   ┌──────────────────────────────────────────────┐    │
            │   │  Dashboards (one per scorecard cadence)       │    │
            │   │  • Daily BC Dashboard                         │    │
            │   │  • Daily Owner Dashboard                      │    │
            │   │  • Weekly Friday Review                       │    │
            │   │  • Monthly Driver Scorecard                   │    │
            │   │  • Monthly P&L Review                         │    │
            │   │  • Quarterly QBR                              │    │
            │   │  • Annual Review                              │    │
            │   └──────────────────────────────────────────────┘    │
            │   ┌──────────────────────────────────────────────┐    │
            │   │  Alerts (email + Slack)                       │    │
            │   │  • Contract standing → YELLOW                 │    │
            │   │  • OT spend > budget                          │    │
            │   │  • Driver rest_candidate = TRUE               │    │
            │   │  • Swan Island break-even hit                 │    │
            │   └──────────────────────────────────────────────┘    │
            └────────────────────────┬──────────────────────────────┘
                                     │
                                     ▼
            ┌───────────────────────────────────────────────────────┐
            │   CONSUMERS                                           │
            │   ─────────                                           │
            │   Roger        — Daily Owner + Weekly + Monthly + QBR │
            │   Vincent/Brandon — Daily BC + Weekly                 │
            │   Jana         — Monthly P&L + Quarterly close        │
            │   Aaliyah      — Coverage Event Log (when built)      │
            │   Drivers      — Monthly Scorecard (printed via WF-4.4)│
            └───────────────────────────────────────────────────────┘

Per-CSA replication pattern:

Each new CSA gets:

• Its own MotherDuck database (or shared database with csa_id filter — same option from WF-6.7 still open)
• A Metabase Collection (folder) cloned from the Cedarfell template
• Same dashboards, same SQL, repointed at the new database
• Estimated setup time per CSA: ~2 hours

A new CSA owner buying into the playbook signs up for Metabase Cloud Starter, imports the dashboard pack as JSON, connects to their MotherDuck database. Total setup time under half a day.

Appendix B — Vendor financial stability snapshot

Concentration risk note. MotherDuck is the data layer; Metabase is the BI layer. Two vendor dependencies, both with open-source fallback paths. AWS S3 (via MotherDuck-managed storage) is the third dependency. None of these create dangerous concentration.