Global expansion: rotating-globe landing page + per-country macro dashboards

[SecondOrderEdge]

Captured idea — tabled for now, not scheduled. This records the vision and the recommended approach so we don't lose the thinking.

Vision

A slowly rotating 3D globe as the landing page. Click a country → drill into that country's macro dashboard (the current US dashboard, replicated per country over time).

Strongest framing: the globe is the signal

Don't make the globe a decorative menu you click through to reach "the real thing." Make it a live world recession-risk heatmap — each country shaded by its current composite risk / growth signal, slowly rotating, click-to-drill. That gives every country a reason to exist on the map before it has a full dashboard, and turns the wow-factor into the product's hero chart.

The hard part is data + models, not the globe UI

The entire current stack is US-wired:

• Data: FRED series IDs (src/data/series_registry.py) and the probit SERIES_CONFIG (src/models/recession_probit.py) are US-hardcoded.
• Recession target: NBER is US-only. Other countries → OECD-derived recession indicators on FRED (…RECD series) or CEPR (euro area, quarterly) — lower quality, often revised. Harder to defend the "every number validated" credibility.
• Yield curve: the 10y–3m predictive record is strongest for the US; many countries lack a clean/liquid/predictive curve.
• Policy path: the Atlanta Fed Market Probability Tracker is US-only; would need OIS-implied paths per central bank (mostly not free).
• CAPE: Shiller is US-only.
• Probit ensemble: coefficients are calibrated on US history and won't transfer — needs re-fit per country where data allows.
• Data depth: the US is uniquely data-rich (weekly claims, JOLTS, monthly everything); most countries are sparser, laggier, lower-frequency.

Tiered honesty (don't fake rigor)

• Tier 1 — full treatment: US ✅, then candidates Canada, UK, euro area/Germany, Japan, Australia.
• Tier 2 — recession-risk lite: yield curve + a few activity/labor series + OECD recession indicator. No bespoke probit.
• Tier 3 — snapshot: OECD CLI, GDP growth, inflation, unemployment. Clearly labeled a snapshot, not a model.

A faked-rigorous dashboard for 150 countries would undercut exactly what makes this project good.

Recommended sequence (resist building the globe first)

1. Generalize the data layer — a country-parameterized registry mapping each concept (policy rate, 10y, 3m, unemployment, IP, CPI, recession target) to per-country series IDs. Refactor series_registry.py + probit SERIES_CONFIG.
2. Prove with ONE second country end-to-end (Canada or UK = gentlest: good data, Anglo curve, OECD dating). This is where you discover what breaks.
3. Then build the globe as the risk choropleth, once ≥2–3 countries justify it.
4. Expand coverage in tiers.

Building the globe first is the trap — polish on a capability that doesn't exist yet.

Key tradeoff

Breadth vs. the rigor that is the project's differentiator. The tiered approach + globe-as-heatmap is how you keep the rigor: even a shallow country contributes one honest number.

Open questions

• Primary goal: portfolio/showcase piece (globe wow-factor justifies more front-end; maybe a custom three.js / globe.gl component or a non-Streamlit shell) vs. analytical tool (data-layer-first, globe later)?
• Country UI polish, Methodology page, curve drill-downs, rename to Dashboard #2: pick by data availability, not size or interest.
• Scaling: N countries × live FRED-on-load won't scale → likely a precomputed nightly pipeline writing per-country data + lazy per-country loading. Streamlit may be the wrong shell for a rich interactive globe.

Tech notes

• Streamlit globe options: Plotly orthographic scattergeo/choropleth (clickable via selection events) — feasible, but auto-rotation is finicky under Streamlit reruns; or a custom JS component (globe.gl / three.js / deck.gl).

---

Discussion captured from a design conversation; refine in comments.