Dashboard Guide

This guide walks you through every tab, metric, and chart on the Cliff Horizon dashboard in plain English. No statistics background needed — just a desire to understand what you are looking at and what to do about it.

What is this dashboard?

The Cliff Horizon dashboard is the window into our weather prediction engine. The engine takes raw weather model data from multiple sources (GFS, ECMWF, FuXi-S2S, satellite imagery) and turns it into calibrated probabilities — statements like "there is a 72% chance the high temperature in New York will exceed 85°F tomorrow."

These probabilities power three products: a forecast validation tool (proving the engine works), a limited performance warranty (guaranteeing accuracy to clients), and parametric weather derivatives (financial products that pay out when weather events occur).

The dashboard has ten tabs. This guide walks through each one.

---

1. Overview

The Overview tab is your executive summary. It answers the question: "How is the engine performing right now?" at a single glance.

Metric cards

Charts and alerts

Signal Edge vs Threshold — the bar chart shows how much our engine disagrees with the market for each contract. Green bars mean we think the market is pricing too low (we would buy). Red bars mean the market is pricing too high (we would sell). Taller bars represent stronger disagreements.

Engine vs Actual vs Market — three lines showing what our engine predicted (cyan), what actually happened (green), and what the market priced (grey dashed). The closer our cyan line is to the green line, the better we are doing.

Confidence Distribution — shows how confident the engine is across all signals. Higher z-scores mean stronger conviction. Z-scores above 1.5 are considered worth trading.

Brier Score by City — compares prediction accuracy across cities. Shorter bars are better. If one city is consistently worse, it may need recalibration.

Top Signal Alerts — the biggest pricing opportunities the engine has found, ranked by severity. High severity means a large mispricing — worth looking at closely.

---

2. Forecast

The Forecast tab has two views, toggled by buttons at the top.

D+1 Contracts view

This view shows tomorrow's weather contracts. It is the engine's core output: for each city and temperature threshold, what does our engine think versus what the market thinks?

Exceedance Probability Curve — the most important chart on this tab. The cyan line is what our engine thinks. The grey dashed line is what the market thinks. Where they diverge, there is an opportunity. The x-axis shows temperature thresholds; the y-axis shows the probability of exceeding each one.

Edge by Threshold — same information as the curve, but shown as a bar chart. Green bars mean we think it is more likely than the market does; red means less likely. The purple line shows the statistical significance (z-score) of each edge.

Signal Detail table — every contract for the selected city, with engine probability, market price, edge, z-score, direction (BUY YES or BUY NO), and status (TRADE or monitor).

Subseasonal view (D+1 to D+42)

This view shows the engine's longer-range outlook, powered by the FuXi-S2S model — an AI weather model that forecasts 42 days ahead using 51 different scenarios (ensemble members).

Fan Chart — the centrepiece. The light band shows the full range of the 51 ensemble members (10th to 90th percentile). The darker inner band is the interquartile range (25th to 75th percentile — the most likely range). The cyan line is the median. As you move further into the future, the bands widen — meaning uncertainty increases. This is expected and healthy.

Exceedance Probability by Lead Day — multiple coloured lines, one per temperature threshold. Watch how the probability of exceeding each threshold changes over the 42-day horizon. Lines converging toward 50% at longer lead times means the model is losing confidence — again, normal.

---

3. Explorer

A guided wizard for assessing weather risk at any location. Walk through four steps: pick a location, select a weather peril (heat, rain, wind, etc.), choose an industry, and see the risk profile.

The output shows four risk tiers — Normal, Elevated, Severe, and Extreme — with the probability, estimated cost impact, and recommended protection for each. The Protection Waterfall chart shows how gross risk exposure is reduced by warranties and derivatives, leaving a manageable net exposure.

Use this tab when onboarding a new client or evaluating a new location. The "Export PDF" button generates a client-ready risk summary.

---

4. Scenario

An interactive "what if" modelling tool. Select a project (e.g. Jakarta Highway, Rajasthan Solar Farm), then use the slider to simulate increasingly severe weather events. The dashboard updates in real time to show the probability of that event, the expected delay or revenue impact, the gross cost, and the net exposure after warranty and derivative protection.

The Sales Messaging panel generates context-specific copy based on the coverage gap — useful for showing clients exactly what they are exposed to.

---

5. Markets

The Markets tab tracks live contract pricing and historical engine performance against the market.

Live Market Scanner — a table of all active contracts showing engine vs market pricing, edge, and signal strength. This is the trading desk view.

Engine vs Market Brier Score — a weekly comparison. When our line (cyan) is below the market's line (grey), we are more accurate. The gap between the lines is our competitive advantage.

---

6. Derivatives

This tab shows parametric weather derivative contracts — financial products that automatically pay out when a weather event occurs (e.g. "if rainfall exceeds 50mm in 24 hours, pay $100,000"). There is no claims process; the payout is triggered by the data.

Select a contract to see its trigger condition, the engine's probability of that trigger firing, the fair premium (what it should cost based on pure probability), and the market premium (what the counterparty is charging). The Pricing Waterfall breaks down how the premium is built: raw probability, plus loadings for basis risk, model uncertainty, and counterparty margin.

The interactive purchase flow (Request Cover, Configure, Review, Submit) lets you structure a derivative cover request.

---

7. Calibration

The proof that the engine works. This is the tab you show to anyone who asks "why should I trust your numbers?"

D+1 calibration (top section)

Reliability Diagram — the most important chart in the entire dashboard. It plots what the engine predicted (x-axis) against what actually happened (y-axis). A perfect engine would produce a straight diagonal line — every point would sit exactly on the dashed line. Our engine's cyan line hugs that diagonal closely. Any deviation means the engine is either overconfident (below the line) or underconfident (above the line).

Calibration Summary — key numbers: Brier Score (lower is better; below 0.10 is excellent), MAE (mean absolute error of the raw forecast), and the number of signals evaluated.

Subseasonal backtest (bottom section)

This section appears once the backtesting scripts have been run. It shows how the engine's accuracy degrades as the forecast horizon extends from 1 day to 42 days.

Skill Decay Curve — the headline chart. The cyan line shows Brier score at each lead day (D+1 through D+42). It starts low (accurate) and rises (less accurate) as the forecast gets further out. The green dashed line at 0.10 is the target; the amber line at 0.15 is the acceptable threshold. The shaded area shows ensemble spread — wider spread means more uncertainty.

Reliability Diagrams (2x2 grid) — four small reliability diagrams, one per time period: D+1-7, D+8-14, D+15-28, D+29-42. Watch how the cyan line drifts away from the diagonal at longer horizons. This is expected — the question is how much.

Per-City Brier Table — shows accuracy broken down by city and time period. Green cells are excellent, cyan is good, amber needs attention, red is poor. If one city is consistently red, it may need additional training data or a different bias correction approach.

---

8. Performance

Historical engine performance over time. While Calibration asks "are the probabilities reliable?", Performance asks "how have we done in practice?"

Accuracy Trend — engine accuracy (cyan) vs market accuracy (grey dashed) over time. The purple shaded area shows the edge — how much better we are than the market. A shrinking edge might mean the market is learning from us or that our models need refreshing.

Edge Attribution — breaks down where our advantage comes from: ensemble bias correction, lead-time calibration, multi-source weighting, satellite data, and behavioural signals. Useful for understanding which components of the engine are pulling their weight.

Accuracy by Lead Time — nested bars comparing engine vs market accuracy at D+1 through D+7. The engine's advantage should be largest at D+1 (where we have the most data) and smallest at D+7 (where uncertainty is highest).

---

9. Warranty

Cliff Horizon offers a Limited Performance Warranty: we guarantee that our forecast accuracy stays within a specified variance threshold (currently ±8%). If we breach that threshold over a quarterly period, the client receives service credits.

Variance Gauge — the green bar shows current variance (e.g. 3.2%) against the 8% threshold. As long as the bar stays in the green zone, the warranty is clear.

Service Credit Formula — SC = (MD - VT) x WA, where MD is measured deviation, VT is the variance threshold, and WA is the warranty amount. This only kicks in if MD exceeds VT.

Warranty History — quarter-by-quarter record of variance and whether the warranty was triggered.

---

10. System

The engine's health monitor. Use this tab to check that everything is running correctly.

Variable Registry — lists every registered weather variable with its statistical distribution, bias correction method, Brier scores (raw and calibrated), and last training date.

Pipeline Steps — shows each step of the daily processing pipeline with its status. Green tick means success, amber triangle means warning, red means failure. Click any step to see details.

Data Flow Diagram — a visual overview: Data Sources, then Processing, then Signals, then Output. Helps you understand the engine's architecture at a glance.

---

Quick Navigation Tips

Keyboard shortcuts — press 1 through 0 on your keyboard to jump directly to any tab (1 = Overview, 2 = Forecast, and so on). Hover over a tab to see its shortcut number.

City selector — most tabs have city pills at the top. Click a city to filter all charts and tables to that city only. Click "All" to see the aggregate view.

Variable toggle — on the Forecast and Overview tabs, you can switch between Daily High and Nighttime Low temperature variables.

Header status — the green pulsing dot in the top-right corner means the engine is running normally. If it turns amber, the pipeline has a warning. The source count next to the clock icon shows how many data feeds are active.

Loading state — when the dashboard first loads, you will see animated placeholder cards. This is normal — it is fetching the latest data. It should resolve in 1-2 seconds.

---

Glossary