> ## Documentation Index
> Fetch the complete documentation index at: https://docs.gdeltcloud.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Worked examples
> Two real events, scored by hand, so you can check our arithmetic against the published definitions.
Each metric's own page under [Event metrics](/reference/metrics) carries worked examples for **that
metric** — several events, one number. This page does the opposite: it takes **two real events** and
scores them on **all four metrics at once**, which is where the interesting disagreements live. A
national blackout is severe, structurally important and highly propagating — and almost worthless to
a market. You only see that by reading the four together.
The variable names below are the sub-factors defined on each metric's own page under
[Event metrics](/reference/metrics). Nothing here redefines them — this is the same rubric,
applied to an instance.
## The two events
Both are real, both are in the corpus, and they were chosen because they disagree in a useful way.
| Event | Domain | Articles |
| ------------------------------------------------- | ---------------- | -------- |
| Nationwide power grid blackout in Havana | `INFRASTRUCTURE` | 18 |
| Anthropic's \$1.5bn copyright settlement approved | `CORPORATE` | 15 |
***
## Magnitude — when nobody dies
> **Havana blackout**
| Sub-factor | Value | The coder's reason |
| ---------- | -------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `path` | `hazard` | No deaths, so the fatality ladder never applies — this scores on the hazard severity tier. |
| `value` | 7 | A nationwide infrastructure failure left approximately 9.6 million people without electricity, halted most public transport, and cancelled tens of thousands of surgeries. |
$$
\operatorname{clamp}(t,\, 0,\, 10) = \operatorname{clamp}(7,\, 0,\, 10) = \mathbf{7.0}
$$
No deaths were reported, so the fatality-based path never applies — and the event is plainly severe.
That is what the other paths exist for. A metric that could only see deaths would score a national
grid collapse at zero.
***
## Systemic importance — a mean of three judgments
> **Anthropic settlement** · node: *Anthropic and the generative-AI copyright/licensing ecosystem*
| Variable | Value | Coder's reason |
| -------------------- | ----- | ----------------------------------------------------------------------------------------- |
| `size` | 0.70 | A major AI developer; reportedly the largest known U.S. copyright settlement of its kind. |
| `interconnectedness` | 0.70 | Claims span authors, publishers, copyrighted books and AI training practice. |
| `irreplaceability` | 0.30 | Important, but there are competing developers and alternative licensing arrangements. |
$$
\frac{s + i + r}{3} = \frac{0.70 + 0.70 + 0.30}{3} = \frac{1.70}{3} = \mathbf{0.57}
$$
***
## Propagation potential — and the one inverted term
> **Havana blackout**
| Variable | Value | Coder's reason |
| ------------- | ----- | ------------------------------------------------------------------------------------------- |
| `channel` | 0.85 | Live operational dependency through electricity-dependent transport and medical services. |
| `coupling` | 0.80 | Total disconnection halted transport and cancelled surgeries immediately — almost no slack. |
| `breach` | 0.55 | A total national grid disconnection crossed a major service-continuity threshold. |
| `primed` | 0.80 | Fuel shortages and a deteriorating grid were already present beforehand. |
| `containment` | 0.35 | Restoration protocols activated — a documented but incomplete damper. |
$$
\frac{c + k + b + p + (1 - m)}{5} = \frac{0.85 + 0.80 + 0.55 + 0.80 + (1 - 0.35)}{5} = \frac{3.65}{5} = \mathbf{0.73}
$$
Note `containment` enters as `(1 − 0.35)`. It is the only sub-factor anywhere in the set that
**suppresses** its metric: high containment means the event is being held, so it should lower the
chance of spread, not raise it.
[The `propagation_potential` page](/reference/metrics/propagation-potential) marks it, and this is
what that marking means in arithmetic.
***
## Market sensitivity — why a huge event scores low
> **Anthropic settlement**
| Variable | Value | Coder's reason |
| ---------------- | ----- | --------------------------------------------------------------------------------------------------- |
| `claim_exposure` | 0.80 | Acts directly on the company's own legal and contractual claim through a court-approved settlement. |
| `economic_bite` | 0.80 | Approval makes a large obligation operative — cash flow and legal standing now, not later. |
$$
\frac{e + b}{2} = \frac{0.80 + 0.80}{2} = \mathbf{0.80}
$$
> **Havana blackout**
| Variable | Value | Coder's reason |
| ---------------- | ----- | ----------------------------------------------------------------------------------------------- |
| `claim_exposure` | 0.10 | No specific traded company or contract identified; exposure is generic sovereign/currency only. |
| `economic_bite` | 0.80 | Already in force nationwide, disrupting core economic capacity. |
$$
\frac{e + b}{2} = \frac{0.10 + 0.80}{2} = \mathbf{0.45}
$$
**This is the metric working, not failing.** The blackout is severe — a magnitude of 7 and a high
systemic importance — but almost nothing traded attaches to it. Market sensitivity measures
*attachment to a traded claim*, not how bad an event is. Reading it as a severity score is the single
most common mistake, which is why
[the `market_sensitivity` page](/reference/metrics/market-sensitivity) lists what it explicitly does
not claim.
***
## Reproducing this yourself
Ask for the sub-factor breakdown and the arithmetic comes back with the event:
```bash theme={null}
curl "https://gdeltcloud.com/api/v2/events?limit=1&sort=significance" \
-H "Authorization: Bearer $GDELT_API_KEY"
```
If a metric comes back `null`, that is the contract working: it means the observable was not found,
never that the value is zero. Each metric's page says exactly when that happens — see
[`magnitude`](/reference/metrics/magnitude), the only one of the four whose scaler returns `NULL`.
What these numbers cannot support, and the measured noise floor.