> ## 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.

# Multi-Surface Synthesis

> Combining GDELT Cloud + macro-finance + prediction markets to find the divergence — the analytical insight

## When to use

Whenever you have data from two or more of GDELT Cloud, macro-finance, and prediction markets. This skill turns three side-by-side surfaces into one finding: the **divergence**.

## GDELT Cloud structured data is the spine

Synthesis depends on a quantified anchor. GDELT Cloud's structured Events, Stories, and metrics provide it:

* `significance` on the underlying narrative (so "narrative is high/moderate/low" carries a number)
* `magnitude`, `systemic_importance`, `propagation_potential` (so divergence has structural anchoring)
* `category` / `subcategory` (so the user knows what *kind* of event)
* Linked entities and a date (which feed `SYMBOL_SEARCH` and prediction-market scenario phrasing)

Without that anchor, divergence framing collapses into impressionistic comparison.

## The 5-step flow

```
1. ANCHOR        →  date, actor, location, entity                  (GDELT)
2. NARRATIVE     →  story significance, volume, trajectory          (GDELT)
2b. DIRECTION    →  NEWS_SENTIMENT confirms bullish/bearish/neutral (macro-finance)  [optional]
3. PRICED        →  market reaction in the relevant instrument      (macro-finance)
4. EXPECTED      →  market-implied probability of the scenario      (prediction markets)
5. DIVERGE       →  where the surfaces disagree — the story         (this skill)
```

Steps 1–4 produce data. Step 5 produces analysis.

## The five divergence shapes

| Shape                               | Pattern                                                                 | Lead                                                                                                                             |
| ----------------------------------- | ----------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------- |
| **1. Priced-in**                    | High narrative + flat market + flat probability                         | "Despite N articles, instrument is unchanged and probability is stable. The market has fully discounted this scenario."          |
| **2. Market sees what news misses** | Low narrative + sharp market move + elevated probability                | "Narrative volume is low but instrument has moved X. The market is pricing a development public reporting has not yet captured." |
| **3. News-driven panic**            | High narrative + sharp market move + flat / declining probability       | "Significant move on heavy coverage, but prediction markets show probability unchanged. Price action is sentiment-driven."       |
| **4. Three-way confirmation**       | All rising and aligned                                                  | State the converging signal; focus on second-order effects.                                                                      |
| **5. Surface-redirect**             | All three converge on a topic *adjacent* to the user's literal question | "You asked about A. All three surfaces are saying the answer to a better question — B — is the more important finding."          |

Shape 2 (early signal) and shape 5 (re-pose the question) are usually the highest-value findings.

## Weighting conflicting signals

When surfaces conflict, weight by:

1. **GDELT Cloud structured metrics first.** Significance > 0.4 outranks high `article_count` with low significance.
2. **Confidence on GDELT events.** `confidence > 0.85` beats lower-confidence event coding.
3. **Liquidity on prediction markets.** A probability with $50K open interest beats one with $500.
4. **Specificity of the contract.** Direct contract beats proxy contract.
5. **Recency.** Last 7 days beats 30-day-aggregate signal for fast-moving stories.
6. **Source diversity.** 50 articles from 30 distinct domains beats 50 articles from 3 domains.

When the conflict can't be resolved, name it: *"Signals diverge; the higher-liquidity prediction-market reading suggests X (probability 0.55, \$87K OI), while the GDELT Story significance is 0.18. We weight the market read because the structured GDELT score doesn't support the narrative-volume framing."*

## When NOT to use

* Single-surface tasks — don't manufacture macro-finance and prediction-market hops.
* Lookups — quote, entity profile, known contract terms.
* The data isn't there. If one surface returned empty, say so honestly. Don't invent a divergence.