> ## Documentation Index
> Fetch the complete documentation index at: https://docs.alterscope.org/llms.txt
> Use this file to discover all available pages before exploring further.

# Graph Intelligence

> Network analysis over Alterscope's knowledge graph — contagion cascades, concentration, communities, and smart-money flow.

Risk in DeFi is rarely isolated. Protocols depend on each other's tokens and prices; capital concentrates in a handful of wallets; a shock in one place propagates to others. Alterscope models these relationships as a **knowledge graph** and runs network analysis over it to surface risks that a per-asset view misses.

The graph links protocols, pools, assets, and wallets through dependency, holding, and price relationships. The analyses below run on top of it.

## Contagion cascades

The cascade model answers: *if this pool or asset takes a shock, what else is affected, and by how much?* It propagates a shock outward across the graph's dependency edges, dampening the impact at each hop:

```text theme={null}
impact(edge) = edge_weight · (1 − liquidity_buffer) · dampening^hop_distance
```

Effects from multiple paths are combined as a union of independent shocks, and propagation stops once impact falls below a threshold or a maximum hop distance is reached. Alterscope runs standard scenarios (a 10%, 25%, and 50% shock) and returns the affected subgraph, which powers the contagion map in the product.

**Calibration.** The same-chain per-hop dampening is **empirically fitted** against a corpus of historical contagion events (including the UST depeg, the stETH dislocation, the USDC/SVB weekend, and the Euler exploit) via grid search. The **cross-chain dampening is an uncalibrated prior** — the historical corpus is Ethereum-only — and is documented as such so it is read as a conservative assumption, not a fitted result. We publish the propagation formula and the calibration approach; the fitted dampening values are calibrated internally.

## Concentration

Concentration measures how much of a pool sits in how few hands — a thin, concentrated holder base is a liquidity and exit risk. Alterscope computes the **Herfindahl-Hirschman Index (HHI)** over holder shares, alongside top-1% / top-5% / top-10% concentration and the largest holders:

```text theme={null}
HHI = Σ (shareᵢ)²
      i
```

A higher HHI means more concentration. Where wallet-level holder data is unavailable for a pool, the model falls back to concentration over protocol dependency shares so a signal is still produced.

## Communities

Alterscope detects **communities** — clusters of protocols and assets that are densely interconnected — using the **Louvain** community-detection algorithm over the graph. Communities reveal which protocols effectively rise and fall together, and each community surfaces its total-value share and whether it spans multiple chains. Community assignments are precomputed on a schedule (see staleness note below) and each response reports when it was last computed.

## Smart-money flow

The smart-money analysis tracks position changes by **whale and labeled entities** — wallets above a size or pool-share threshold — and classifies each move as entering, exiting, or holding. The intent is to surface where sophisticated capital is rotating before it shows up in aggregate metrics.

<Warning>
  Smart-money and wallet-level concentration depend on **wallet-ingestion coverage**, which varies by environment and is still expanding. Treat these as a *capability* whose signal quality tracks how completely the relevant wallets have been ingested — not as comprehensive coverage of all market participants. See [Coverage & gaps](/trust/data/coverage-and-gaps).
</Warning>

## Supporting network metrics

The graph also precomputes standard network-centrality metrics as node properties, used as inputs to the analyses above and available on graph endpoints:

| Metric                      | What it captures                                                              |
| --------------------------- | ----------------------------------------------------------------------------- |
| PageRank                    | Overall influence of a node in the dependency network.                        |
| Degree (in / out / total)   | How many direct relationships a node has.                                     |
| Weakly-connected components | Which nodes belong to the same connected region.                              |
| Betweenness                 | How often a node sits on paths between others — a bridge / chokepoint signal. |

<Note>
  **Batch staleness.** Graph metrics and community assignments are recomputed on a schedule (faster-moving metrics hourly, heavier ones every few hours), so they can lag live state by up to their refresh interval. Endpoints report the last-computed timestamp so you can see exactly how current a result is.
</Note>

## Where this shows up

Cascade, concentration, community, and smart-money results are returned by the graph endpoints in the [API reference](/api-reference/overview), each carrying [freshness and quality metadata](/trust/data/freshness-and-quality). The contagion / systemic signals also feed the [systemic factor](/trust/methodology/risk-factors) in the composite score. The honest boundaries — coverage dependence and the uncalibrated cross-chain prior — are on the [Limitations](/trust/methodology/limitations) page.