> For the complete documentation index, see [llms.txt](https://docs.rubyscore.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.rubyscore.io/how-rubyscore-works/ai-scoring.md).

# AI Scoring

AI-driven, anti-Sybil reputation scoring that converts raw on-chain behavior into a **bot-likelihood** signal and **risk tiers** (**Clean / Monitor / High-Risk**). It blends temporal patterns, transaction-graph structure, action diversity, and economic rationality to help gate rewards and access **before** budgets get drained by bots.

***

### How it works (at a glance)

#### 1) Feature engineering from on-chain data

* **Temporal patterns:** activity rhythm, streaks vs. bursts, synchronicity with campaign “waves.”
* **Transaction graph:** connectivity density, cyclic routes, fan-out/fan-in, shared counterparties.
* **Action diversity:** share of unique contracts, depth of sequences (e.g., bridge → deposit → swap → LP → vote).
* **Economic sense:** fee/volume balance, net inflows/outflows, asset retention vs. instant withdrawals.
* **Cross-chain signals:** recurring routes and migrations across networks.

#### 2) Model ensemble

A combination of gradient-boosting and sequence/graph models produces a **Bot Likelihood** score (0–1) plus explanatory flags (e.g., `sync-batch`, `cycle-flow`, `fee-volume-outlier`).

#### 3) Risk tiers & explainability

* **Clean** — low risk
* **Monitor** — medium risk
* **High-Risk** — high bot probability

For integrations, we return a brief explanation of which features pushed the score up/down.

#### 4) Access-rule integration

Use risk tiers **together with** MRS / PoH / Score Level:

* Example rule: **allow access if** `High-Risk = false` **AND** `MRS ≥ 500` **AND** `Level ≥ 6`.
* Tighten thresholds for airdrops; relax them for onboarding flows.

#### 5) Feedback loop

Appeals and manual samples improve quality over time: we tune thresholds per campaign/ecosystem and update features/weights based on outcomes.

***

### Best practices

* **Combine AI risk with reputation metrics** (MRS / PoH / Levels) to reduce false positives/negatives.
* **Set per-campaign thresholds** (stricter for airdrops, softer for onboarding).
* For **Monitor** tier, prefer **soft limits** (reduced caps) over hard bans.

The AI anti-bot layer turns raw transactions into a clear risk signal, enabling informed gating and rewards decisions **before** bots consume your campaign budget.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://docs.rubyscore.io/how-rubyscore-works/ai-scoring.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.