Oryx Sentinel — System Documentation

Overview

Oryx Sentinel is an autonomous market-intelligence system. It observes social platforms and Solana market activity, detects emerging narratives, correlates them with newly created on-chain assets, scores the result and monitors qualified opportunities continuously.

Every decision the system takes — including the decision to do nothing — is recorded with its full evidence trail. This interface is a demonstration; data shown may be simulated.

Architecture

The system is organized as a pipeline of independent services connected by an event stream: collectors, a normalization layer, the narrative engine, the correlation engine, the scoring engine, the monitor and the decision engine.

Each stage is stateless with respect to its neighbors and communicates through typed events, so individual stages can be replayed, audited or upgraded independently.

Signal Detection

Collectors emit raw observations: posts, engagement deltas, pair creations, liquidity events. Each observation carries source, timing and reliability metadata.

Detection operates on normalized streams and fires only on sustained deviation from a per-cluster rolling baseline, across consecutive observation windows.

Narrative Engine

The narrative engine clusters semantically related content across platforms into narrative objects. A narrative tracks its volume, velocity, acceleration, source diversity and age.

Clusters merge and split as content evolves; lineage is preserved so a derivative wave can be traced back to its parent narrative.

Correlation Engine

New Solana tokens are evaluated against active narratives within minutes of deployment. Correlation combines metadata similarity, symbol alignment, deploy timing relative to narrative onset, and early holder behavior.

Output is a scored narrative–token pair. Pairs below the correlation floor are archived, not deleted — they remain queryable for research.

Scoring

The scoring engine produces a 0–100 composite from six weighted factors. The current demonstration weights are: narrative strength 25%, social velocity 20%, cross-platform signal 15%, token correlation 15%, market conditions 15%, momentum 10%.

The factor breakdown is stored with every score. A score is never persisted without the evidence that produced it.

Monitoring

Qualified opportunities are re-scored at 30-second resolution. The monitor tracks social decay, liquidity migration, holder concentration and momentum, and emits state-change events consumed by the decision engine.

Decision Engine

The decision engine is deliberately simple: it applies configured conditions to live scores. Entry, hold, exit and rejection are all explicit conditions, and every evaluation produces a reasoning receipt.

Receipts contain the decision, confidence, full evidence set, the reasoning summary and the eventual outcome. Receipts are immutable once written.

Data Sources

Social: X, TikTok, Instagram and Reddit are observed for public content and engagement metadata. On-chain: Solana pair creation, liquidity, swaps and holder data.

All sources are read-only observation. The demonstration interface ships with simulated data in a single mock layer designed to be replaced by live APIs.

API

A read API exposing narratives, correlations, scores and reasoning receipts is planned. Endpoint shapes mirror the internal event schema: /narratives, /correlations, /signals, /receipts.

The API surface documented here is a preview and subject to change before release.

Security

Observation infrastructure is isolated from any execution capability. Decision output is a recorded event, and any downstream execution layer is a separate, independently gated system.

Receipts are append-only. No component can modify a recorded decision.