Marigold Insights | Citizen Volunteer Network

Prerequisites & True Zero-Cloud Architecture

Unlike traditional monolithic architecture, Marigold Insights strictly enforces a Zero-Cloud, 100% Local Compute model. We do not expose a REST API for data ingestion, because we refuse to accept your sensitive civic data over the wire.

Runtime Requirements: You must have Node.js 18+ or a compatible edge runtime (e.g., Cloudflare Workers, Next.js Edge, Bun).
No Network Egress: Our NPM package executes entirely in your local process using WebAssembly (WASM) and heavily optimized JavaScript. Your data never leaves your infrastructure.
Open Standards: The core algorithms (Fellegi-Sunter, NCOA anomaly heuristics) are fully open-source and auditable.

The Zero-PII Guarantee

Because everything runs locally, you do not need to hash, salt, or encrypt your Personally Identifiable Information (PII) before analysis. You process the plaintext data directly in your own secure RAM, and the library outputs the flags. No rate limits, no network latency, no data breaches.

Step 1: Installing the Core Engine

To get started, install the @marigold/local-core package into your project. We provide TypeScript definitions out of the box for strict type safety.

Terminal

1# Install the Marigold Local Compute Engine
2npm install @marigold/local-core
3
4# If you are using TypeScript, types are included automatically.

Step 2: Running Your First Audit

Because Marigold operates entirely in your local process, there is no need to chunk payloads, implement exponential backoffs, or manage REST API tokens. You simply pass your dataset array into the engine.

audit.ts

1import { MarigoldEngine, AnomalyType } from '@marigold/local-core';
2
3// Initialize the local engine
4const engine = new MarigoldEngine();
5
6// Your local, plaintext dataset (never leaves your RAM)
7const localDataset = [
8  { id: "ROW_001", address: "123 Main St", density_code: "APT" },
9  { id: "ROW_002", address: "123 Main St", density_code: "APT" }
10];
11
12// Execute the audit locally (Synchronous or Async)
13const results = engine.detect(localDataset, {
14  anomalyType: AnomalyType.HIGH_DENSITY
15});
16
17console.log(`Found ${results.anomalies_found} issues.`);
18console.log(results.flags);

Step 3: Scaling Locally

Civic datasets can exceed 10 million rows. While our engine is highly optimized (capable of evaluating ~500k rows/second on a standard M1 Macbook), analyzing massive datasets on a single thread will block the Node.js event loop or the browser UI.

We strongly recommend the Web Worker Strategy for browser implementations, or Worker Threads in Node.js.

Multithreading Support