Most of what you monitor is noise.

Signal Scout pulls from the sources you choose, runs everything through a 12-layer scoring engine, and surfaces only what's worth your attention. Niche-agnostic: point it at enterprise software, at orchids, at anything.

Install View on GitHub
What it does

Ingest, score, decide.

Monitoring tools give you more to read. Signal Scout decides what's worth reading, then helps you say something original about it.

Ingest

Pull from news, communities, research, trends, and company sources. Add your own feeds. The tool doesn't assume your domain.

Score

Every item runs the 12 layers. Emergence, question gaps, source trust, engagement, timing, hype position. Ranked, not just collected.

Decide

Review a ranked queue. Generate structured outputs from what earns it. Sources are cited, never republished.

The engine

Twelve layers, one score.

Five weighted signals form the base score. Three multipliers dial it up or down. Two overrides bypass the formula entirely. Every layer is configurable.

01EmergencePast too-early, before everyone's covering itWeight
02Thought leaderA trusted name posts it, it jumps the queueOverride
03Question gapRepeated questions nobody has answered wellWeight
04Practitioner vs analystWhere the experts and the builders disagreeSignal
05Competitive gapWhat everyone covers, nobody covers, or covers poorlySignal
06TemporalSeasonal timing and pre-event windowsMultiplier
07Source trustEarned over time, decays with low-signal outputWeight
08Engagement velocityComments, shares, upvotes at ingestWeight
09Cross-platform heatSame topic, multiple platforms, 48 hoursMultiplier
10RelevanceDensity of the topics you said you care aboutWeight
11Noise filterClearly off-topic junk drops to zeroOverride
12Hype cycleRewards early and practical, discounts saturatedMultiplier
Architecture

Thick server, thin client.

All logic and every secret stay server-side. The frontend renders and calls the API. Nothing else. No external database: persistence is stdlib SQLite.

PresentationReact + TypeScript, Next.js static export
/api
Logic — FastAPI (Python)One origin: serves the API and the built frontend
Ingestion
Concurrent fetchers
Scoring engine
The 12 layers
Generation
Prompts, LLM client
Storage access
Schema and CRUD
sqlite3
Data — SQLiteSignals, config, outputs, learned state
Install

Running locally.

Placeholder — replace before publishing

These steps are a template. Fill them in from the real repo: actual Python and Node versions, the real install and run commands, required environment variables, and what a first run looks like. Verify each one on a clean machine.

01

Requirements

Python 3.x, Node.js 18+. No external database.

02

Clone and install

git clone https://github.com/halfman-halfmachine/signal-scout.git
cd signal-scout
03

Configure

Copy the example env file and add your key. Generation falls back to templates without one, so the app runs either way.

cp .env.example .env
# ANTHROPIC_API_KEY=...  optional; enables generation
04

Run

The backend serves the API and the built frontend from one origin.

# backend
# frontend build
05

First run

Name your domain, take the suggested sources, and run an ingest. The queue fills with what scored highest.