Skip to main content
b1e55ed is easiest to understand as an event-sourced brain with pluggable producers and an operator-facing workflow.

The data flow

1) Brain (append-only)

The brain is the source of truth. Instead of mutating state directly, b1e55ed records events (signals, updates, confirmations, invalidations) and derives state from them. Why it matters:
  • auditability (what changed, when, and why)
  • reproducibility (rebuild views from events)
  • attribution (who contributed what)

2) Producers

A producer is anything that emits signals:
  • a human operator at the CLI
  • an agent calling the API
  • a connector / scraper / rules engine
Producers should be cheap to add and safe to run. They don’t “control” the brain — they submit events.

3) Synthesis

Synthesis is the layer that turns raw signals into something you can act on:
  • rollups by symbol / timeframe
  • conflict detection
  • aggregation by contributor reputation / karma

4) Conviction

Conviction is the system’s representation of how strongly we believe something right now. It can be:
  • set directly by a producer
  • adjusted by synthesis over time
  • decayed or overridden by contradictory evidence

5) Action

Actions are what you do with the output:
  • notifications / alerts
  • suggested positions
  • execution adapters (manual or automated)
b1e55ed is intentionally modular here: you can keep it “analysis-only” or plug in execution as your risk tolerance allows.

Where to go deeper

  • Internal architecture docs: docs/architecture.md, docs/authority-model.md, docs/learning-loop.md
  • API surface: API Overview
  • Producers: Producer Overview
The repo also includes an older/longer explanation at docs/how-it-works.md.