Micro Brain Exists Because a Registry Cannot Know the Whole World

LightBI's Semantic Registry is very good at one thing: recognizing business concepts that the product already knows how to describe precisely.

The problem is that real businesses do not live inside a finite dictionary. New industries, local terminology, abbreviations and long-tail concepts keep appearing.

A registry should not become an infinite pile of aliases

I wanted a softer understanding layer that could connect terms and relations without turning every new dataset into another hardcoded exception.

That became the Micro Brain.

It is not a small chatbot

Micro Brain is a local deterministic retrieval layer over structured knowledge. It broadens semantic candidate discovery without replacing the canonical registry, resolver, grain gates, readiness, metric preflight or runtime authority.

The Brain may suggest where to look. It does not get to declare truth on behalf of evidence.

Negative knowledge matters as much as similarity

Cards contain positive clues, negative clues, confusion pairs and relations. COD amount is not carrier revenue by default. Stock value is not COGS. Bank inflow is not automatically revenue.

Recall and authority remain separate

Micro Brain → candidate clue
Semantic Registry → canonical identity
Resolver → evidence-bound meaning
Grain/readiness → safety
Metric preflight → execution authority

An upstream hypothesis can be restricted downstream. It cannot grant downstream authority by itself.

Open concepts are allowed to stay open

If the Brain recognizes a useful concept that has no governed canonical bridge yet, LightBI keeps it open instead of forcing it into the nearest supported signal.

The goal is not to make LightBI answer everything. The goal is to understand more while preserving the ability to say “not enough evidence.”