Intelligence Is a Primitive

The Three Questions

In CANONIC, INTEL is one of the three primitives. It’s the knowledge layer every service inherits. Not a database. Not a training set. Not a folder of PDFs. A governed body of knowledge where every claim traces to proof and every insight traces to its source.

INTEL answers three questions:

Where did this come from? Every insight has a source. Every source has a tier. Gold evidence is timestamped and hashed — git commits, transcript archives, cryptographic artifacts. Silver is ledgered — deal records, meeting logs. Bronze is reconstructed — conference notes, post-hoc summaries. The tier tells you how much to trust it.

Is it still true? Intelligence goes stale. What was accurate last quarter might be dangerously wrong today. The governance framework detects drift — when patterns diverge from prior intelligence, the system flags it. Stale intelligence is more dangerous than no intelligence, because stale intelligence is confident.

Can you prove it? When MammoChat recommends additional imaging, the intelligence behind that recommendation traces through clinical guidelines (gold), validated by credentialed clinicians (VITAE), timestamped on the LEDGER. The intelligence doesn’t just exist. It’s auditable. Pull the thread. It holds.

Four Domains

Every governed system generates intelligence across four domains:

Strategy answers “where are we going?” Patterns from execution — not from slide decks. What worked. What didn’t. Where the network has hidden edges.

Sales answers “who needs this?” Buyer segments. Objection patterns. Close rates. The intelligence that turns a cold call into a warm handoff.

Technical answers “does it work?” Validation scores. Compliance architecture. The mathematical proof that the system does what it claims.

Market answers “what’s changing?” Regulatory timelines. Competitive gaps. The EU AI Act deadline that’s no longer theoretical.

Why “Governed” Matters

The word “governed” does heavy lifting. Anyone can build a knowledge base. Drag some PDFs into a vector store. Point a RAG pipeline at it. Ship it.

But when someone asks “why did the AI recommend this?” — the ungoverned system shrugs. The model said so. The embedding was close. The retrieval seemed relevant.

The governed system answers. This recommendation traces to this guideline (committed, hashed), validated by this clinician (VITAE on file), current as of this timestamp (on the LEDGER). The intelligence doesn’t just inform the output. It proves the output.

That’s the difference between data and intelligence. Data fills a binder. Intelligence answers the oncologist’s question.

Context

Type

Data

post

audit-trail

items: Guideline → Validate → Timestamp → Output

Every AI company has data. We have governed intelligence.

A pharma rep walks into a tumor board meeting with a binder full of clinical trial data. Eighty pages. Charts. P-values. Kaplan-Meier curves. Impressive.

The oncologist glances at the binder and asks one question: “Which of these trials enrolled patients who look like mine?”

Silence. The rep has data. He doesn’t have intelligence.

That’s the gap. Data is what you collect. Intelligence is what you know — with evidence, with context, with provenance. And in most AI systems, the gap between them is a canyon.