Skip to main content
Concept Engine

Guide

Operational intelligence, defined by evidence.

The term is widely used and rarely defined. This is the definition we work to: what operational intelligence means once every material fact is graded, source-linked, and accountable to its origin.

What it is not

Retrieval and analytics are not the same thing.

Retrieval

Surfaces documents that look relevant. Treats every retrieved passage as equally true. Cannot tell a superseded standard from a current one.

Analytics

Aggregates events after the fact. Answers what happened at scale, not whether a specific claim in a specific document is supported by the underlying system.

Operational intelligence

Treats each material fact as an accountable artefact — graded, source-linked, checked for contradictions, and refused when evidence is missing.

Four properties

What has to be true for the phrase to mean something.

01
GRADED

Every fact carries its evidential strength

MEASURED facts come from a system of record — a table, a log, an instrument reading. INFERRED facts are derived by explicit reasoning from measured inputs. STAKEHOLDER-CLAIMED facts come from an interview or a document with a named author. The grade travels with the fact.

02
SOURCE-LINKED

Every claim points back to its origin

Not a citation of a document, but a link to the exact row, transition, module or paragraph that produced the claim. A reader can verify without asking the analyst.

03
CHECKED

Contradictions surface before decisions

Automated checks compare facts against each other and against the systems they describe. Superseded versions, orphaned facts and disagreements between sources are flagged, not silently averaged away.

04
REFUSED

Silence where the evidence stops

Where the evidence is missing, the answer is a flagged gap or a refusal. A plausible-sounding claim is a failure mode, not a fallback.

What it looks like in practice

The shape of the artefact, from a live engagement.

These are measured figures from a founding engagement — a 25-year-old core system inside a global testing, inspection and certification company. Every row is a fact carrying its evidential strength.

Embedded SQL decoded
~110,000 lines
Workflow engine mapped
24 classes / 184 states / 557 transitions
Production log rows validated against
~24.5 million
Database tables attributed
~466
Certification standards reconstructed
71
Cross-linked documents
300+
Evidence-graded facts
~400
Integrity-check false positives tuned to zero across
230 files

Source: /work — anonymised engagement record.

05 / Questions

Frequently asked, precisely answered.

01

What is operational intelligence?

Operational intelligence treats every material fact as an accountable artefact — graded by evidential strength (measured, inferred, or stakeholder-claimed), linked to the exact source that produced it, checked for contradictions against other facts and the systems they describe, and refused when the evidence is missing. It is not retrieval and it is not analytics.

02

How is operational intelligence different from retrieval-augmented generation (RAG)?

Retrieval surfaces documents that look relevant and treats each retrieved passage as equally true. It cannot tell a superseded standard from a current one, and it will not refuse when the evidence is thin. Operational intelligence grades every fact, links it to a specific row, transition, or paragraph, and returns silence or a flagged gap where evidence stops.

03

What does an evidence grade mean?

Every fact carries one of three grades. MEASURED facts come from a system of record — a database table, a production log, an instrument reading. INFERRED facts are derived by explicit reasoning from measured inputs. STAKEHOLDER-CLAIMED facts come from an interview or a document with a named author. The grade travels with the fact into every downstream use.

04

What happens when the evidence is missing?

The answer is a flagged gap or a refusal. A plausible-sounding claim without evidence is treated as a failure mode, not a fallback. This is why the method is usable in high-consequence environments where an unsupported assertion has a cost.

05

Where has the method been applied?

Two founding engagements — a 25-year-old core system inside a global testing, inspection and certification company, and a top-3 German automotive OEM programme. The certification engagement decoded ~110,000 lines of embedded SQL, mapped a workflow engine of 24 classes / 184 states / 557 transitions against ~24.5 million production log rows, and produced ~400 evidence-graded facts across 300+ cross-linked documents.

The method is what makes the definition operational.

Grade, source-link, check, refuse. Four working rules that keep every fact accountable to its origin. Read the method, or start from the specifics of your own dispersed knowledge.