Where the engines come from.
Caugia’s diagnostic logic is not a black box and is not an LLM guess. It is a corpus of peer-reviewed work, foundational texts, named-practitioner essays and applied frameworks, encoded into deterministic engines and audited on a public rubric.
Three pillars under one constraint engine.
Boards do not buy diagnostic systems whose answers shift between sessions. The corpus, the engines and the audit rubric are designed around that single requirement.
Cited sources, not synthesised opinion.
Every framework slug in the engine maps to a citable source: a peer-reviewed paper with a DOI, a foundational book with an ISBN, or a named-operator essay with a verifiable URL. The provenance is part of the engine, not a marketing layer.
Deterministic, not probabilistic.
The 72 scoring engines apply the corpus to a structured set of business inputs. Given the same inputs, you get the same constraint, the same cost, the same action cascade. LLMs do not decide outcomes. They translate, summarise and present.
Public rubric, weekly re-scored.
Citations are re-audited against a public rubric every week. Hallucinations are tracked. When a source becomes obsolete, the engine is re-pointed and the change-log is published. The audit is not a slide. It is a process.
Fact, inference, hypothesis. Tagged.
Every output Sophie produces is tagged with the type of knowledge underneath it. Investors who confuse projection for evidence lose deals. So we do not let our own system confuse them either.
What the corpus is made of.
We do not publish a count, because counts shift weekly. What matters is composition.
Journal articles with a verifiable DOI.
Marketing science, organisational design, operations research and revenue strategy. We weight peer-reviewed work above commentary, and we cite the article, not a blog summary of it.
Books that shaped the field, with ISBNs.
Goldratt on constraints. Christensen on disruption. Reichheld on loyalty. Mohr-Sengupta on growth. The canon, plus the modern additions that earned their place. Each indexed to a chapter, not a quote.
Named practitioners writing about live problems.
The best growth writing of the last decade lives outside journals: operators publishing what worked, what failed, and why. We include them when the author and the company are verifiable.
Frameworks that earn their slug.
A framework enters the engine only when it has a clear input contract, a deterministic output and a citable source. Frameworks without provenance, or with overlapping definitions, get rejected.
Proprietary signal from the platform itself.
As GRIP OS runs across the customer base, the constraint-distribution and resolution-time distributions become a benchmark layer. This is the only proprietary part of the corpus, and it is anonymised before it influences anyone else’s output.
Governance, not garnish.
The corpus is reviewed on a weekly cadence. Engines are versioned. When a foundational source is superseded, the change is published and existing reports get a citation update note.
We measure citations against a public rubric. The current target: zero hallucinations measured in the rolling sample. When we miss it (rare, but it happens) we publish what was wrong, what changed, and why. Trust is a maintenance job.
Published artefacts.
What is publicly readable today, and what is on the publish-queue.
Methodology paper. Corpus selection and engine determinism.
Publication queued. Email [email protected] to receive the preprint.
The public audit rubric. How we measure Sophie’s citation health.
Annotated rubric and the scoring code will be linked here when the schema stabilises.
Cohort distribution of structural constraints.
Once a privacy-preserving aggregate is stable, the constraint distribution by industry, ARR band and growth archetype will be published quarterly.
The engines are what stand behind the price.
Buy the Intelligence Report to see the methodology applied. Reach out if you want to read the methodology paper before it is public.