the engine behind kerte.io
kerte is an assessment engine that answers those three questions with a measure you can audit — not a score you have to trust.
measures before maps.
This is one measure. Not a dashboard — the smallest honest unit the engine produces.
→ supports a shift toward Defined (3.0)
Every number on this card points to something. The band points to a quote. The quote was confirmed by a person. The floor explains what the engine refused to conclude. A month from now, “where did this come from” has a file for an answer.
One engine. Three ways of knowing.
Most assessment tools support exactly one way of reaching a number, and hide which one. kerte makes it a setting:
- Deterministic — where the world has already drawn the rule.
- Stochastic — where the truth only arrives as a story, and has to be read out of evidence.
- Composite — where both are true at once.
The code does not change between them. The setting does. That is what we mean by epistemology is a parameter.
How a measure is made
Evidence, not opinion
In stochastic mode the engine reads behavioral evidence and extracts literal quotes. Each quote nudges a belief — a Dirichlet posterior over the five CMMI bands. Nothing moves without a citation.
A floor under the intelligence
Beneath every reading sits a deterministic floor: the band definitions, the allowed vocabulary, the rules that bound a plausible reading. The model proposes; the floor disposes. The engine can use a language model without inheriting its imagination.
The human is the judge
The reader confirms, flips, or skips every cited marker. Confirmation overrides the engine. The engine calculates; it does not decide.
The log is the product
Which question, what answer, which band proposed, what challenged, how it updated, where it landed — recorded. Transparency here is not a claim. It is a record.
The discipline behind the engine
The engine is proprietary. How it’s built is not a secret — the decisions, the reasoning, and the gates are on the record:
- /governance →Decisions on the record. Every architectural choice is an ADR — argued, dated, and kept.
- /how-it-works →Reasoning on the record. The audit log is not paperwork around the engine; it is the engine’s output.
- /governance →Contracts that block, not suggest. Test, traceability, bug, scope, topology — five contracts the work cannot bypass.
- /governance →A bar it must clear. The engine has a published readiness standard for its hardest claim — and the standard holds even when the answer is “not yet.”