Evidrel — relational patterns and research records
Distinct prototypes for pattern extraction, evidence selection and review after source changes.
Concepts in this work
Follow a concept to discover related writing and work.
More concepts (3)
Overview
Keeping knowledge with its grounds
Can a record retain not just what it says, but the relations, sources and validity conditions governing its use? Work under the Evidrel name addresses different parts of this question.
A relational research kernel
The status report dated 14 September describes pattern extraction in defined relational languages, selection of evidence separating competing explanations and controls on use after grounds change. The reported results concern narrow synthetic tasks, not open-ended representation generation or broad cross-domain generalisation.
An important negative lesson is that agreement between explanations, one remaining candidate or reduced uncertainty does not guarantee real-world correctness. Shared wrong assumptions need separate testing.
A research-notebook prototype
A separate local tool stores versioned text sources, supports review beside the source and links corrections to records. Source changes send records from the old version back for review.
Review approval is not scientific validation. Search is text matching in this version; language-independent meaning learning, team permissions and cloud operation are not established capabilities.
The limit of a shared name
The kernel, notebook and executor are not one completed product. Shared design questions with BMK do not transfer evidence between these mechanisms.
Sources and open work
Sources reviewed on 14 October 2026: the Örgü status report (14 September) and notebook README. Tests were not rerun for this entry. A shared contract, external comparisons and a human pilot remain separate evaluation tasks.
Naming note — 18 October 2026: This work, previously called Örgü, continues under the name Evidrel. Historical sources retain their original titles. The name change does not mean that the separate prototypes have been integrated or that a new product version has been released.
Experiments and research records
No separate experiment or research records have been published yet. This does not mean no experiments have been run; the current scope is described in the research text.
Go to the research textWriting from this research
How do we publish an unsuccessful experiment?
From an ADL experiment that missed its target to Evidrel’s decision traces: keeping expectations, observations and later explanations distinct.
The trace of a correction in Evidrel
Source, scope and review decisions: how a correction preserves an old record and changes its subsequent use in Evidrel’s local research notebook.
When should we stop using a result?
Source changes, corrections and conditional inferences in Evidrel: when should we stop relying on a record while keeping it?