Dknowledge: memory that can be examined and corrected

Dknowledge connects information with its source, context and use. The public repository currently brings together papers, architecture and decisions. The broader proposal extends this principle across public knowledge, project records and private contexts, with distinct permissions for each.

From a source to a decision

A useful record identifies where a claim came from, when it was observed, which assumptions qualify it and which revision was used. An inference should remain distinguishable from an observation, and an agent’s explanation should be linked to evidence of the action or result it describes.

A correction carried forward

Consider a project whose resource plan relies on a measurement. A later review finds that the instrument was miscalibrated. The corrected record should preserve the original reading and explain the change. The project can then locate the analysis that used it, rerun the affected calculation and record whether the plan needs revision.

This example defines a behaviour to develop: source → use → correction → review of dependent conclusions. A correction should reach the places where it matters, with responsibility for deciding which actions need to change.

A correction adds context; it never erases the trace. The error stays where it was, linked to the evidence that corrected it. What changes is how it is perceived, through the relations added around it. The same rule holds at the personal, local and global scales: in a person’s own Dknowledge, errors and traumas are not removed either; the relations added around them change how they are perceived. The record and its origin stay intact while the relations around it and its contextual reputation are remade with each test in the world: memory stays alive without having to lie about the past.

Whoever decides what is retrieved helps define what seems important now, and whoever decides what is forgotten helps define which links to the past a person can still recognize. Personal Dknowledge makes those decisions understandable and correctable by the person, and keeps each decision together with the conditions in which it was taken, so that a decision suited to conditions that no longer exist does not harden into a permanent identity: an ‘I have always been like this’ can be checked against the record.

Context and permission

Public evidence, project knowledge and a person’s private notes serve different purposes. Reuse requires an appropriate permission boundary and enough provenance to preserve meaning. Two passages between scales are designed: deliberate contribution, under explicit permission and provenance, and anonymous federated learning, in which experience stays where it was lived and only unnamed patterns circulate. Their privacy properties still have to be specified and tested: what each update and its metadata could reveal.

Words change, relations stay

The words that name concepts change over time. What changes is the configuration of the network of relations, not the record of what was said. The genealogy of words is the natural state of the Dknowledge graph: which term was used, when, in which context, and which term came after it. A decision recorded in older words therefore stays legible, because whoever reads it later can follow the relations from the term used then to the meaning it has now.

First worked case

A small demonstrator could connect one source, one inference and one decision, then publish a correction and show which dependent records are flagged for review. That would provide a concrete basis for discussing versioning, responsibility and access controls before extending the design.

See the Dknowledge overview and the Dk architecture.