Evidence is not truth.

A reference architecture for turning incomplete and conflicting claims into an attributable current account through exact, reviewed change.

Start here Read the specification Adopt the architecture

The system in one minute

A truth machine keeps five meanings separate:

Evidence records what a source said or showed.
Reconciliation compares that evidence with current truth under bounded source authority.
A proposed change states the exact delta, rationale, evidence, uncertainty, actor, and expected state.
Review and validation either advance the complete coherent transition or advance nothing.
Current truth and change history move together, preserving why every consequential value stands.

Truth means the current best reviewed account on which the system’s people may responsibly act. It does not mean objective certainty. Unknown and conflict are valid current states.

evidence -> reconciliation -> exact reviewed change -> current truth
       -> publication or encounter -> contribution or observation -> evidence

The architecture is independent of storage engine, programming language, interface, AI model, and domain schema. Its conformance requirements govern meaning and transition boundaries.

What this project provides

a normative semantic specification with stable requirement identifiers;
a self-governing, Git-native reference instance;
JSON schemas, synthetic fixtures, an offline governed reader, and tests;
optional Git-native and AI-Peer implementation profiles;
Facets as an optional boundary-projection pattern;
implementation reports tied to exact source revisions; and
static HTML, raw Markdown, compact and full AI context, and a structured manifest generated from the same canonical documents.

Maturity: specification 0.1.0 is accepted repository truth. It has not yet been published as a public release or assessed by outside implementers. Implementation evidence is real; universality beyond the stated requirements remains unclaimed.