Start Here

Truth Machine is a reference architecture for maintaining attributable current truth from incomplete, overlapping, or conflicting evidence.

Read this correctly

A source claim is evidence, not current truth.
Source authority is narrow and declared; importance or confidence does not expand it.
Reconciliation produces an exact proposed delta against an identified current state and coherence boundary.
Review binds the exact proposal. Stale, invalid, or partly applicable work advances nothing.
A successful transition advances current truth and appends its reason together.
Unknown, disagreement, and uncertainty remain explicit.
Publications and outside contributions do not become parallel truth paths.
An AI Peer may perform substantial work but gains no authority from fluency.

Normative core

The normative specification is under /specification/. Requirements use stable identifiers such as TM-EVD-001 and the BCP 14 terms MUST, MUST NOT, SHOULD, and MAY. Conformance applies only to those identified requirements—not to examples, profiles, patterns, or this orientation.

Optional material

Git-native profile — one concrete implementation using commits, text packets, and an offline reader.
AI Peer profile — how to make a truth system safely inhabitable by an AI collaborator.
Facets — disposable, non-authoritative projections at an outside boundary.

None is required for core conformance.

Fast assessment

The architecture is a candidate when all of these are real:

people act against durable shared claims;
sources may be incomplete, conflicting, or differently authoritative;
changing the current account has consequences worth attributing; and
the prior account and reason for change must remain recoverable.

Use an ordinary record, script, or application when those conditions do not hold.

Machine-readable entry points

The published site provides:

/llms.txt — curated reading map;
/llms-full.txt — complete documentation context;
/context/core.md — compact normative context;
/manifest.json — versions, classifications, paths, and content hashes;
page-level Markdown at each HTML route plus index.html.md; and
/schemas/ — reference packet schemas.

All are generated from the same documents as the human site. If two generated forms disagree, the build is invalid; docs/ remains canonical.