Skip to content

SPEC-EVO-005 — Reasoning Engine

Purpose

The reasoning engine preserves how evidence is transformed into provisional understanding without pretending that computation establishes reality.

Required chain

A publishable meaning SHOULD be traceable through:

Question → Evidence → Observation → Reasoning → Meaning → Decision → Revision

Not every investigation must contain every object type, but omitted stages MUST be visible rather than silently inferred.

Requirements

  1. Every reasoning object MUST identify its premises and conclusion.
  2. Assumptions MUST be declared separately from observed evidence.
  3. A meaning marked verified MUST reference at least one reasoning object and one evidence object, directly or transitively.
  4. Competing hypotheses MUST be allowed to coexist.
  5. Contradictory evidence MUST remain visible until explicitly resolved, superseded, or scoped.
  6. Confidence MUST be separate from truth status.
  7. Automated consistency checks MAY detect invalid structure but MUST NOT declare a claim true.
  8. Revisions MUST preserve the previous state through Git history and explicit supersession metadata where applicable.

Evidence classes

Implementations SHOULD distinguish observed, derived, referenced, and external evidence. These classes describe provenance, not correctness.

Acceptance criteria

  • A verified meaning without traceable support fails validation.
  • Two competing hypotheses can be rendered side by side.
  • A contradiction report identifies the exact evidence and reasoning paths involved.
  • Updating a conclusion does not erase the earlier conclusion’s history.

Non-goals

The reasoning engine is not an autonomous judge, theorem prover for all domains, or mechanism for hiding interpretive choices.