Public doctrine, vocabulary, governance signals, and contact surface. Operational methods remain private and are discussed only under engagement.
Interpretive risk

Source hierarchy: the minimal condition for an AI answer to be defensible

An AI answer does not become defensible because it sounds right. It becomes defensible, at minimum, when it is publicly intelligible which sources prevail, which remain secondary, and when the system must suspend arbitration. Source hierarchy is not a refinement; it is the minimum threshold of defence.

Reading markers — Interpretive risk
  • Publicly establish which sources dominate interpretation.
  • Separate priority, subsidiarity, and inability to arbitrate.
  • See why a good-sounding answer without hierarchy remains fragile.

Sources are not equal

A recent official source, a media paraphrase, a historical trace, product documentation, and a community comment do not share the same status or normative force. Treating them as interchangeable creates false neutrality.

Source hierarchy does not eliminate every contradiction. It prevents a system from turning mere coexistence of traces into automatic truth.

What it means to hierarchy sources publicly

Public hierarchy is not just a list of sources. It is a declaration of which surfaces are canonical, which are complementary, which document caution or conflict, and which must not ground affirmative output.

The hierarchy can remain simple. It must, however, be clear enough that a third party can understand why an answer should rely on one layer rather than another.

When arbitration must stop

A good hierarchy does not only authorise answering. It also authorises suspension. When dominant sources conflict, are stale, or permit only a conditional answer, the system must know how to stop.

Silence is not weakness here. It is the normal effect of a hierarchy explicit enough to refuse false arbitration.

The minimum condition of defence

A defensible answer is not necessarily perfect. It is at least attachable to an intelligible evidence architecture: dominant sources, secondary sources, scope limits, and non-answer cases.

Without that architecture, even a plausible answer remains fragile because no public frame exists to explain why it should have been believed or refused.

Publication boundary

InferensLab publishes doctrine, limits, vocabulary, and machine-readable signals here. Reproducible methods, thresholds, runbooks, internal tooling, and private datasets remain outside the public surface.

Topic compass

Continue from this note

This note belongs to the Interpretive risk hub. Use this topic when the output has consequences: legal exposure, false certainty, silent misclassification, decision risk, and interpretive debt.

Lane: Governance boundaries and decision risk · Position: Doctrinal note · Active corpus: 16 notes

Go next toward

  • AI governance — Policies, boundaries, proof obligations, change control, and machine-first publication.
  • Interpretation phenomena — Recurring phenomena: fusion, smoothing, invisibilization, coherent hallucinations, etc.
  • Agentic era — Agents, delegation, non-answers, safety, and proxy governance.

Source lineage

This note builds on a post published on gautierdorval.com (2026-01-27). This InferensLab edition reframes the material for institutional legibility, public doctrine, and machine-first indexing.

Related machine-first surfaces